Claude · vision · prompt workflow

Turn an image into a prompt with Claude

Create a structured positive prompt, negative prompt, target-model settings, and warnings from visible details. The result is an interpretation—not the source image’s original prompt or a reproduction guarantee.

Open Image to Prompt Download SKILL.md Connect MCP

REST clients call POST /api/prompt-director/from-image; MCP clients call image_to_prompt.

Use it responsibly

Use describe_image instead when the user only needs a caption.
Do not claim identity preservation or the original hidden prompt.
Review multi-subject and composition warnings before generation.
Ask before enabling the optional Advanced Optimizer.

{"image_url":"https://example.com/reference.png",
 "medium":"realistic","direction":"soft studio light"}

The standard vision operation uses 2 flexible credits. Advanced adds 1 credit only on success. Uploads are capped at 8 MB.

Complete API guide · All skills