Image generation

Replies can carry pictures. When the model writes an [image=...] tag, ScyllaChat generates that image in the background and drops it into the message.

Setup

In the Preset tab, under Image Generation:

1. Enter your Image API URL and Image API Key. NovelAI is the reference target; any compatible endpoint works.
2. Set the model name and default width and height (256 to 2048, in steps of 64).
3. The chat history limit controls how many recent messages the generator may look at for context.

How it works

1. The model includes a tag like [image=a lighthouse in a storm, oil painting] in its reply.
2. ScyllaChat detects the tag, sends the prompt to your image API, and shows a placeholder while it renders.
3. The finished image replaces the placeholder and is cached with your chat.

Images are cached by a hash of the prompt and dimensions, so re-rendering a message or re-opening a chat never pays for the same image twice. Clear Image Cache in the settings wipes the folder when you want the disk back.

Getting the model to draw

The model only emits tags it was asked to emit. Two reliable ways in:

• Add an instruction to your preset or Author's Note: "When a scene changes dramatically, include one [image=...] tag describing it visually."
• Ask in character: "show me what the lighthouse looks like."

Prompt tags work best written like image prompts (subject, style, mood), not prose. A short style suffix in your instruction ("always end image prompts with: detailed, cinematic lighting") keeps a chat's art consistent.