gptdiff · AI-liftoff comic

A comic page built from a multi-file project — a JSON art-direction config (style + character bible) and one JSON file per panel (scene, shot, cast, caption, dialogue, FX). Describe a change and gptdiff rewrites all of them (adding or removing panels as it sees fit). The harness then organizes every panel into one prompt and draws the whole portrait page in a single image render (the image model is set in config.json — ideogram:4@0 by default) via NanoGPT. Export the finished page as a PNG.

Project files (gptdiff edits, adds & removes these — config + one file per panel)

… Goal (what to change)

Diff & apply → re-render ↶ undo

◐ starting… · ⏱ 0.0s

↓ 0 in · ↑ 0 out · $0.00000

Generated diff

Comic page · one image render auto-render 🎨 render page 🎲 new take ⬇ export PNG

Hit 🎨 render page to draw the whole comic in one image render — or run a goal and let it auto-render.

◐ drawing…

Page prompt (every panel organized into one render)

…

The prompt = the shared style + the character bible + each panel's scene, caption, dialogue and FX, assembled in reading order — so it's drawn as a single coherent page (panels, lettering and all). Results are cached per prompt, so re-running an unchanged page is free. Click the page to view it fullscreen (click again for actual size).