Generating images

With the image generation add-on, the agent creates images locally through ComfyUI (Stable Diffusion) — logos, social graphics, ad creative, concept art. No cloud image API, no per-image cost.

Setup is automatic

The agent finds an existing ComfyUI install on its own — including Stability Matrix installs — and starts it with the dashboard and on demand. If you don't have ComfyUI at all, Settings → Image Generation has a one-click installer. If the status dot in the dashboard is red, clicking it attempts a start and reports what happened.

Just describe what you want

"Generate a realistic photo of…" or "draw a cartoon of…" in the chat is enough. The style words matter: the agent picks the image model (checkpoint) to match — "realistic"/"photo" selects a photorealism checkpoint if you have one installed, "anime"/"illustration" selects a stylized one. You can also pin one yourself with /checkpoint <filename> or by naming it in your request.

For a quick test that skips the language model entirely, use /image your prompt here.

Iterating on a result

• Variations — "same image but make it night time" reuses the previous image (img2img) instead of starting over.
• Upscaling — "make it print quality" produces a 2x version with added detail.
• Style presets — if you use a recurring style (a brand look, a comic style), ask the agent to save it as a preset and reuse it by name later.
• Custom workflows — advanced: the agent can author, save, and run full ComfyUI node graphs for pipelines the standard modes can't do.

Honest limits

Some asks don't work reliably with Stable Diffusion alone, and the agent is instructed to say so up front instead of burning minutes: putting a specific person's face into an image, rendering exact text, and reproducing a precise existing logo. It will offer the closest achievable alternative.

Where images go

Results show inline in the chat (click to open full size) and are saved locally under the agent's outputs/images folder — the 🗂 files drawer lists them all.