Feature request: Use Library images directly as image-generation/editing references

ChatGPT Feature requests
1
  • Goal: Let users pick an existing image from Library and attach it as a real visual input to the image model (not just filename/metadata), avoiding re-uploads.
  • Flow: Library → Select image → (A) Use as visual reference or (B) Use as source image to edit.
  • Scope/platform: Works for both image generation and image editing; ideally web + mobile (web first is fine).
  • Where it appears: Anywhere users can currently add image references—generation prompt/attachment picker and image editor.
  • Reference modes (separate behaviors):
    • Use as visual reference: preserve the identity/appearance of the referenced asset while creating a new image.
    • Use as source image to edit: treat the selected image as the actual image being modified (edit/inpaint/outpaint).
  • Multiple references: Support 3–5 per generation/edit; allow roles per reference and reordering/priority.
  • Versioning: Generations/edits should bind to the exact version used at the time; later Library updates shouldn’t alter old chats.
  • Projects & sharing: Project-shared Library assets available to project members according to permissions; personal Library stays private unless explicitly shared.
  • UI transparency/consent: Clearly show which Library images are being used as references, with remove/replace options before generating.
  • Reliability control: Simple toggle such as “Preserve closely” vs “Use as inspiration.” This is particularly important for approved products, logos, characters, and other assets that should not be visually reinterpreted.
  • Storage/efficiency: Reference the stored asset via a pointer rather than duplicating/re-uploading the file; this could reduce repeated uploads, failed generations, and wasted compute.
  • If a referenced image is deleted: Warn and allow replacement before generating; past generations remain unaffected.
  • Key requirement: Selecting a Library reference must make the actual visual asset available to the image model, rather than merely providing its filename, metadata, or textual description.
2

Thanks for the detailed suggestion! You can reuse saved files through Add from library: https://help.openai.com/en/articles/20001052. We'll share the ideas for clearer image-reference roles, preservation and version controls with the product team. No timeline to share.