-
Notifications
You must be signed in to change notification settings - Fork 0
Troubleshooting
Open Agents — the reason is reported there, and the failure panel on Generate names it too.
The common causes:
- Out of memory. A 1024×1024 SDXL render wants 10–11 GB. Use a smaller shape, fewer steps, or SD-Turbo at 512px. Compute tells you whether a job fits before you start it.
- The engine is not installed. System → install it.
The model is loading into the graphics card, which happens once per session per model. Later renders are much faster. Switching models pays it again.
History logging is off by default (System → Settings). While it is off, renders are not recorded, so there is nothing to rate. Turn it on; renders made after that can be rated.
ImageForge does not include the trainer. If kohya sd-scripts is not installed
in ~/sd-scripts, a run writes its configuration and stops, and says so. See
Training a LoRA.
If it reports it could not reach HuggingFace, check your connection. Note that gated models need a token — see System and licences.
- The configuration only fills in after the engine is installed — before that the API screen says so.
- Paths outside ImageForge's own folders are refused by design. Set
IMAGEFORGE_IMAGE_DIRSto allow a folder of your own. See Agents and the API.
ImageForge is not code-signed. More info → Run anyway, or verify the SHA-256 against the release first. See Installing.
You are probably on the processor build. System reports the graphics card and which engine build is installed. Without a supported NVIDIA card, minutes per render is expected.
Everything runs on your own computer. No account, no subscription.
Getting started
Using it
Beyond the app
Reference