Image generation¶

Pluma talks to two image backends, both via their HTTP APIs:

A1111 (Stable Diffusion web UI) via /sdapi/v1/txt2img
ComfyUI via its /prompt API with a workflow JSON

The in-chat wand button derives the SD prompt from the last few conversation turns.

Set up a backend¶

A1111¶

Start A1111 with --api so it exposes the HTTP API:

./webui.sh --api

It serves on http://127.0.0.1:7860 by default.

In Pluma: Settings → Images → Auto-detect. Pluma probes 7860 and creates a profile. Or click + A1111 to add one manually.

ComfyUI¶

Start ComfyUI normally; it serves on http://127.0.0.1:8188.

In Pluma: Settings → Images → Auto-detect (probes 8188). Or + ComfyUI for manual entry.

ComfyUI needs a workflow JSON. The format is the same one ComfyUI's UI exports. Pluma's %placeholder% substitution convention matches what SillyTavern uses, so existing ST workflows drop in unchanged.

Workflow library¶

<datadir>/workflows/ holds workflow JSONs you can pick from the dropdown when configuring a ComfyUI profile. Pluma ships flat-anime-portrait on first run.

To add your own:

Export a workflow from ComfyUI's UI (Save As → JSON).
Drop the file into <datadir>/workflows/.
Refresh the workflow dropdown in the image profile editor.

Pasted multi-line prompts auto-normalise so workflows exported from the ComfyUI UI drop in directly without manual escaping.

Placeholders¶

Workflow JSONs reference these placeholders. Pluma substitutes them at generation time.

Placeholder	Replaced with
`%prompt%`	The SD prompt (composer wand-button derivation OR explicit input).
`%negative_prompt%`	The negative prompt (server-side floor adds anti-realism / anti-crop negatives even when the workflow's negative encode is bare).
`%char_avatar%`	Active character's avatar (for img2img workflows). Only present in chat-derived generations.
`%seed%`, `%steps%`, `%cfg%`, `%sampler%`	Override-able knobs from the profile.
`%width%`, `%height%`	Output dimensions.

In-chat wand button¶

The composer's wand icon shows up when an SD profile is active. Click it to:

Derive an SD prompt from the last few chat turns.
Generate against the active workflow.
Drop the result into the conversation as an attachment.

Generation status streams to the composer; click again to cancel.

Test from settings¶

The image-profile editor has a Try it input. Type a prompt, click Generate test image, and Pluma runs the workflow once. Useful for verifying connection + workflow without going through the chat path.

Preview generates without a character avatar; workflows that need %char_avatar% only run from inside a chat.