Skip to main content
In v2 all prompt boosters use a single unified endpoint — POST /api/v2/prompts/enhancements. You select the booster with the type field (v2 dot notation) and the target model via model_slug.

​
Endpoints

​
Request fields (/api/v2/prompts/enhancements)

​
type values

Only prompt-driven inference types have a booster. Types that take no meaningful prompt — upscaling, background removal, OCR, transcription, embeddings — are not supported and are rejected with 422.

​
Workflow

1

Enhance your prompt

Call POST /api/v2/prompts/enhancements with prompt, type, and model_slug (plus image for images.edits / videos.animations).
2

Use the enhanced prompt

Plug the returned prompt into the matching v2 generation endpoint — /api/v2/images/generations, /api/v2/images/edits, /api/v2/videos/generations, /api/v2/audio/speech, etc.
See the Prompt Booster page for the full request/response schema.

​
Inline boosting (enhance_prompt)

Instead of calling the booster first and passing the result on yourself, you can set enhance_prompt: true directly on a generation request. The boost then runs as an async pre-step inside the job, and the boosted prompt is what gets generated. Supported on POST /api/v2/images/generations, POST /api/v2/images/edits, POST /api/v2/videos/generations and POST /api/v2/videos/animations. The job status response reports what happened:
  • prompt_boosted — whether the boost actually ran.
  • prompt_boost — the boosted and original prompts. Non-null only once prompt_boosted is true; it stays null while the boost is still queued, and when no boost was requested.
Inline boosting requires a prompt-booster guide configured for the target model. The request is rejected with 422 if no guide exists for that model, or if your account balance cannot cover the boost fee on top of the generation.