Free resource
GPT Image skill
Cinematic, uncluttered, non-generic pictures on any image model, in Claude Code, Codex or the API.
Locked
GPT Image skill · free
Pictures that don't look generic.
Left alone, GPT Image dresses every set with props you never asked for, writes fake signs and brand names on everything, and poses people like stock models. This skill turns a plain request into a frame pulled from a film: a real camera look, only the things you ask for, true-to-life textures, and no stray text. Words only, no reference images.
Built and tested on GPT Image 2, 2.5 and 2.6 across 102 pictures. The same prompts hold up on Nano Banana, Seedream and Ideogram.
- Claude Code
- Codex
- The API
Works on other image models too.
The same skill prompt, sent to five other models. Same person, same place, no stray text.
- OpenAI
- ByteDance
- Ideogram
Nano Banana 2.1Google
Seedream 5 ProByteDance
Seedream 5 FlashByteDance
Ideogram 4.5Ideogram
The library.
Every picture from testing, made from words alone. Pick a shelf.