From ff8855ade75a859c082c11f26a62df67acf1bf5f Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Fri, 17 Jul 2026 15:12:03 +0000 Subject: [PATCH 01/16] docs: restructure AMD auto-applied env vars as tables and add legacy GPU variables - Convert the Windows ROCm auto-applied environment variables list from bare text to formatted markdown tables with notes column - Add table documenting SDP backend override variables automatically set for legacy AMD GPU architectures - Broaden scope description from ComfyUI-specific to all Windows ROCm WebUI package launches - Wrap existing note paragraphs in GitHub-style lockquote note callouts for Linux AMD installs and ComfyUI-Zluda --- docs/advanced/environment-variables.md | 37 +++++++++++++++----------- 1 file changed, 21 insertions(+), 16 deletions(-) diff --git a/docs/advanced/environment-variables.md b/docs/advanced/environment-variables.md index bcd534d02..b26c62ac2 100644 --- a/docs/advanced/environment-variables.md +++ b/docs/advanced/environment-variables.md @@ -143,26 +143,31 @@ Most users should leave these alone unless they are troubleshooting a specific R | `COMFYUI_ENABLE_MIOPEN` | `1` | Tells ComfyUI to keep the MIOpen-backed path enabled on ROCm builds where it may otherwise be disabled by default. Without this enabled, ComfyUI disables the `cudnn` backend path in its backend calls for RDNA3, RDNA3.5, and RDNA4 AMD GPUs, which in turn disables the MIOpen-backed functions that rely on that path. This variable is needed for MIOpen to function properly in those setups. | | `TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL` | `1` | Enables the experimental ROCm AOTriton path in compatible PyTorch builds. In Stability Matrix's Windows ROCm ComfyUI integration, this is used for TheRock technical-preview PyTorch builds to enable AOTriton-provided built-in Flash Attention and PyTorch SDPA memory-efficient attention paths. | -For some Windows ROCm-based ComfyUI launches, Stability Matrix already applies several of these optimizations automatically in package code, including: +For Windows ROCm-based WebUI package launches, Stability Matrix already applies several of these optimizations automatically at package launch as a general all-round baseline. Variables automatically utilized at package launch include: -`MIOPEN_FIND_MODE=2` - -`MIOPEN_SEARCH_CUTOFF=1` - -`MIOPEN_FIND_ENFORCE=1` - -`TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1` (RDNA3 / RDNA3.5 / RDNA4 only, and additionally excluded on the gfx1152/gfx1153 APU architectures where AOTriton isn't yet supported) - -`FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE` - -`COMFYUI_ENABLE_MIOPEN=1` (RDNA3 / RDNA3.5 / RDNA4 only, no gfx1152/gfx1153 exclusion) - -`PYTORCH_ALLOC_CONF=max_split_size_mb:512,garbage_collection_threshold:0.8` +| Variable | Value | Notes | +|---|---|---| +| `MIOPEN_FIND_MODE` | `2` | | +| `MIOPEN_SEARCH_CUTOFF` | `1` | | +| `MIOPEN_FIND_ENFORCE` | `1` | | +| `TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL` | `1` | Enables the experimental ROCm AOTriton path. Applied only on RDNA3 / RDNA3.5 / RDNA4 GPUs, and additionally excluded on gfx1152/gfx1153 APU architectures where AOTriton isn't yet supported. | +| `FLASH_ATTENTION_TRITON_AMD_ENABLE` | `TRUE` | Enables the Triton-based Flash Attention implementation for AMD GPUs, providing an efficient attention path when the native ROCm Flash Attention is unavailable and instead provided by a seperate FA python package. | +| `COMFYUI_ENABLE_MIOPEN` | `1` | Keeps the MIOpen-backed path enabled in ComfyUI on RDNA3 / RDNA3.5 / RDNA4 GPUs. | +| `PYTORCH_ALLOC_CONF` | `max_split_size_mb:512,garbage_collection_threshold:0.8` | | + +Environment Variables automatically enabled for legacy AMD GPU architectures (Vega/GCN5, RDNA1, RDNA2) where certain backend paths are not supported by the hardware or supported well: +| Variable | Value | Notes | +|---|---|---| +| `TORCH_BACKENDS_CUDA_FLASH_SDP_ENABLED` | `0` | Disables the Flash SDP backend. | +| `TORCH_BACKENDS_CUDA_MEM_EFF_SDP_ENABLED` | `0` | Disables the memory-efficient SDP backend. | +| `TORCH_BACKENDS_CUDA_MATH_SDP_ENABLED` | `1` | Enables the math SDP backend as the fallback attention implementation. | -Linux installs do not currently get the same automatic overrides, so they will need to be enabled by the user. +> [!NOTE] +> Linux installs do not currently get the same automatic overrides, so they will need to be enabled by the user. -If you're using the ComfyUI-Zluda package specifically, it also sets its own environment variables at launch on top of the above: `FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE`, `MIOPEN_FIND_MODE=2`, `MIOPEN_LOG_LEVEL=3`, and `ZLUDA_COMGR_LOG_LEVEL=1`. If you're wondering why those already appear to be set for a ZLUDA install, this is why. +> [!NOTE] +> ComfyUI-Zluda is not managed by Stability Matrix's internal ROCm handling. If you're using the ComfyUI-Zluda package specifically, it also sets its own environment variables at launch similar to some of the above: `FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE`, `MIOPEN_FIND_MODE=2`, `MIOPEN_LOG_LEVEL=3`, and `ZLUDA_COMGR_LOG_LEVEL=1`. If you're wondering why those already appear to be set for a ZLUDA install, this is why. Whatever you set in Stability Matrix's own environment-variable editor is applied last, so it always overrides these auto-applied defaults if the same variable name is used. From fe26f85fc51a22fab70bcb6f6ebfc441a4852869 Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Fri, 17 Jul 2026 15:55:27 +0000 Subject: [PATCH 02/16] docs: remove TheRock preview branding from Windows ROCm section, expand package support and caveats --- docs/advanced/hardware-support.md | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-) diff --git a/docs/advanced/hardware-support.md b/docs/advanced/hardware-support.md index de17a74ce..6012b2242 100644 --- a/docs/advanced/hardware-support.md +++ b/docs/advanced/hardware-support.md @@ -40,11 +40,11 @@ The lists below describe what the code checks for. Because hardware detection wo ## AMD on Windows -AMD support on Windows is the most involved case, because there are three different paths depending on your GPU and package: native ROCm (technical preview), ZLUDA, and DirectML. +AMD support on Windows is the most involved case, because there are three different paths depending on your GPU and package: native ROCm, ZLUDA, and DirectML. -### Native ROCm (TheRock technical preview) +### Native ROCm -Stability Matrix can install AMD's native ROCm PyTorch on Windows using AMD's TheRock multi-architecture wheels. This path is gated to a specific set of GPU architectures. The code recognizes the following `gfx` architectures as supported on Windows: +Stability Matrix can install AMD's native ROCm PyTorch on Windows using AMD's official multi-architecture wheels. This path is gated to a specific set of GPU architectures. The code recognizes the following `gfx` architectures as supported on Windows: - **RDNA4** — `gfx120x` (e.g. RX 9070, RX 9060 families). - **RDNA3 / RDNA3.5** — `gfx110x` (RDNA3 desktop and mobile) and `gfx115x` (RDNA3.5 APUs such as the 890M / 8060S / Z2 Extreme families). @@ -54,15 +54,17 @@ Architectures in the `gfx110x`, `gfx115x`, and `gfx120x` ranges are treated as " **What Stability Matrix does automatically:** -- When a supported AMD GPU is present on Windows, ROCm becomes the recommended backend for ROCm-capable packages (currently ComfyUI and Comfy-based flows via SwarmUI). -- Torch is installed from AMD's ROCm multi-arch index (`repo.amd.com/rocm/whl-multi-arch/`) as device-specific wheels (`torch[device-gfxNNNN]`). Vega parts (`gfx900` / `gfx906`) pull from the nightly multi-arch feed instead, since TheRock currently only publishes their device packages there. +- When a supported AMD GPU is present on Windows, ROCm becomes the recommended backend for ROCm-capable packages. +- Torch is installed from AMD's ROCm multi-arch index (`repo.amd.com/rocm/whl-multi-arch/`) as device-specific wheels (`torch[device-gfxNNNN]`). Vega parts (`gfx900` / `gfx906`) pull from the 'TheRock' nightly multi-arch feed instead, since these architecture builds currently are only avaliable there instead of the stable production distribution stream. - On modern architectures it applies a set of ROCm performance and attention environment variables at launch (MIOpen find-mode tuning, AOTriton experimental flash attention, `COMFYUI_ENABLE_MIOPEN`, and an allocator tuning string). AOTriton is excluded on the `gfx1152` / `gfx1153` APU architectures, which it does not yet support. Legacy architectures instead force a math SDP fallback. The full variable list and exactly which ones are auto-applied are documented in [Environment Variables](environment-variables.md#amd-and-rocm-variables). -- ComfyUI offers optional extra commands for supported AMD GPUs, including **Install Triton and SageAttention (ROCm)**, **Install Flash Attention (ROCm)** (legacy architectures), an **Install ROCm Development SDK** step, and an **Install bitsandbytes (ROCm)** step for Python 3.12 environments. +- ComfyUI offers optional extra commands for supported AMD GPUs, including **Install Triton and SageAttention (ROCm)** (Sage Attention 1.x), **Install Flash Attention (ROCm)** (legacy architectures), an **Install ROCm Development SDK** step, and an **Install bitsandbytes (ROCm)** step for Python 3.12 environments. +- **Packages:** ComfyUI, Stable Diffusion WebUI Reforge, InvokeAI, SwarmUI, and Wan2GP are supported by this install path. + > [!NOTE] While not managed by Stability Matrix for ROCm installs, SD.Next has Windows-native ROCm install when the "Rocm" Pytorch index is selected in Advanced Installation Options during initial package install and `--use-rocm` is set in launch options. This install path is internally handled by SD.Next itself and currently only supports RDNA2 dedicated GPUs, RDNA3, RDNA3.5, and RDNA4 GPUs. **Caveats:** -- Windows AMD ROCm is explicitly experimental. Stability Matrix prints a notice asking you to report issues to Stability Matrix first, since the setup may not be officially supported by the upstream package developers. -- Only the architectures listed above are eligible. If your AMD GPU is not on the list, the recommended default becomes ZLUDA or DirectML instead. +- Windows AMD ROCm implementation for Stability Matrix is explicitly in an experimental state. While these package installations are generally in a working state, untested or edge-case issues and incompatibilities may appear. Stability Matrix prints a notice asking you to report issues to Stability Matrix first, since the setup may not be officially supported by the upstream package developers. +- Only the architectures listed above are eligible. If your AMD GPU is not on the list, the recommended default becomes ZLUDA or DirectML instead. For a more detailed list of compatible AMD GPU architectures, please refer to the ROCm/TheRock [GPU Support](`https://github.com/ROCm/TheRock/blob/main/SUPPORTED_GPUS.md#rocm-on-windows`) table. ### ZLUDA From a0ad577ccb4c82a57896078889c4ed4b19fa3674 Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Fri, 17 Jul 2026 15:58:26 +0000 Subject: [PATCH 03/16] docs: update Linux AppImage runtime packages from libfuse2/libappimage to fuse3 --- docs/getting-started/overview.