Images
Illustrations generated from the scene, right in the chat.
Turn on image generation per conversation in the settings drawer, or set it as a preset default. Once enabled, replies can produce a scene illustration automatically.
After a reply lands, a prompt-writer model reads it and writes an image prompt; the image model then renders it. The reply is never delayed: the image appears in place when ready.
A placeholder marks the spot while generating; if generation fails, the chat simply continues without the image.
Pick any catalog image model, or let the engine choose a free one. The prompt-writer uses your utility model. Custom-provider image models are dispatched directly to your own endpoint.
Upload reference images in the settings drawer to keep faces, outfits, or styles consistent across generations; the primary character's avatar can be included automatically.
Up to six references ride along per request, capped by what the image model accepts.
Style templates prefill the prompt instruction with a look: photorealistic, anime, painterly, and more. Edit the instruction after applying a template to fine-tune your default look.
With preview enabled, the written image prompt pauses for your review: edit it, approve it, or skip the image entirely.
Any generated image can be regenerated in place from its overlay menu; the new render replaces the old one in the same spot.
Images live in messages as {{inlay::<id>}} tokens rendering from local storage. Exported conversations carry the bytes, so images survive the trip.