Images: Generation, Editing, Gallery, and Annotations

TatsuCode v0.9.107 adds a complete image workflow — generate new images, edit existing ones with masks and reference images, browse everything in a unified gallery, and mark up images directly in chat to point things out for the agent.

This page covers all four features in one place. They share concepts (model selection, references, lifecycle), so it's easier to read them together.

Image Generation

A built-in workflow for creating images — concept art, mockups, sprite sheets, icons, marketing visuals, asset references, and anything else you'd reach for an image model to do.

Configure the Provider and Defaults

/image-generation

Pick a provider, a model, and your default size / aspect ratio / quality / background. Locked settings here override per-call values, so once you've configured your defaults, you don't need to specify them every time.

Supported Providers and Models

Each model has its own supported sizes, aspect ratios, qualities, and backgrounds — /image-generation shows what's available for the model you've picked.

Generate

Just ask:

> Generate a pixel-art potion bottle sprite sheet with 4 idle frames
> Make me a square cover image for a synthwave podcast
> Generate a clean architecture diagram for a 3-tier web app

Edit

Edit an existing image rather than starting from scratch. References can come from any of:

• Drag-and-drop an image into the chat
• Paste an image from clipboard
• @img-NN to reference an image already in the conversation
• A file path or URL

> Add a baseball cap and sneakers to the dog in the attached image
> Take @img-03 and turn the sky into a dramatic sunset
> Use these three references and combine them into one stylized game icon

Masked Inpaint

Restrict an edit to a specific region by supplying a mask image alongside the source. White pixels = edit; black pixels = keep. Useful for swapping skies, replacing logos, or cleaning up a single corner without disturbing the rest.

Quota Detection

If your image-gen quota or balance is exhausted, TatsuCode surfaces a clear quota error rather than failing silently. You'll know exactly which provider tripped and can switch via /image-generation.

Image Gallery

/image-gallery

Browse every image in your current conversation — and across the project — in one searchable, scrollable gallery. Stable with 10,000+ images, with deduping, lifecycle cleanup, and quota-aware loading. Click any image to open it full-size; click again to use it as a reference in your next prompt.

The gallery is the same store the agent uses, so anything that ended up in chat (drag-dropped, pasted, generated, screenshot, browser-captured) is available.

Image Tools

A built-in skill that gives the agent direct image-editing operations: resize, crop, trim, blur, and more. Useful when:

• Prepping an asset to a specific size before handing it off
• Cropping a screenshot down to the relevant region for a clearer review
• Blurring sensitive content before sharing
• Trimming whitespace around a logo or icon

The agent uses these tools automatically when the task calls for it (e.g., "crop this screenshot to just the error dialog and zoom in"). You can also ask explicitly:

> Resize this image to 1024x1024 and trim the transparent edges
> Crop @img-02 to the top-right quadrant
> Blur the email addresses in this screenshot

Image Annotations

Mark up images directly in chat to point things out for the agent — circle a misaligned button, highlight a broken layout, drop a note next to a confusing UI element. Annotations are passed through to the model as visual context, so the agent sees exactly what you're flagging.

Use cases:

• UI feedback — circle the part that's wrong; the agent fixes it
• Bug reports with visuals — highlight an artifact in a screenshot to debug
• Design review — point at specific elements for targeted feedback

Annotations live with the image in the gallery, so they survive session save / restore.

Tips

• For best edit fidelity, use @img-NN to reference an image already in the conversation — it preserves the original at full resolution rather than re-uploading.
• For complex edits, use masked inpaint to keep the rest of the image stable.
• Save reusable defaults in /image-generation so you're not repeating size/aspect ratio every prompt.
• The gallery is your friend on long sessions — it's faster than scrolling chat to find an earlier generation.

Next Steps

• Agent Skills — the Image Generator and Image Tools skills
• Tools Reference — what tools the agent invokes for image work
• Commands — /image-generation, /image-gallery