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/getting-started/overview.md b/docs/getting-started/overview.md index 6b45dff2b..bf1393d29 100644 --- a/docs/getting-started/overview.md +++ b/docs/getting-started/overview.md @@ -43,7 +43,7 @@ Stability Matrix is cross-platform, but the exact release formats and hardware t | Operating System | Version / Target | Architecture | Notes | |---|---|---|---| | Windows | Windows 10 and Windows 11 | x64 | Official release builds are published for `win-x64`. This is the broadest-supported desktop target for Stability Matrix and most package workflows. | -| Linux | Modern x86-64 desktop distributions | x64 | Official Linux releases are published for `linux-x64`, primarily as an AppImage, with an AUR package also available for Arch-based systems. Depending on the distribution, AppImage/runtime support packages such as `libfuse2`, `libappimage`, or `libxcrypt-compat` may be needed if they are not already provided by the system. | +| Linux | Modern x86-64 desktop distributions | x64 | Official Linux releases are published for `linux-x64`, primarily as an AppImage, with an AUR package also available for Arch-based systems. Depending on the distribution, runtime support packages such as `fuse3`, or `libxcrypt-compat` may be needed if they are not already provided by the system. | | macOS | Apple Silicon Macs, with macOS 12.3 or later recommended for AI workflows | arm64 | Official macOS releases are published for Apple Silicon (`osx-arm64`) as a `.dmg`. The app's AI workflows rely on the MPS backend on Apple Silicon. | In other words, the practical supported release targets are Windows x64, Linux x64, and Apple Silicon macOS. Some project files include additional runtime identifiers, but the documented source-build support and the release pipeline currently focus on `win-x64`, `linux-x64`, and `osx-arm64`. For work from a local checkout instead of a packaged release, a planned Building from Source and Contributing page will serve as the documentation entry point and link to the repository's contributor guide. From fe125f00f0a81c47712ee2cd265a373e25866910 Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Fri, 17 Jul 2026 19:13:01 +0000 Subject: [PATCH 04/16] docs: add Accounts settings documentation page --- docs/settings/accounts.md | 192 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 192 insertions(+) create mode 100644 docs/settings/accounts.md diff --git a/docs/settings/accounts.md b/docs/settings/accounts.md new file mode 100644 index 000000000..24c9f2b87 --- /dev/null +++ b/docs/settings/accounts.md @@ -0,0 +1,192 @@ +# Accounts + +The Accounts settings page manages external service integrations. Connecting accounts allows Stability Matrix to download models that require authentication, access membership features, and use cloud-based image generation services. + +[`Section Overview`](overview.md) | [`Home`](../README.md) + +## Table of Contents + +- [Membership and Lykos Account](#membership-and-lykos-account) +- [CivitAI API Key](#civitai-api-key) +- [HuggingFace Token](#huggingface-token) +- [Image Generation APIs](#image-generation-apis) + +--- + +## Membership and Lykos Account + +A Lykos account is the central identity for Stability Matrix. Lykos is the team behind Stability Matrix, and signing in unlocks membership benefits and account-linked features across the application. + +### Membership + +Lykos offers a membership program that supports ongoing development, access to pre-release preview builds of Stability Matrix, and unlocks additional features on the [Lykos.ai](https://lykos.ai) website. Membership is managed entirely in-house by the Lykos team at [lykos.ai/membership](https://lykos.ai/membership?ref=a1). + +| Tier | Description | +|---|---| +| **Visionary** | Highest monthly credits and vote power. Shout-out in both release and Preview/Dev changelog. Maximum support for development. | +| **Pioneer** | Higher monthly credits and vote power. Shout-out in release changelog. | +| **Insider** | Increased credits and vote power over Supporter. Accellerated Model Discovery | +| **Supporter** | 1,000 monthly credits, 2× vote power. The entry-level tier that supports development and server costs. | + +All support tiers receive special roles and dedicated supporter channel on the Stability Matrix Discord server and early access to new features through **Preview** and **Dev** update channels in [Update Settings](updates.md). + +When you are an active subscriber, the Accounts page shows your tier, the date you started supporting, and a quick link to manage your membership. A thank-you message and community Discord link are also displayed. + +If you are not subscribed, a membership card appears with a **Become a Supporter** button that opens the Lykos membership page in your browser. + +### Connecting a Lykos Account + +1. Navigate to **Settings → Accounts**. +2. Under the Lykos section, click **Connect**. +3. A device authentication dialog appears with a code. +4. Open your browser and go to the URL shown in the dialog. +5. Enter the code to authorize Stability Matrix. +6. Once authorized, the dialog closes and your account appears on the Accounts page. + +> [!NOTE] +> The device code flow does not require entering your email or password into Stability Matrix itself. Authentication happens entirely in your browser on the Lykos website. + +Your profile image is fetched from [Gravatar](https://gravatar.com/) using the email address associated with your Lykos account. You can update your avatar at any time through Gravatar, accessible from the profile image context menu. + +### Managing Your Account + +Once connected, clicking your profile image provides quick actions: + +- **Manage Lykos Account** — opens the Lykos account management page in your browser, where you can update your subscription, email, and password +- **Edit Gravatar** — opens gravatar.com to change your profile picture +- **Copy User ID** — copies your Lykos user ID to the clipboard + +### Disconnecting + +To disconnect your Lykos account, click **Disconnect** under the Lykos section. This removes your stored tokens and signs you out. Your membership can be restored by reconnecting. + +--- + +## CivitAI API Key + +A CivitAI API key enables model downloads from CivitAI that require authentication and unlocks additional integrations. + +### What the API Key Unlocks + +- **Gated model downloads** — models behind an "early-access" paywall on CivitAI +- **NSFW content browsing** — accessing NSFW-tagged models in the Model Browser (requires enabling the NSFW toggle in the Model Browser itself) + +### How to Get an API Key + +1. Log in to [civitai.com](https://civitai.com). +2. Click on avatar in upper-right. Go to **Account Settings → API Keys**. +3. Click **Create API Key** and give it a name (e.g., "Stability Matrix"). +4. Copy the generated key. Save this key somewhere safe (ie: text file, password/note manager, etc.) as CivitAI will only provide this key once. + +### Connecting in Stability Matrix + +1. Navigate to **Settings → Accounts**. +2. Under the CivitAI section, click **Connect**. +3. Paste your API key into the dialog and click **Connect**. + +Stability Matrix validates the key by fetching your CivitAI profile. Once validated, your CivitAI username appears on the Accounts page. + +### How the Key Is Used + +The API key is sent as a `Bearer` token in the `Authorization` header for all requests to `civitai.com`, including model downloads and API calls. This allows Stability Matrix to download models that would otherwise return an authentication error. + +### Security + +API keys are stored encrypted at rest in the `user-secrets.data` file inside Stability Matrix's application data directory. The key is not stored in plain text, and the file is encrypted using a key derived from system-specific identifiers. + +The `user-secrets.data` location depends on your operating system: + +| Platform | Path | +|---|---| +| Windows | `%APPDATA%\StabilityMatrix\user-secrets.data` (`C:\Users\{username}\AppData\Roaming\StabilityMatrix`) | +| Linux | `~/.config/StabilityMatrix/user-secrets.data` | +| macOS | `~/Library/Application Support/StabilityMatrix/user-secrets.data` | + +### Disconnecting + +To remove your CivitAI API key, click **Disconnect** under the CivitAI section. This deletes the stored key and stops sending authentication headers on CivitAI requests. + +--- + +## HuggingFace Token + +A HuggingFace access token enables downloading gated models from the Hugging Face Hub. + +### What the Token Unlocks + +- **Gated models** — models that require accepting a license agreement before download (e.g., FLUX,) + +### Important: Accept the Model License First + +Even with a valid token, gated models require you to accept the model's license terms on the Hugging Face website before the download will succeed. Visit the model's page on [huggingface.co](https://huggingface.co) and click **Agree and access repository** before attempting to download through Stability Matrix. + +### How to Get a Token + +1. Log in to [huggingface.co](https://huggingface.co). +2. Go to **Settings → Access Tokens**. +3. Click **Create new token**. +4. Select the **Read** permission type (sufficient for model downloads). +5. Give the token a name (e.g., "Stability Matrix"). +6. Copy the generated token. Save this key somewhere