ChatGPT, Claude & Gemini
Use the default format with chat assistants — for text and for their built-in image generation. The style block leads; your request follows.
Placement
Put the style block first as a leading constraint, then your request:
<compiled default-format style block>
---
REQUEST: a landing page hero headline for a developer toolYou have three places to put the block, from most to least durable:
System prompt
Set the style block as the system prompt (Claude, or any API/system-prompt field). It then constrains every message in the conversation.
Tips
- The style is a hard constraint: it overrides conflicting instructions in your request. If you want to break the style deliberately, say so explicitly.
- For image modes, the same
defaultblock works — describe the subject in your request and let the style set the look. - To apply a style in one step from an agent, use
apply_style_to_prompt, which returns a ready-to-paste prompt. See Tools reference.
What to expect
Within one tool, the style holds consistently across generations because each one receives the same constraints. Across ChatGPT, Claude, and Gemini, the same spec produces related but not identical results — each model reads it differently.
Next steps
- Midjourney
- FLUX & diffusion models
- MCP tools reference —
apply_style_to_prompt.
Last updated on