remove-photo-filter

Neutralize an applied photographic filter or color grade and recover a natural-looking color, contrast, white balance, and texture while preserving scene content and acknowledging that hidden original

SKILL.md

GitHub

Remove Photo Filter

Catalog

  • Display name: Remove Photo Filter
  • Categories: Utility
  • Status: Ready
  • Route label: GPT Image 2 consistency-sensitive edit
  • Tone: sky
  • Short description: Remove an artificial grade and restore a natural look.
  • Similar skills: apply-photo-filter, recolor-object

What this skill does

  • Identifies visible filter symptoms and asks the image model to neutralize them coherently.
  • Preserves people, geometry, objects, text, crop, and genuine scene lighting.

How to use

Provide filtered image, known original references if available, and desired natural-light target.

Remove a heavy filter

Remove the strong orange-teal grade, crushed blacks, vignette, and added grain; restore plausible daylight color.

Workflow

  1. Describe filter symptoms: cast, curve, saturation, clipping, grain, bloom, vignette, and beauty effects.
  2. Record any known neutral references; state that unrecoverable hidden colors are inferred.
  3. Run gpt-image v1.0 with quality: "medium", imageSize: "2K", the supplied references, the requested or source-preserving ratio, batchCount: 1, and one result.
  4. Forbid content, face, texture, crop, pose, background, and lighting-direction changes.
  5. Inspect neutrals, skin, highlights, shadows, banding, residual cast, texture, and content stability.

User-facing output

  • Media type: One naturalized photograph
  • Default quantity: 1
  • Content per image: Same source with filter symptoms reduced
  • Default layout: Source composition
  • Model policy: GPT Image 2 Medium/2K for consistency-sensitive final image editing
  • Downstream use: Photo correction and asset normalization