Old photos generated with gpt-image-2, hard to tell real from fake!

Put the cover under the lamp. Each of the five things that made it work is a layer you can switch on: the words, the grid it sits on, the colours it uses, the subject, the light.

Cover of the original post: Old photos generated with gpt-image-2, h — Cover · 1 of 1Original on Xiaohongshu ↗

Layers

Hook · Text layer / Text layerNo text in the image
Palette · Color layer / Palette layerFaded warm gray-pink
Mood · Mood layer / Mood layerMisty old-photo feel

Spec sheet

1Hook

. The image itself has no text; it relies on a group photo that looks like an old photo to make people pause first, then look back at the title.

Readable text: None
Note: The text in the title does not appear on the image

3 likes 1 saves 1 shares

Posted by momo on Xiaohongshu, 24 Apr 2026. Shown with credit and a link back.

Keep all five layers and swap what’s under them: your shop, your budget, your city. You get four covers and the post text.

Make your version4 cards Stick it together

The prompt, as used

Verbatim. Copy it, or let Queenie swap in your own details.

Referencing the old photo restoration approach, vintage texture, visual storytelling, and publishing tone of the case study "Old photos generated with gpt-image-2, hard to tell real from fake!", generate a new set of retro old photo image-text content. New topic: {fill in the photo topic you want to restore, rebuild, or recreate} Reference platform: Xiaohongshu Content direction: retro old photo image-text Creation method: retain the nostalgic emotion, film/photo paper texture, character relationships, scene era feel, and restoration constraints of the original case; if it is photo restoration, it must stay faithful to the original image's identity, facial features, composition, and historical context, avoiding excessive beautification, repainting, false historical information, and modern digital plastic feel. Please output: 1. Cover title and subtitle; 2. 6-9 page image-text script, each page including visual description, main copy, era/texture keywords, and save prompt; 3. Visual prompts that can be directly used for image generation or photo restoration, explaining requirements for characters, scene, lighting, film grain, damage repair, and color restoration; 4. Publishing suggestions: title length, tags, image ratio, and platform tone. Constraints: do not replicate the original author, original people, or brand logos; do not fabricate historical facts; when involving real people's photos, prioritize preserving identity consistency, and do not do age reduction, beautification, face slimming, or facial feature rewriting.