Skip to content

Latest commit

 

History

History
58 lines (41 loc) · 1.4 KB

File metadata and controls

58 lines (41 loc) · 1.4 KB
name reverse-prompt
description Use when turning an image into a reusable prompt for image generation, visual design, or cinematic direction.
phase 1
status available

Reverse Prompt

Use this skill to convert an image into a structured prompt that captures its visual character.

Evals

Should trigger

  • "turn this image into a generation prompt"
  • "give me a prompt to recreate this style"
  • "cinematic prompt from this frame"
  • "reverse-engineer a prompt for this art"

Should not trigger

  • "score the image quality" → use image-quality-audit
  • "what objects are in it?" → use visual-reasoning
  • "review this UI screenshot" → use screenshot-ui-review

Core workflow

  1. Run visual-reasoning on the image
  2. Extract subject, environment, lighting, palette, composition
  3. Translate signals into prompt language
  4. Return single-string prompt and token list

Inputs

  • image_path — local path to image file

Outputs

  • prompt — complete prompt string
  • tokens — individual prompt components

Include in prompt

  • Subject / objects
  • Environment and setting
  • Palette description
  • Lighting (inferred from brightness + contrast)
  • Composition style
  • Depth / focus treatment
  • Mood and tone

Avoid

  • Claiming exact camera or lens data unless EXIF confirms it
  • Overfitting prompt language to a single model provider
  • Hallucinated details not visible in the image