safe (ie: text file, password/note manager, etc.) as HuggingFace will only provide this key once. + +### Connecting in Stability Matrix + +1. Navigate to **Settings → Accounts**. +2. Under the HuggingFace section, click **Connect**. +3. Paste your token into the dialog and click **Connect**. + +Stability Matrix validates the token by calling the HuggingFace API. Once validated, your HuggingFace username appears on the Accounts page. + +### How the Token Is Used + +The token is used in two ways: + +1. **Direct downloads** — when Stability Matrix downloads a model directly from `huggingface.co`, the token is sent as a `Bearer` token in the `Authorization` header +2. **User-configurable environment variable** — you can inject the token into launched packages by setting `HF_TOKEN` in the [Environment Variables editor](../advanced/environment-variables.md). Stability Matrix does **not** automatically inject the token into package environments. + +### Security + +The token is stored encrypted at rest in the `user-secrets.data` file inside Stability Matrix's application data directory, using the same encryption mechanism as the CivitAI API key. + +### Disconnecting + +To remove your HuggingFace token, click **Disconnect** under the HuggingFace section. This deletes the stored token. Direct downloads from Hugging Face will no longer include authentication headers. + +> [!IMPORTANT] +> If you previously set `HF_TOKEN` manually in the Environment Variables editor, disconnecting here does **not** remove it from there. You must remove it from the Environment Variables editor separately. + +--- + +## Image Generation APIs + +The Image Generation APIs section contains keys for cloud-based image generation services used by the **Image Lab** feature. + +### Gemini API Key + +A Gemini API key enables the **Nano Banana** image generation provider in the Image Lab, which uses Google's Gemini models for AI-powered image generation. + +> [!NOTE] +> The Gemini API key requires a paid-tier Google AI account with billing enabled. Free-tier keys will not work for Image Lab generation. + +#### How to Get an API Key + +1. Go to [Google AI Studio](https://aistudio.google.com/api-keys). +2. Sign in with your Google account. +3. Click **Create API Key**. +4. Copy the generated key. + +#### Connecting in Stability Matrix + +1. Navigate to **Settings → Accounts**. +2. Under the **Image Generation APIs** section, click **Connect** next to Gemini API. +3. Paste your API key into the dialog and click **Connect**. + +Stability Matrix checks that a key is present. The key is validated at generation time when the Image Lab makes its first request. + +#### Security + +The API key is stored encrypted at rest in the `user-secrets.data` file inside Stability Matrix's application data directory, using the same encryption mechanism as other stored credentials. + +#### Disconnecting + +To remove your Gemini API key, click **Disconnect** under the Gemini API section. This deletes the stored key. The Nano Banana provider in the Image Lab will no longer be available until a new key is added. From 0c55bd084c009757e9422263c104f6329db78744 Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Fri, 17 Jul 2026 22:34:57 +0000 Subject: [PATCH 05/16] docs: add General settings documentation page --- docs/settings/general.md | 112 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 112 insertions(+) create mode 100644 docs/settings/general.md diff --git a/docs/settings/general.md b/docs/settings/general.md new file mode 100644 index 000000000..851a7c719 --- /dev/null +++ b/docs/settings/general.md @@ -0,0 +1,112 @@ +# General Settings + +General settings is the first section in the Settings view. It provides quick navigation to the two primary configuration areas: **Inference** and **Notifications**. + +[`Section Overview`](overview.md) | [`Home`](../README.md) + +## Table of Contents + +- [Inference](#inference) +- [Notifications](#notifications) + +--- + +## Inference + +Inference settings control the behavior of Stability Matrix's built-in image generation and prompt editing tools. These settings are organized into three subsections: **Prompt**, **General**, and **Dimensions**. + +### Prompt + +#### Auto Completion + +The auto completion feature provides tag suggestions as you type in prompt and negative prompt text boxes. When enabled, a dropdown appears with matching tags from a configurable CSV file. + +| Setting | Description | +|---|---| +| **Enable Auto Completion** | Toggle the tag suggestion feature on or off. | +| **Prompt Tags** | Select which CSV tag file to use for completions. Custom tag CSVs can be imported via the **Import** button below this setting. | +| **Replace underscores with spaces** | When enabled, underscores in completed tags are displayed as spaces (e.g., `oil_painting` shows as `oil painting`). | + +> [!NOTE] +> Imported CSVs are copied to the `Tags/` folder inside the data directory. After importing, the new file appears in the **Prompt Tags** dropdown for selection. + +--- + +### General + +#### Filter Extra Networks by Base Model + +When enabled, LoRA and other extra network model lists in the Inference UI are filtered to only show models compatible with the currently selected base model (e.g., SDXL LoRAs when SDXL is selected). This reduces clutter and helps avoid loading incompatible models. + +#### Image Viewer + +| Setting | Description | +|---|---| +| **Show pixel grid at high zoom levels** | When enabled, a pixel grid overlay appears when zooming into generated images, making it easier to inspect fine details. | +| **Move files to the Recycle Bin when deleting** | When enabled, deleted images are sent to the system Recycle Bin / Trash instead of being permanently erased. This option is only visible on platforms where the Recycle Bin is available. | + +#### Output Image Files + +Controls how generated images are named when saved. + +**File name pattern** — A format string that defines the naming convention for saved output images. Uses curly-brace variables like `{model_name}` and `{seed}`. Clicking into the text field shows a tooltip with all available format variables and examples. The default pattern is the same as the Inference UI template: + +``` +{model_name}_{seed}_{width}x{height} +``` + +A live preview of the resulting filename is shown below the input. + +#### Floating Search + +When enabled, all combo boxes in the Inference UI use a floating search mode: typed text appears as an overlay and auto-selects the first match. When disabled, each combo box uses its own configured search behavior. + +--- + +### Dimensions + +#### Step Size + +Controls how much the width and height values increase or decrease per step when using the arrow buttons or scroll wheel on dimension input fields. The value is in pixels and can range from 8 to 1024. + +#### Favorite Dimensions + +A configurable list of frequently used width × height pairs. These appear as quick-select options in the Inference UI's dimensions section, letting you switch between common resolutions without typing. + +| Action | How | +|---|---| +| **Add** | Click **+ Add**, enter a width and height, then save. | +| **Remove** | Select a row in the grid and click **Remove**. | + +The grid supports sorting by dimension value. Favorite dimensions are saved automatically and persist across restarts. + +--- + +## Notifications + +The Notifications page controls how Stability Matrix alerts you about events such as completed downloads, finished inference prompts, and package installations. + +Each notification type can be independently configured to use one of three channels: + +| Channel | Description | +|---|---| +| **None** | No notification is sent for this event. | +| **In-App** | A toast notification appears inside the Stability Matrix window. | +| **Desktop** | A native OS push notification is sent (e.g., Windows Action Center, macOS Notification Center). | + +> [!NOTE] +> On Linux, native push notifications depend on the desktop environment's notification daemon. The default for Linux is **In-App** since native push support varies across distributions. + +### Available Notification Events + +| Event | Description | Default Channel | +|---|---|---| +| **Inference Prompt Completed** | A single prompt has finished generating. | Native Push (In-App on Linux) | +| **Inference Batch Completed** | An entire batch of prompts has finished. | Native Push (In-App on Linux) | +| **Download Completed** | A model or file download has finished successfully. | Native Push (In-App on Linux) | +| **Download Failed** | A download has failed due to an error. | Native Push (In-App on Linux) | +| **Download Canceled** | A download was manually canceled by the user. | Native Push (In-App on Linux) | +| **Package Install Completed** | A package install has finished successfully. | Native Push (In-App on Linux) | +| **Package Install Failed** | A package install has failed. | Native Push (In-App on Linux) | + +Each event row shows the event name and a dropdown to select the notification channel. Changes take effect immediately and are saved automatically. From ee0041f34c7618acaaf5fa1bb4b2dd1d8c5840b3 Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Sat, 18 Jul 2026 01:24:09 +0000 Subject: [PATCH 06/16] docs: deleted General and Accounts settings docfiles. All Settings documentation consolidated into single file --- docs/settings/accounts.md | 192 --------------------- docs/settings/general.md | 112 ------------- docs/settings/settings.md | 342 ++++++++++++++++++++++++++++++++++++++ 3 files changed, 342 insertions(+), 304 deletions(-) delete mode 100644 docs/settings/accounts.md delete mode 100644 docs/settings/general.md create mode 100644 docs/settings/settings.md diff --git a/docs/settings/accounts.md b/docs/settings/accounts.md deleted file mode 100644 index 24c9f2b87..000000000 --- a/docs/settings/accounts.md +++ /dev/null @@ -1,192 +0,0 @@ -# Accounts - -The Accounts settings page manages external service integrations. Connecting accounts allows Stability Matrix to download models that require authentication, access membership features, and use cloud-based image generation services. - -[`Section Overview`](overview.md) | [`Home`](../README.md) - -## Table of Contents - -- [Membership and Lykos Account](#membership-and-lykos-account) -- [CivitAI API