GPT Image-2 anime & manga prompts

Anime and manga prompts are the trickiest aesthetic to write reliably. The same words ("anime style") in two different prompts produce wildly different outputs — sometimes 90s cel-shaded, sometimes 2020s shounen, sometimes Ghibli pastel, sometimes Korean webtoon. The prompts here pin the era and studio influence explicitly so you stop rolling the dice.

A working anime prompt looks more like a model sheet than a paragraph. It names the era (late 80s cel-anime, 2010s Kyoto Animation, 2020s mobile-game keyart), the character anatomy (head-body ratio, eye style, hair line-weight), the shot type (full-body sheet, three-quarter keyframe, manga panel splash), and the lighting/post (flat cel + hard shadow, painterly soft shading, halftone for manga). It then writes the scene with cinematic language — "low-angle hero shot, hair caught mid-action by a side wind" — not generic noun lists.

The cases here are organised by use case rather than series. Character sheets show the same figure from three angles. Transformation sequences split the frame into stages (resting → mid-transform → fully transformed). Magical-girl burst pages are written in manga splash-page idiom. Keyframes from imagined episode cuts have the right pacing language. Chibi sets stay tonally coherent across multiple expressions.

A practical note on character consistency: GPT Image-2 still cannot perfectly hold a single character across multiple separate generations. You will get the same archetype, but small details (hair tie, scar position) drift between renders. The workaround is to copy the prompt, generate the first image, then in the next message paste the image back and ask for "the same character" — visual anchoring outperforms re-prompting.

If you are using these for fan art or original IP, every prompt is original and license-free. Studio names appear in the cases only as stylistic descriptors, not as direct lifts of existing copyrighted characters.