How you prompt depends on the engine you're using.
Forge & Photoreal — natural language
Forge and Photoreal understand plain English. Describe the scene like you'd tell a person:
A woman in a leather jacket leaning against a neon-lit wall at night, cinematic, light rainFull sentences, moods, and details all work. Photoreal is tuned for photorealism, so realistic wording (natural light, shot on camera, realistic skin) plays to its strengths.
Prism & Anime — Booru tags
Prism and Anime are driven by Booru tags: short, comma-separated keywords, not full sentences. "Boorus" are anime/art image boards (like Danbooru and Gelbooru) whose community-tagged images trained these models, so their tags are the vocabulary they understand best.
Simple:
1girl, red dress, beach, sunset, smiling, looking at viewerMore complex, with weighting:
1girl, (red dress:1.2), windswept hair, beach, sunset, (freckles:0.8), looking at viewer, depth of field- Default weight is 1.0;
(tag:1.2)emphasizes a tag,(tag:0.8)plays it down. - Don't push much past 1.7, or you'll get artifacts.
- Look up the exact tag on a booru's tag list (Danbooru tags, Gelbooru tags): type a word to see the canonical tag and how common it is.
You usually don't start from scratch
For every Peep image, AI Peeps already builds a full, correctly-formatted prompt behind the scenes — from the Peep and the chat — so most of the time you just tweak. Want a blank slate instead? Switch on Full Control Mode in custom generation and only your own prompt is used. See Images.