Key](#civitai-api-key) -- [HuggingFace Token](#huggingface-token) -- [Image Generation APIs](#image-generation-apis) - ---- - -## Membership and Lykos Account - -A Lykos account is the central identity for Stability Matrix. Lykos is the team behind Stability Matrix, and signing in unlocks membership benefits and account-linked features across the application. - -### Membership - -Lykos offers a membership program that supports ongoing development, access to pre-release preview builds of Stability Matrix, and unlocks additional features on the [Lykos.ai](https://lykos.ai) website. Membership is managed entirely in-house by the Lykos team at [lykos.ai/membership](https://lykos.ai/membership?ref=a1). - -| Tier | Description | -|---|---| -| **Visionary** | Highest monthly credits and vote power. Shout-out in both release and Preview/Dev changelog. Maximum support for development. | -| **Pioneer** | Higher monthly credits and vote power. Shout-out in release changelog. | -| **Insider** | Increased credits and vote power over Supporter. Accellerated Model Discovery | -| **Supporter** | 1,000 monthly credits, 2× vote power. The entry-level tier that supports development and server costs. | - -All support tiers receive special roles and dedicated supporter channel on the Stability Matrix Discord server and early access to new features through **Preview** and **Dev** update channels in [Update Settings](updates.md). - -When you are an active subscriber, the Accounts page shows your tier, the date you started supporting, and a quick link to manage your membership. A thank-you message and community Discord link are also displayed. - -If you are not subscribed, a membership card appears with a **Become a Supporter** button that opens the Lykos membership page in your browser. - -### Connecting a Lykos Account - -1. Navigate to **Settings → Accounts**. -2. Under the Lykos section, click **Connect**. -3. A device authentication dialog appears with a code. -4. Open your browser and go to the URL shown in the dialog. -5. Enter the code to authorize Stability Matrix. -6. Once authorized, the dialog closes and your account appears on the Accounts page. - -> [!NOTE] -> The device code flow does not require entering your email or password into Stability Matrix itself. Authentication happens entirely in your browser on the Lykos website. - -Your profile image is fetched from [Gravatar](https://gravatar.com/) using the email address associated with your Lykos account. You can update your avatar at any time through Gravatar, accessible from the profile image context menu. - -### Managing Your Account - -Once connected, clicking your profile image provides quick actions: - -- **Manage Lykos Account** — opens the Lykos account management page in your browser, where you can update your subscription, email, and password -- **Edit Gravatar** — opens gravatar.com to change your profile picture -- **Copy User ID** — copies your Lykos user ID to the clipboard - -### Disconnecting - -To disconnect your Lykos account, click **Disconnect** under the Lykos section. This removes your stored tokens and signs you out. Your membership can be restored by reconnecting. - ---- - -## CivitAI API Key - -A CivitAI API key enables model downloads from CivitAI that require authentication and unlocks additional integrations. - -### What the API Key Unlocks - -- **Gated model downloads** — models behind an "early-access" paywall on CivitAI -- **NSFW content browsing** — accessing NSFW-tagged models in the Model Browser (requires enabling the NSFW toggle in the Model Browser itself) - -### How to Get an API Key - -1. Log in to [civitai.com](https://civitai.com). -2. Click on avatar in upper-right. Go to **Account Settings → API Keys**. -3. Click **Create API Key** and give it a name (e.g., "Stability Matrix"). -4. Copy the generated key. Save this key somewhere safe (ie: text file, password/note manager, etc.) as CivitAI will only provide this key once. - -### Connecting in Stability Matrix - -1. Navigate to **Settings → Accounts**. -2. Under the CivitAI section, click **Connect**. -3. Paste your API key into the dialog and click **Connect**. - -Stability Matrix validates the key by fetching your CivitAI profile. Once validated, your CivitAI username appears on the Accounts page. - -### How the Key Is Used - -The API key is sent as a `Bearer` token in the `Authorization` header for all requests to `civitai.com`, including model downloads and API calls. This allows Stability Matrix to download models that would otherwise return an authentication error. - -### Security - -API keys are stored encrypted at rest in the `user-secrets.data` file inside Stability Matrix's application data directory. The key is not stored in plain text, and the file is encrypted using a key derived from system-specific identifiers. - -The `user-secrets.data` location depends on your operating system: - -| Platform | Path | -|---|---| -| Windows | `%APPDATA%\StabilityMatrix\user-secrets.data` (`C:\Users\{username}\AppData\Roaming\StabilityMatrix`) | -| Linux | `~/.config/StabilityMatrix/user-secrets.data` | -| macOS | `~/Library/Application Support/StabilityMatrix/user-secrets.data` | - -### Disconnecting - -To remove your CivitAI API key, click **Disconnect** under the CivitAI section. This deletes the stored key and stops sending authentication headers on CivitAI requests. - ---- - -## HuggingFace Token - -A HuggingFace access token enables downloading gated models from the Hugging Face Hub. - -### What the Token Unlocks - -- **Gated models** — models that require accepting a license agreement before download (e.g., FLUX,) - -### Important: Accept the Model License First - -Even with a valid token, gated models require you to accept the model's license terms on the Hugging Face website before the download will succeed. Visit the model's page on [huggingface.co](https://huggingface.co) and click **Agree and access repository** before attempting to download through Stability Matrix. - -### How to Get a Token - -1. Log in to [huggingface.co](https://huggingface.co). -2. Go to **Settings → Access Tokens**. -3. Click **Create new token**. -4. Select the **Read** permission type (sufficient for model downloads). -5. Give the token a name (e.g., "Stability Matrix"). -6. Copy the generated token. Save this key somewhere safe (ie: text file, password/note manager, etc.) as HuggingFace will only provide this key once. - -### Connecting in Stability Matrix - -1. Navigate to **Settings → Accounts**. -2. Under the HuggingFace section, click **Connect**. -3. Paste your token into the dialog and click **Connect**. - -Stability Matrix validates the token by calling the HuggingFace API. Once validated, your HuggingFace username appears on the Accounts page. - -### How the Token Is Used - -The token is used in two ways: - -1. **Direct downloads** — when Stability Matrix downloads a model directly from `huggingface.co`, the token is sent as a `Bearer` token in the `Authorization` header -2. **User-configurable environment variable** — you can inject the token into launched packages by setting `HF_TOKEN` in the [Environment Variables editor](../advanced/environment-variables.md). Stability Matrix does **not** automatically inject the token into package environments. - -### Security - -The token is stored encrypted at rest in the `user-secrets.data` file inside Stability Matrix's application data directory, using the same encryption mechanism as the CivitAI API key. - -### Disconnecting - -To remove your HuggingFace token, click **Disconnect** under the HuggingFace section. This deletes the stored token. Direct downloads from Hugging Face will no longer include authentication headers. - -> [!IMPORTANT] -> If you previously set `HF_TOKEN` manually in the Environment Variables editor, disconnecting here does **not** remove it from there. You must remove it from the Environment Variables editor separately. - ---- - -## Image Generation APIs - -The Image Generation APIs section contains keys for cloud-based image generation services used by the **Image Lab** feature. - -### Gemini API Key - -A Gemini API key enables the **Nano Banana** image generation provider in the Image Lab, which uses Google's Gemini models for AI-powered image generation. - -> [!NOTE] -> The Gemini API key requires a paid-tier Google AI account with billing enabled. Free-tier keys will not work for Image Lab generation. - -#### How to Get an API Key - -1. Go to [Google AI Studio](https://aistudio.google.com/api-keys). -2. Sign in with your Google account. -3. Click **Create API Key**. -4. Copy the generated key. - -#### Connecting in Stability Matrix - -1. Navigate to **Settings → Accounts**. -2. Under the **Image Generation APIs** section, click **Connect** next to Gemini API. -3. Paste your API key into the dialog and click **Connect**. - -Stability Matrix checks that a key is present. The key is validated at generation time when the Image Lab makes its first request. - -#### Security - -The API key is stored encrypted at rest in the `user-secrets.data` file inside Stability Matrix's application data directory, using the same encryption mechanism as other stored credentials. - -#### Disconnecting - -To remove your Gemini API key, click **Disconnect** under the Gemini API section. This deletes the stored key. The Nano Banana provider in the Image Lab will no longer be available until a new key is added. diff --git a/docs/settings/general.md b/docs/settings/general.md deleted file mode 100644 index 851a7c719..000000000 --- a/docs/settings/general.md +++ /dev/null @@ -1,112 +0,0 @@ -# General Settings - -General settings is the first section in the Settings view. It provides quick navigation to the two primary configuration areas: **Inference** and **Notifications**. - -[`Section Overview`](overview.md) | [`Home`](../README.md) - -## Table of Contents - -- [Inference](#inference) -- [Notifications](#notifications) - ---- - -## Inference - -Inference settings control the behavior of Stability Matrix's built-in image generation and prompt editing tools. These settings are organized into three subsections: **Prompt**, **General**, and **Dimensions**. - -### Prompt - -#### Auto Completion - -The auto completion feature provides tag suggestions as you type in prompt and negative prompt text boxes. When enabled, a