AI creates the assets. Humans direct the story. The creator remains the author.
A manga studio that runs entirely on your machine — reusable characters, polygon panels, tones, dialogue and a natural-language Manga Agent. No account. No subscription. Your API keys, your projects, your pages.
The Kumanga studio: project library, manga assets, the page canvas and the Manga Agent — one editor.
Most AI comics tools generate a finished page and leave you no way to change it. Kumanga generates reusable assets — characters, poses, expressions, backgrounds — and gives you a real non-destructive editor to compose them into pages you control.
Move a character without regenerating the panel. Reframe a full-body asset into a close-up with zero new generations. Rewrite one line of dialogue without touching the art. That's what a harness is: the AI is powerful, and you hold the reins.
Everything below is already built and running in the v0.1 open-source preview.
Create a character once, use them across every panel. Structured pose & expression libraries, reference-aware generation, and semantic slot switching keep them consistent page after page.
Backgrounds, props, uploads, screentones, speed lines, focus lines and impact bursts — all organized, all drag-to-panel, all yours to reuse in any project.
An infinite canvas with polygon panels you can reshape vertex by vertex, classic layouts (including yonkoma), viewport framing from full-body to close-up, and full undo/redo.
Tell it what you want in plain language — "make them blush in panel 2" — and it plans, reuses what exists, generates only what's missing, and commits a verified, fully editable result.
Projects and images live on your machine, not our cloud. Connect any compatible AI provider with your own key — Gemini, OpenAI-compatible, or a custom endpoint. No account, ever.
High-quality PNG export straight from the live scene graph at 1× or 2×, with provenance metadata on every generated asset. Your pages leave the studio ready to publish.
The Manga Agent treats the AI as a creative director — but code owns the state, the identity of every asset, and the final commit. Nothing lands on your page unverified.
Describe the scene, the shot, the mood, the dialogue — in your own words. Your exact wording is locked as evidence before any AI sees it.
A single creative-director pass maps your intent to cameras, interactions and bubbles — validated against a strict contract, never guessed IDs.
Existing characters and assets are reused or transformed first; new images are generated only for what's genuinely missing. Your tokens are respected.
A deterministic verifier checks every actor is placed and dialogue is byte-exact before committing. The run lands whole — or rolls back cleanly.
Two systems that make Kumanga different — recorded straight from the studio.
Three demo pages composed in the studio — same characters, page after page, scene after scene. Click any page to read it full size.
Same two characters across every panel — poses, expressions and interactions generated as reusable assets, composed by hand.
This isn't a concept video. The open-source preview is running in production today — backing accelerates what comes next.
The v0.1 foundation is done. Crowdfunding pays for the features creators ask for most.
What people ask before trying an AI manga studio — straight answers, no fine print.
Kumanga is an open-source, local-first AI manga harness — a complete manga studio that runs in your browser. AI generates reusable assets such as characters, poses and backgrounds, while you direct the story in a non-destructive editor. The creator remains the author: your API keys, your projects, your pages.
Yes. Kumanga v0.1 is free and open source under the MIT license, and the preview is running today. A rewards-based crowdfunding campaign runs on Pozible until 29 September 2026 (all-or-nothing, AU$5,000), GoFundMe takes direct donations, and a Kickstarter campaign is in review — the core studio stays free.
No. Kumanga has zero accounts, sign-ups or lock-in. Projects and images live on your machine, not on a company cloud, and you connect AI providers with your own API key.
Any compatible provider with your own key: Google Gemini, OpenAI-compatible endpoints, or a custom endpoint. You bring the key, so you choose the model and control the cost — Kumanga itself never bills you.
Most AI comics tools generate a finished page and leave you no way to change it. Kumanga generates reusable assets — characters, poses, expressions, backgrounds — and gives you a real non-destructive editor, so you can move a character without regenerating the panel or rewrite one line of dialogue without touching the art.
Yes — everything stays editable. The Manga Agent plans, reuses existing assets and commits a verified, fully editable result. A deterministic verifier checks every actor placement and that dialogue is byte-exact before a run lands; failed runs roll back cleanly.
Today: high-quality PNG at 1× or 2×, straight from the live scene graph, with provenance metadata on every generated asset. PDF, webtoon-strip and print-ready export are on the roadmap, funded by the current campaign.
No. Kumanga is local-first: projects and images stay on your machine. It is independent and open source — no VC, no ad model, no data harvesting — and the v0.1 codebase ships with 991 passing automated tests, including 98 dedicated security tests.
Kumanga is independent and open source — no VC, no ad model, no data harvesting. Every contribution goes straight into building the features on the roadmap. A rewards-based crowdfunding campaign is live on Pozible until 29 September 2026 — all-or-nothing, AU$5,000 — funding PDF, webtoon and print export, multi-project management and collaboration. GoFundMe is open for direct donations, and a Kickstarter campaign is in review.