feat: add dirtberry

[Simplereally]

Summary by CodeRabbit

Release Notes

• New Features
  • Introduced Dirtberry image generation model with a fixed 9:16 portrait aspect ratio (832×1144 pixels).
  • Pricing set at 0.001 per image. Currently in alpha status.
  • Automatic image optimization for Dirtberry-generated outputs to ensure optimal dimensions.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
bloomstudio	Ignored	Preview	Mar 11, 2026 11:43am

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: dd220d08-4df2-4e37-bafc-2dc76721feae

📥 Commits

Reviewing files that changed from the base of the PR and between fbd7052 and 12057dd.

⛔ Files ignored due to path filters (1)

• convex/_generated/api.d.ts is excluded by !**/_generated/**

📒 Files selected for processing (9)

• convex/batchProcessor.ts
• convex/lib/dirtberryCrop.ts
• convex/lib/index.ts
• convex/singleGenerationProcessor.ts
• hooks/queries/use-generate-image.ts
• hooks/use-generation-settings.ts
• lib/config/models.ts
• lib/schemas/pollinations-pricing.schema.ts
• lib/schemas/pollinations.schema.ts

---

📝 Walkthrough

Walkthrough

This PR introduces Dirtberry image generation model support with specialized image cropping. It adds a new jimp-based cropping utility, integrates it into batch and single generation processors, registers the Dirtberry model in the configuration and pricing schemas, and updates generation hooks to handle dimension overrides from cropped outputs.

Changes

Cohort / File(s)	Summary
Dirtberry Cropping Infrastructure convex/lib/dirtberryCrop.ts, convex/lib/index.ts	New utility module providing Dirtberry-specific image cropping via jimp, computing crop regions based on trim fraction, and returning crop metadata. Exported constants define source/output dimensions. Library index extends public API with cropping functions and types.
Generation Processors convex/batchProcessor.ts, convex/singleGenerationProcessor.ts	Both processors now detect Dirtberry models, fetch source dimensions to override generation request parameters, apply cropping post-generation, and use cropped buffers for upload. Updated generationParams stored to reflect final dimensions when cropping occurs, with fallback to original dimensions if cropping fails.
Model Configuration & Pricing lib/config/models.ts, lib/schemas/pollinations.schema.ts, lib/schemas/pollinations-pricing.schema.ts	New Dirtberry model entry added to MODEL_REGISTRY with fixed 9:16 aspect ratio (832×1144), constraints, and icon/logo definitions. Schema enums extended to recognize "dirtberry" as a known image model. Pricing schema includes Dirtberry entry with 0.001 per-image cost and alpha flag.
Generation Hooks hooks/queries/use-generate-image.ts, hooks/use-generation-settings.ts	GeneratedImage construction now merges original params with dimension overrides from cropped output (width/height). Single-tier fixed-size model detection now uses explicit preset dimensions instead of computed standard dimensions; multi-tier behavior unchanged.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• feat: concurrent single generation queue with cancellation support #30: Also modifies singleGenerationProcessor.ts with cleanup/cancellation logic, creating a direct code-level dependency on the same processor control flow.

Poem

🐰 A new model hops into view,
Dirtberry arrives with dimensions true,
Cropped just right by jimp's keen eye,
Nine by sixteen—a portrait sky!
Generation flows now know the way. ✨

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch feat/dirtberry-image-model

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: db11a87d64

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Sync persisted dimensions when Dirtberry crop falls back

When cropDirtberryImageBuffer throws, this branch logs and keeps going with the original buffer, but outputWidth/outputHeight stay at the pre-crop request values (params.width/height, typically 832x1144 for Dirtberry). Because the upstream request is forced to source dimensions (832x1216), a crop failure stores metadata that no longer matches the uploaded asset, which can skew aspect-ratio-dependent rendering and downloaded dimension expectations; the same pattern is present in convex/batchProcessor.ts.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve output MIME when Dirtberry is re-encoded

This fallback forces outputMime to image/jpeg for any input MIME Jimp cannot encode, but the processors still upload with the original response content-type header and derive the file extension from it. If Dirtberry (or a future variant) returns something like image/webp, the code will upload JPEG bytes labeled as WebP, leading to incorrect metadata/content-type and potential decode issues for clients.

Useful? React with 👍 / 👎.