dropdown appears with matching tags from a configurable CSV file. - -| Setting | Description | -|---|---| -| **Enable Auto Completion** | Toggle the tag suggestion feature on or off. | -| **Prompt Tags** | Select which CSV tag file to use for completions. Custom tag CSVs can be imported via the **Import** button below this setting. | -| **Replace underscores with spaces** | When enabled, underscores in completed tags are displayed as spaces (e.g., `oil_painting` shows as `oil painting`). | - -> [!NOTE] -> Imported CSVs are copied to the `Tags/` folder inside the data directory. After importing, the new file appears in the **Prompt Tags** dropdown for selection. - ---- - -### General - -#### Filter Extra Networks by Base Model - -When enabled, LoRA and other extra network model lists in the Inference UI are filtered to only show models compatible with the currently selected base model (e.g., SDXL LoRAs when SDXL is selected). This reduces clutter and helps avoid loading incompatible models. - -#### Image Viewer - -| Setting | Description | -|---|---| -| **Show pixel grid at high zoom levels** | When enabled, a pixel grid overlay appears when zooming into generated images, making it easier to inspect fine details. | -| **Move files to the Recycle Bin when deleting** | When enabled, deleted images are sent to the system Recycle Bin / Trash instead of being permanently erased. This option is only visible on platforms where the Recycle Bin is available. | - -#### Output Image Files - -Controls how generated images are named when saved. - -**File name pattern** — A format string that defines the naming convention for saved output images. Uses curly-brace variables like `{model_name}` and `{seed}`. Clicking into the text field shows a tooltip with all available format variables and examples. The default pattern is the same as the Inference UI template: - -``` -{model_name}_{seed}_{width}x{height} -``` - -A live preview of the resulting filename is shown below the input. - -#### Floating Search - -When enabled, all combo boxes in the Inference UI use a floating search mode: typed text appears as an overlay and auto-selects the first match. When disabled, each combo box uses its own configured search behavior. - ---- - -### Dimensions - -#### Step Size - -Controls how much the width and height values increase or decrease per step when using the arrow buttons or scroll wheel on dimension input fields. The value is in pixels and can range from 8 to 1024. - -#### Favorite Dimensions - -A configurable list of frequently used width × height pairs. These appear as quick-select options in the Inference UI's dimensions section, letting you switch between common resolutions without typing. - -| Action | How | -|---|---| -| **Add** | Click **+ Add**, enter a width and height, then save. | -| **Remove** | Select a row in the grid and click **Remove**. | - -The grid supports sorting by dimension value. Favorite dimensions are saved automatically and persist across restarts. - ---- - -## Notifications - -The Notifications page controls how Stability Matrix alerts you about events such as completed downloads, finished inference prompts, and package installations. - -Each notification type can be independently configured to use one of three channels: - -| Channel | Description | -|---|---| -| **None** | No notification is sent for this event. | -| **In-App** | A toast notification appears inside the Stability Matrix window. | -| **Desktop** | A native OS push notification is sent (e.g., Windows Action Center, macOS Notification Center). | - -> [!NOTE] -> On Linux, native push notifications depend on the desktop environment's notification daemon. The default for Linux is **In-App** since native push support varies across distributions. - -### Available Notification Events - -| Event | Description | Default Channel | -|---|---|---| -| **Inference Prompt Completed** | A single prompt has finished generating. | Native Push (In-App on Linux) | -| **Inference Batch Completed** | An entire batch of prompts has finished. | Native Push (In-App on Linux) | -| **Download Completed** | A model or file download has finished successfully. | Native Push (In-App on Linux) | -| **Download Failed** | A download has failed due to an error. | Native Push (In-App on Linux) | -| **Download Canceled** | A download was manually canceled by the user. | Native Push (In-App on Linux) | -| **Package Install Completed** | A package install has finished successfully. | Native Push (In-App on Linux) | -| **Package Install Failed** | A package install has failed. | Native Push (In-App on Linux) | - -Each event row shows the event name and a dropdown to select the notification channel. Changes take effect immediately and are saved automatically. diff --git a/docs/settings/settings.md b/docs/settings/settings.md new file mode 100644 index 000000000..c5d90f30c --- /dev/null +++ b/docs/settings/settings.md @@ -0,0 +1,342 @@ +# Settings + +The Settings page is the central configuration hub for Stability Matrix. It is organized into categories that cover general application behavior, integrations, package management, browsing, appearance, and system-level configuration. These settings can be accessed via the Settings link in the bottom-left of the Stability Matrix UI. + +[`Home`](../README.md) + +## Table of Contents + +- [General](#general) +- [Integrations](#integrations) +- [Checkpoint Manager](#checkpoint-manager) +- [Package Environment](#package-environment) +- [Model Browser](#model-browser) +- [Workflow Browser](#workflow-browser) +- [Console](#console) +- [Appearance](#appearance) +- [System](#system) +- [About](#about) +- [Debug Options](#debug-options) + +--- + +## General + +Controls Inference UI behavior, prompt auto-completion, and notification preferences. + +### Inference + +Inference settings control the behavior of Stability Matrix's built-in image generation and prompt editing tools. They are organized into three groups: **Prompt**, **General**, and **Dimensions**. + +#### Prompt + +**Auto Completion** provides tag suggestions as you type in prompt and negative prompt text boxes. When enabled, a dropdown appears with matching tags from a configurable CSV file. + +| Setting | Description | +|---|---| +| **Enable Auto Completion** | Toggle the tag suggestion feature on or off. | +| **Prompt Tags** | Select which CSV tag file to use for completions. Custom tag CSVs can be imported via the **Import** button. Imported CSVs are copied to the `Tags/` folder inside the data directory. | +| **Replace underscores with spaces** | When enabled, underscores in completed tags are displayed as spaces (e.g., `oil_painting` shows as `oil painting`). | + +#### General + +| Setting | Description | +|---|---| +| **Filter Extra Networks by Base Model** | When enabled, LoRA and extra network model lists are filtered to only show models compatible with the currently selected base model (e.g., SDXL LoRAs when SDXL is selected). | +| **Image Viewer — Show pixel grid at high zoom** | When enabled, a pixel grid overlay appears when zooming into generated images. | +| **Image Viewer — Move files to Recycle Bin** | When enabled, deleted images go to the system Recycle Bin / Trash instead of being permanently erased. Only shown on platforms where Recycle Bin is available. | +| **Output Image Files — File name pattern** | A format string for naming saved output images, using variables like `{model_name}` and `{seed}`. Clicking the text field shows all available variables. Default: `{model_name}_{seed}_{width}x{height}`. | +| **Floating Search** | When enabled, all combo boxes use floating search mode where typed text appears as an overlay and auto-selects the first match. | + +#### Dimensions + +| Setting | Description | +|---|---| +| **Step Size** | How much width/height values change per step when using arrow buttons or scroll wheel on dimension inputs. Range: 8–1024 pixels. | +| **Favorite Dimensions** | A configurable list of frequently used width × height pairs that appear as quick-select options in the Inference UI. Add via the **+ Add** button, remove by selecting a row and clicking **Remove**. Saved automatically. | + +### Notifications + +Controls how Stability Matrix alerts you about events. Each notification type can be independently set to one of three channels: + +| Channel | Description | +|---|---| +| **None** | No notification is sent. | +| **In-App** | A toast notification appears inside the Stability Matrix window. | +| **Desktop** | A native OS push notification (Windows Action Center, macOS Notification Center). | + +> [!NOTE] +> On Linux, the default channel is **In-App** for all events since native push support varies across desktop environments. + +| Event | Default | +|---|---| +| Inference Prompt Completed | Desktop (In-App on Linux) | +| Inference Batch Completed | Desktop (In-App on Linux) | +| Download Completed | Desktop (In-App on Linux) | +| Download Failed | Desktop (In-App on Linux) | +| Download Canceled | Desktop (In-App on Linux) | +| Package Install Completed | Desktop (In-App on Linux) | +| Package Install Failed | Desktop (In-App on Linux) | + +--- + +## Integrations + +The Integrations section connects Stability Matrix to external services and platforms. + +### Accounts + +The Accounts sub-page manages external service credentials. Connecting accounts allows Stability Matrix to download models that require authentication, access membership features, and use cloud-based image generation services. + +#### Membership and Lykos Account + +A Lykos account is the central identity for Stability Matrix. Signing in unlocks membership benefits and account-linked features. Membership is managed at [lykos.ai/membership](https://lykos.ai/membership?). + +| Tier | Description | +|---|---| +| **Visionary** | Highest monthly credits and vote power. Shout-out in both release and Preview/Dev changelog. Maximum support for development. | +| **Pioneer** | Higher monthly credits and vote power. Shout-out in release changelog. | +| **Insider** | Increased credits and vote power over Supporter. Accelerated Model Discovery | +| **Supporter** | 1,000 monthly credits, 2× vote power. The entry-level tier that supports development and server costs. | + +All support tiers receive early access to new features through **Preview** and **Dev** update channels. + +When you are an active subscriber, the Accounts page shows your tier, the date you started supporting, and a quick link to manage your membership. A thank-you message and community Discord link are also displayed. + +If you are not subscribed, a membership card appears with a **Become a Supporter** button that opens the Lykos membership page in your browser. + +**Connecting**: Click **Connect** under the Lykos section. A device authentication dialog appears with a code. Open the URL in your browser, enter the code, and authorize. Authentication happens entirely in your browser — no credentials are entered into Stability Matrix. + +Your profile image is fetched from [Gravatar](https://gravatar.com/) using your Lykos account email. Click the image to manage your Lykos account, edit your Gravatar, or copy your user ID. + +#### CivitAI API Key + +Enables downloading early-access models from CivitAI and browsing NSFW-tagged models (requires enabling the NSFW toggle in the Model Browser). + +**To connect**: Log in to [civitai.com](https://civitai.com), go to **Account Settings → API Keys**, create a key, and paste it into the Connect dialog in Stability Matrix. The key is validated by fetching your CivitAI profile. + +The API key is sent as a `Bearer` token in the `Authorization` header for all requests to `civitai.com`, including model downloads. + +#### HuggingFace Token + +Enables downloading gated models (e.g., FLUX) from the Hugging Face Hub. You must also accept the model's license terms on the Hugging Face website before downloading. + +**To connect**: Log in to [huggingface.co](https://huggingface.co), go to **Settings → Access Tokens**, create a **Read** token (starts with `hf_`), and paste it into the Connect dialog. Stability Matrix validates the token by calling the HuggingFace API. + +The token is used in two ways: sent as a `Bearer` header for direct downloads from `huggingface.co`, and available for manual injection into package environments via `HF_TOKEN` in the Environment Variables editor. + +> [!IMPORTANT] +> Disconnecting the token here does **not** remove a manually-set `HF_TOKEN` from the Environment Variables editor. Remove it from both places if needed. + +#### Gemini API Key + +Enables the **Nano Banana** image generation provider in the **Image Lab**, which uses Google's Gemini models. Requires a paid-tier Google AI account with billing enabled. + +**To connect**: Go to [Google AI Studio](https://aistudio.google.com/api-keys), create an API key, and paste it into the Connect dialog under the **Image Generation APIs** section. The key is validated at generation time when the Image Lab makes its first request. + +> [!IMPORTANT] +> When API keys are generated on their respective service website, it is recommended to save your API keys somewhere safe (ie: text file, password/note manager, etc.) as these services will typically only provide this key once. + +#### How Credentials Are Stored + +All API keys and tokens are stored encrypted at rest in the `user-secrets.data` file. Keys are never written to plaintext. The file is encrypted using a key derived from system-specific identifiers, meaning it can only be decrypted on the same machine that wrote it. + +The `user-secrets.data` location depends on your operating system: + +| Platform | Path | +|---|---| +| Windows | `%APPDATA%\StabilityMatrix\user-secrets.data` (`C:\Users\{username}\AppData\Roaming\StabilityMatrix`) | +| Linux | `~/.config/StabilityMatrix/user-secrets.data` | +| macOS | `~/Library/Application Support/StabilityMatrix/user-secrets.data` | + +This is the fixed application data directory — distinct from the user-configurable library / data directory where packages and models are stored. + +### Discord Rich Presence + +When enabled, your Discord status shows that you are using Stability Matrix. + +--- + +## Checkpoint Manager + +Controls how model files are handled when imported and how shared model folders are managed. + +| Setting | Description | +|---|---| +| **Import Behaviour** | Choose whether to **Copy** or **Move** files when dragging and dropping them into the Checkpoint Manager. Copy leaves the original in place; Move relocates it. | +| **Remove Symlinks on Shutdown** | When enabled, Stability Matrix removes shared model symlinks and junction points when the application closes. | +| **Reset Checkpoints Cache** | Clears the cached checkpoint index. Use this if models are not appearing correctly or if the index becomes out of date. | + +--- + +## Package Environment + +Manages environment variables, Python tooling, and Git configuration used when launching packages. + +### Environment Variables + +Opens the global environment variable editor. Variables set here are injected into every launched package's process environment. This is useful for configuring PyTorch behavior, setting cache locations, or adding API tokens without editing scripts. + +Variables are defined as name/value pairs. Click **Edit** to open the editor, then click **+** to add a new row. Changes apply to future package launches — restart any running packages for new variables to take effect. + +Common uses include `PYTORCH_ALLOC_CONF` for GPU memory tuning, `HF_TOKEN` for HuggingFace authentication in packages, `CUDA_VISIBLE_DEVICES` for GPU isolation, and `PATH` modifications. See the [Environment Variables guide](../advanced/environment-variables.md) for a full reference. + +> [!NOTE] +> User-set environment variables take priority over any defaults Stability Matrix applies automatically at package launch. + +### Embedded Python + +Stability Matrix bundles its own Python installation (Python 3.10) stored in the `Assets/Python310/` folder inside the data directory. This embedded Python is separate from the various standalone Python versions used by package virtual environments and is used exclusively for Stability Matrix's own internal operations. + +Internally, the embedded Python powers: + +- **Virtual environment bootstrapping** — When setting up a package for the first time, the embedded Python is used to install `pip` and `virtualenv` into itself via Python.NET. These tools are then invoked to create the package's own isolated virtual environment. This same bootstrap process runs during one-click installs, data directory migration, and standard package installs. +- **Python version inspection** — The **Check Version** button in Settings uses Python.NET to initialize the embedded Python runtime in-process and retrieve detailed version information. + +The embedded Python section in Settings displays the current version and provides actions to run an arbitrary Python command utilizing the embedded Python, clear the Pip cache, or clear the uv cache. + +### Git + +Runs Git commands against the bundled Git installation. On Windows, also provides an option to enable long path support in Git. + +### Show Unsupported Python Versions + +When enabled, Python versions outside the supported range for a package are still shown as installation options. Typically left disabled unless troubleshooting. + +--- + +## Model Browser + +Configures the behavior of the CivitAI model browser. + +| Setting | Description | +|---|---| +| **Auto Search on Load** | When enabled, the Model Browser automatically runs a search with your last-used sort, period, and model type filters when the page opens. Results are sorted by Highest Rated by default. When disabled, you must manually trigger a search. Previous search parameters are restored from your last session regardless of this setting. | +| **Base Model Filter** | Toggle which base model types (SD1.5, SDXL, Pony, Illustrious, Anima, etc.) are shown or hidden in the `Base Model` filter for the Model Browser. | + +--- + +## Workflow Browser + +| Setting | Description | +|---|---| +| **Infinite Scrolling** | When enabled, the Workflow Browser continuously loads more results as you scroll. When disabled, results are paginated. | + +--- + +## Console + +| Setting | Description | +|---|---| +| **History Size** | Maximum number of log entries retained in the Console view. Set to `-1` for unlimited or `0` to disable history. (Default: 9001) | + +--- + +## Appearance + +| Setting | Description | +|---|---| +| **Theme** | Switch between Light, Dark, and System themes. Changes take effect immediately. | +| **Language** | Select the UI language from available community-contributed translations. | +| **Number Format** | Choose how numbers are formatted. Options include default system format or forced period/comma decimal separators. A live preview is shown in the dropdown. | +| **Holiday Mode** | Toggles a festive santa hat overlay on Model Browser card images. **Automatic** shows hats during December, **Enabled** forces them year-round, **Disabled** never shows them. | +| **Always Show Scrollbars** | When enabled, scrollbars are always visible. When disabled, they appear only while scrolling. | + +--- + +## System + +System-level settings for updates, data management, hardware preferences, and diagnostic information. + +### Updates + +| Setting | Description | +|---|---| +| **Update Channel** | Choose between **Stable** (tested releases), **Preview** (newer features, minor issues possible), and **Dev** (latest development builds). Preview and Dev channels require an active Lykos membership. | +| **Auto-Update Frequency** | How often Stability Matrix checks for updates: on startup, daily, or never (manual only). A notification appears when an update is available; updates are never installed without confirmation. | + +### Analytics + +When enabled, Stability Matrix sends anonymous usage data to Lykos to help prioritize development. The data collected is minimal: + +- **Package installs** — package name, version, and whether the install succeeded or failed +- **First-time setup** — which package was selected, which recommended models were chosen, and whether setup was skipped + +No hardware information, model browsing history, personal data, or API keys are included. Additionally, each app launch sends the Stability Matrix version, OS description, and runtime identifier to the analytics API at most once per day. + +This setting is also presented during first-run setup. Analytics are enabled by default (opt-out); you can change your choice here at any time. + +### Add to Start Menu + +(Windows only) Adds Stability Matrix to the Start Menu for the current user or all users. + +### Maximum Simultaneous Downloads + +Limits the number of concurrent model downloads. Does not apply to package installs. Set to `0` for unlimited. + +### Select New Data Directory + +Change the library location where Stability Matrix stores packages, models, images, and settings. Displays the current path. Changing the path does not move existing data — you must relocate files manually. See [Data Directory](../getting-started/data-directory.md#changing-the-data-directory-later) for further details. + +### Select New Models Folder + +Change the shared models directory independently from the main data directory. Useful for keeping models on a separate drive. + +### App Folders + +Quick-access buttons to open key Stability Matrix folders in your file manager: Application Data directory, Logs, Data directory, Checkpoints (Models) shared directory, and Packages directory. + +### System Settings + +| Setting | Description | +|---|---| +| **Default GPU** | Select which GPU Stability Matrix should prefer for package installs. This influences which PyTorch backend index is automatically selected. | +| **Enable Long Paths** | (Windows only) Enables Windows long path support in the registry to avoid path-length errors with deeply nested package or model folders. | + +### System Information + +Read-only diagnostic information about your hardware: + +- **CPU** — processor model +- **Memory** — total installed and available RAM +- **GPU** — detected graphics cards with driver version and VRAM + +--- + +## About + +The About section appears at the bottom of the Settings view. It shows: + +- The Stability Matrix app icon and name +- The current version number (e.g., `Version 2.16.1`) +- A **License and Open Source Notices** button that opens a dialog listing third-party licenses for bundled open-source components + +Tapping the version number repeatedly (7 times) enables hidden debug options in the Settings UI. + +## Debug Options + +When debug mode is enabled, a **Debug Options** section appears in the Settings view. It contains diagnostic tools and utilities intended for development and troubleshooting: + +### Diagnostics + +| Item | Description | +|---|---| +| **Paths** | Displays the current working directory, app directory, base directory, and AppData path. | +| **Compat Info** | Shows platform, portable mode status, library directory state, and other environment details. | +| **GPU Info** | Lists all detected GPUs with their properties. | + +### Utilities + +| Item | Description | +|---|---| +| **Animation Scale** | Adjusts UI animation speed from 0× (instant) to 2×. Lower values make the UI feel snappier during testing. | +| **Notification** | Fires a test in-app notification to verify the notification system. | +| **Content Dialog** | Opens a test content dialog to verify dialog rendering. | +| **Exceptions** | Buttons to trigger unhandled and dispatcher exceptions for testing error handling and crash reporting. | +| **Download Manager tests** | Adds a test tracked download entry. | +| **Refresh Models Index** | Manually refreshes the local model index used by the Checkpoint Manager. | +| **Make image grid** | Opens a file picker to select images and generates a combined grid image. | +| **Image metadata parser** | Chooses an image and parses its metadata for debugging. | +| **Gemini error dialogs** | Previews the various Image Lab / Nano Banana error states (key not configured, invalid key, quota exceeded, access forbidden, generic failure) without triggering a real API call. | + +Additional debug commands may appear dynamically at the bottom of the section. From 2eeb43f4046b51feddc0c4a406802822bc37a72e Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Sat, 18 Jul 2026 17:39:37 +0000 Subject: [PATCH 07/16] docs: remove TheRock preview branding from ROCm backend description in installing-packages doc --- docs/package-manager/installing-packages.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/package-manager/installing-packages.md b/docs/package-manager/installing-packages.md index be594f12b..a060a36e6 100644 --- a/docs/package-manager/installing-packages.md +++ b/docs/package-manager/installing-packages.md @@ -96,7 +96,7 @@ The **PyTorch backend** determines which GPU acceleration library your package u | Backend | Platform | GPU | Notes | |---------|----------|-----|-------| | **CUDA** | Windows, Linux | NVIDIA (GTX 900-series and newer) | Best performance and broadest compatibility. CUDA toolkit is bundled with PyTorch; no separate driver installation beyond standard NVIDIA drivers. Turing (RTX 2000-series) or newer recommended. | -| **ROCm** | Windows, Linux | AMD (select GPUs per platform) | Native AMD GPU acceleration. On Linux, requires system-level ROCm installation. On Windows, uses AMD's TheRock technical preview builds. Per-chip compatibility details are planned for a future Hardware Support page. | +| **ROCm** | Windows, Linux | AMD (select GPUs per platform) | Native AMD GPU acceleration. On Linux, requires system-level ROCm installation. Package support is more limited on Windows. Per-chip compatibility details are planned for a future Hardware Support page. | | **DirectML** | Windows | AMD, Intel, some NVIDIA | Microsoft's DirectML API. Broad compatibility but slower performance than CUDA or ROCm. Development is largely stagnant; consider native ROCm, or ZLUDA if need be, as an alternative for AMD GPUs. | | **ZLUDA** | Windows | AMD (via CUDA translation layer) | Experimental CUDA-to-AMD translation layer. Used by the ComfyUI-Zluda, SD.Next, and AMDGPU Forge packages. Generally faster than DirectML for supported operations. | | **IPEX** | Windows, Linux | Intel Arc (discrete and integrated) | Intel Extension for PyTorch. Requires Intel Arc GPU (A-series, B-series) or modern Intel Core Ultra with integrated Arc graphics. | From 596611529ed7c4eecf0af79c802a2cca77cb3dca Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:39:44 -0400 Subject: [PATCH 08/16] Update GPU information in settings documentation --- docs/settings/settings.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/settings/settings.md b/docs/settings/settings.md index c5d90f30c..1d0162373 100644 --- a/docs/settings/settings.md +++ b/docs/settings/settings.md @@ -299,7 +299,7 @@ Read-only diagnostic information about your hardware: - **CPU** — processor model - **Memory** — total installed and available RAM -- **GPU** — detected graphics cards with driver version and VRAM +- **GPU** — detected graphics cards and VRAM --- From 18a2516d55399ef59d40912cbc3f7fdd6d229c21 Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:52:28 -0400 Subject: [PATCH 09/16] Update docs/advanced/environment-variables.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- docs/advanced/environment-variables.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/advanced/environment-variables.md b/docs/advanced/environment-variables.md index b26c62ac2..0e1ef004b 100644 --- a/docs/advanced/environment-variables.md +++ b/docs/advanced/environment-variables.md @@ -151,7 +151,7 @@ For Windows ROCm-based WebUI package launches, Stability Matrix already applies | `MIOPEN_SEARCH_CUTOFF` | `1` | | | `MIOPEN_FIND_ENFORCE` | `1` | | | `TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL` | `1` | Enables the experimental ROCm AOTriton path. Applied only on RDNA3 / RDNA3.5 / RDNA4 GPUs, and additionally excluded on gfx1152/gfx1153 APU architectures where AOTriton isn't yet supported. | -| `FLASH_ATTENTION_TRITON_AMD_ENABLE` | `TRUE` | Enables the Triton-based Flash Attention implementation for AMD GPUs, providing an efficient attention path when the native ROCm Flash Attention is unavailable and instead provided by a seperate FA python package. | +| `FLASH_ATTENTION_TRITON_AMD_ENABLE` | `TRUE` | Enables the Triton-based Flash Attention implementation for AMD GPUs, providing an efficient attention path when the native ROCm Flash Attention is unavailable and instead provided by a separate FA Python package. | | `COMFYUI_ENABLE_MIOPEN` | `1` | Keeps the MIOpen-backed path enabled in ComfyUI on RDNA3 / RDNA3.5 / RDNA4 GPUs. | | `PYTORCH_ALLOC_CONF` | `max_split_size_mb:512,garbage_collection_threshold:0.8` | | From 1bdd97a599d2c8f1619cbc151009ec693dd5791e Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:53:10 -0400 Subject: [PATCH 10/16] Update docs/advanced/hardware-support.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- docs/advanced/hardware-support.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/advanced/hardware-support.md b/docs/advanced/hardware-support.md index 6012b2242..c69e0ed2f 100644 --- a/docs/advanced/hardware-support.md +++ b/docs/advanced/hardware-support.md @@ -59,7 +59,7 @@ Architectures in the `gfx110x`, `gfx115x`, and `gfx120x` ranges are treated as " - On modern architectures it applies a set of ROCm performance and attention environment variables at launch (MIOpen find-mode tuning, AOTriton experimental flash attention, `COMFYUI_ENABLE_MIOPEN`, and an allocator tuning string). AOTriton is excluded on the `gfx1152` / `gfx1153` APU architectures, which it does not yet support. Legacy architectures instead force a math SDP fallback. The full variable list and exactly which ones are auto-applied are documented in [Environment Variables](environment-variables.md#amd-and-rocm-variables). - ComfyUI offers optional extra commands for supported AMD GPUs, including **Install Triton and SageAttention (ROCm)** (Sage Attention 1.x), **Install Flash Attention (ROCm)** (legacy architectures), an **Install ROCm Development SDK** step, and an **Install bitsandbytes (ROCm)** step for Python 3.12 environments. - **Packages:** ComfyUI, Stable Diffusion WebUI Reforge, InvokeAI, SwarmUI, and Wan2GP are supported by this install path. - > [!NOTE] While not managed by Stability Matrix for ROCm installs, SD.Next has Windows-native ROCm install when the "Rocm" Pytorch index is selected in Advanced Installation Options during initial package install and `--use-rocm` is set in launch options. This install path is internally handled by SD.Next itself and currently only supports RDNA2 dedicated GPUs, RDNA3, RDNA3.5, and RDNA4 GPUs. + > [!NOTE] While not managed by Stability Matrix for ROCm installs, SD.Next has a Windows-native ROCm install when the "ROCm" Pytorch index is selected in Advanced Installation Options during initial package install and `--use-rocm` is set in launch options. This install path is internally handled by SD.Next itself and currently only supports RDNA2 dedicated GPUs, RDNA3, RDNA3.5, and RDNA4 GPUs. **Caveats:** From 593a0448c0b4b1cba7587d4e3a8cb235865b92c7 Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:53:29 -0400 Subject: [PATCH 11/16] Update docs/advanced/hardware-support.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- docs/advanced/hardware-support.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/advanced/hardware-support.md b/docs/advanced/hardware-support.md index c69e0ed2f..76a8cca16 100644 --- a/docs/advanced/hardware-support.md +++ b/docs/advanced/hardware-support.md @@ -64,7 +64,7 @@ Architectures in the `gfx110x`, `gfx115x`, and `gfx120x` ranges are treated as " **Caveats:** - Windows AMD ROCm implementation for Stability Matrix is explicitly in an experimental state. While these package installations are generally in a working state, untested or edge-case issues and incompatibilities may appear. Stability Matrix prints a notice asking you to report issues to Stability Matrix first, since the setup may not be officially supported by the upstream package developers. -- Only the architectures listed above are eligible. If your AMD GPU is not on the list, the recommended default becomes ZLUDA or DirectML instead. For a more detailed list of compatible AMD GPU architectures, please refer to the ROCm/TheRock [GPU Support](`https://github.com/ROCm/TheRock/blob/main/SUPPORTED_GPUS.md#rocm-on-windows`) table. +- Only the architectures listed above are eligible. If your AMD GPU is not on the list, the recommended default becomes ZLUDA or DirectML instead. For a more detailed list of compatible AMD GPU architectures, please refer to the ROCm/TheRock [GPU Support](https://github.com/ROCm/TheRock/blob/main/SUPPORTED_GPUS.md#rocm-on-windows) table. ### ZLUDA From a23bcc9b4c068e141a476b87e7d8648aeb491431 Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:53:46 -0400 Subject: [PATCH 12/16] Update docs/getting-started/overview.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- docs/getting-started/overview.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/getting-started/overview.md b/docs/getting-started/overview.md index bf1393d29..96209b748 100644 --- a/docs/getting-started/overview.md +++ b/docs/getting-started/overview.md @@ -43,7 +43,7 @@ Stability Matrix is cross-platform, but the exact release formats and hardware t | Operating System | Version / Target | Architecture | Notes | |---|---|---|---| | Windows | Windows 10 and Windows 11 | x64 | Official release builds are published for `win-x64`. This is the broadest-supported desktop target for Stability Matrix and most package workflows. | -| Linux | Modern x86-64 desktop distributions | x64 | Official Linux releases are published for `linux-x64`, primarily as an AppImage, with an AUR package also available for Arch-based systems. Depending on the distribution, runtime support packages such as `fuse3`, or `libxcrypt-compat` may be needed if they are not already provided by the system. | +| Linux | Modern x86-64 desktop distributions | x64 | Official Linux releases are published for `linux-x64`, primarily as an AppImage, with an AUR package also available for Arch-based systems. Depending on the distribution, runtime support packages such as `fuse3` or `libxcrypt-compat` may be needed if they are not already provided by the system. | | macOS | Apple Silicon Macs, with macOS 12.3 or later recommended for AI workflows | arm64 | Official macOS releases are published for Apple Silicon (`osx-arm64`) as a `.dmg`. The app's AI workflows rely on the MPS backend on Apple Silicon. | In other words, the practical supported release targets are Windows x64, Linux x64, and Apple Silicon macOS. Some project files include additional runtime identifiers, but the documented source-build support and the release pipeline currently focus on `win-x64`, `linux-x64`, and `osx-arm64`. For work from a local checkout instead of a packaged release, a planned Building from Source and Contributing page will serve as the documentation entry point and link to the repository's contributor guide. From 8c1ae147d2b262e608a0b30e79fbadfec699676c Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:54:38 -0400 Subject: [PATCH 13/16] Update docs/settings/settings.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- docs/settings/settings.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/settings/settings.md b/docs/settings/settings.md index 1d0162373..fb52a6b2b 100644 --- a/docs/settings/settings.md +++ b/docs/settings/settings.md @@ -90,7 +90,7 @@ The Accounts sub-page manages external service credentials. Connecting accounts #### Membership and Lykos Account -A Lykos account is the central identity for Stability Matrix. Signing in unlocks membership benefits and account-linked features. Membership is managed at [lykos.ai/membership](https://lykos.ai/membership?). +A Lykos account is the central identity for Stability Matrix. Signing in unlocks membership benefits and account-linked features. Membership is managed at [lykos.ai/membership](https://lykos.ai/membership). | Tier | Description | |---|---| From 99189b659e584de2a9007d7ec646ee40824634d1 Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:54:48 -0400 Subject: [PATCH 14/16] Update docs/settings/settings.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- docs/settings/settings.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/settings/settings.md b/docs/settings/settings.md index fb52a6b2b..69069b0e6 100644 --- a/docs/settings/settings.md +++ b/docs/settings/settings.md @@ -135,7 +135,7 @@ Enables the **Nano Banana** image generation provider in the **Image Lab**, whic **To connect**: Go to [Google AI Studio](https://aistudio.google.com/api-keys), create an API key, and paste it into the Connect dialog under the **Image Generation APIs** section. The key is validated at generation time when the Image Lab makes its first request. > [!IMPORTANT] -> When API keys are generated on their respective service website, it is recommended to save your API keys somewhere safe (ie: text file, password/note manager, etc.) as these services will typically only provide this key once. +> When API keys are generated on their respective service website, it is recommended to save your API keys somewhere safe (e.g., text file, password/note manager, etc.) as these services will typically only provide this key once. #### How Credentials Are Stored From 68e4d8d02b22837ebb625d439a65c27cf41900ec Mon Sep 17 00:00:00 2001 From: NeuralFault <65365345+NeuralFault@users.noreply.github.com> Date: Sat, 18 Jul 2026 17:55:18 -0400 Subject: [PATCH 15/16] Update docs/advanced/hardware-support.md Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- docs/advanced/hardware-support.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/advanced/hardware-support.md b/docs/advanced/hardware-support.md index 76a8cca16..ad74692cb 100644 --- a/docs/advanced/hardware-support.md +++ b/docs/advanced/hardware-support.md @@ -55,7 +55,7 @@ Architectures in the `gfx110x`, `gfx115x`, and `gfx120x` ranges are treated as " **What Stability Matrix does automatically:** - When a supported AMD GPU is present on Windows, ROCm becomes the recommended backend for ROCm-capable packages. -- Torch is installed from AMD's ROCm multi-arch index (`repo.amd.com/rocm/whl-multi-arch/`) as device-specific wheels (`torch[device-gfxNNNN]`). Vega parts (`gfx900` / `gfx906`) pull from the 'TheRock' nightly multi-arch feed instead, since these architecture builds currently are only avaliable there instead of the stable production distribution stream. +- Torch is installed from AMD's ROCm multi-arch index (`repo.amd.com/rocm/whl-multi-arch/`) as device-specific wheels (`torch[device-gfxNNNN]`). Vega parts (`gfx900` / `gfx906`) pull from the 'TheRock' nightly multi-arch feed instead, since these architecture builds currently are only available there instead of the stable production distribution stream. - On modern architectures it applies a set of ROCm performance and attention environment variables at launch (MIOpen find-mode tuning, AOTriton experimental flash attention, `COMFYUI_ENABLE_MIOPEN`, and an allocator tuning string). AOTriton is excluded on the `gfx1152` / `gfx1153` APU architectures, which it does not yet support. Legacy architectures instead force a math SDP fallback. The full variable list and exactly which ones are auto-applied are documented in [Environment Variables](environment-variables.md#amd-and-rocm-variables). - ComfyUI offers optional extra commands for supported AMD GPUs, including **Install Triton and SageAttention (ROCm)** (Sage Attention 1.x), **Install Flash Attention (ROCm)** (legacy architectures), an **Install ROCm Development SDK** step, and an **Install bitsandbytes (ROCm)** step for Python 3.12 environments. - **Packages:** ComfyUI, Stable Diffusion WebUI Reforge, InvokeAI, SwarmUI, and Wan2GP are supported by this install path. From f7fb5ac7a23fb45036aa177ecf6cf7ae62b9d911 Mon Sep 17 00:00:00 2001 From: NeuralFault Date: Sat, 18 Jul 2026 21:59:36 +0000 Subject: [PATCH 16/16] Update docs/README.md --- docs/README.md | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) diff --git a/docs/README.md b/docs/README.md index 695271779..29cbcbac2 100644 --- a/docs/README.md +++ b/docs/README.md @@ -52,11 +52,7 @@ Current Status: In-progress - Structure is in-place and planned docs are current - Community Workflows *(planned)* — Browsing community workflows via OpenArt ### Settings -- Overview *(planned)* — Navigating the settings hub -- General *(planned)* — Theme, language, data directory, and shared folder settings -- Accounts *(planned)* — Lykos account, OAuth login, and API tokens -- Inference Settings *(planned)* — Inference UI behavior and defaults -- Updates *(planned)* — Auto-update channel and frequency settings +- [Settings](settings/settings.md) — All application settings covering General, Accounts, Checkpoint Manager, Package Environment, Model Browser, Workflow Browser, Console, Appearance, System, and Debug Options ### Advanced - [Overview](advanced/overview.md) — Advanced configuration and technical reference