⚠️ Experimental – first version. MOSS-Image is new and this kit was built and tested in a few days. Expect bugs and rough edges. Feedback and bug reports in the comments are very welcome.

🖥️ Now available: Fumika, the standalone app. The same doujin pipeline in a free desktop app for Windows and Linux: one installer, models downloaded and managed for you, series and chapters, a story editor with panel cards, drawn covers, a reader and PDF / CBZ export. No nodes, no workflows. GitHub · Civitai article. This ComfyUI kit stays available if you prefer ComfyUI.

Make complete manga / doujin pages, or a whole book in one Run, with MOSS-Image. You get panel borders, readable speech bubbles, sound effects and title lettering, and the same characters on every page. No JSON, no photo editing, no node spaghetti.

How it works (easy mode)

Open Templates → ComfyUI-MOSS-Doujin → Make a whole manga and fill in 3 fields:

characters:  Tifa: Tifa Lockhart, an adult woman with long black hair, red eyes, white tank top, black mini skirt
             Aerith: Aerith Gainsborough, an adult woman with a long brown braid with a pink ribbon, green eyes, long pink dress

story:       Tifa and Aerith pose together in a neon-lit street at night | #SLUM NIGHTS

             Tifa wipes the counter of a cosy bar at night | Slow night...
             Aerith bursts through the door holding a flyer | I found us a job!
             Tifa and Aerith lean over the counter face to face | Night work... good pay?

• One block per page, with an empty line between pages.

• One line per panel: what happens | what she says. The number of lines (1–5) picks the panel layout.

• ! before a text = sound effect, # = big title (covers).

• Outfit changes: define the same character twice with different names, e.g. Hooker Tifa: …, and use that name in the story.

• Press Run. Every page is drawn 3 times and the version with the fewest gibberish bubbles is kept. Pages + PDF are saved to output/doujin/<title>/.

There is also Make a manga page (one page per Run) for trying things out.

Optional settings below the main fields (all have working defaults):

• character LoRA and its strength

• base model (MOSS Base or a fine-tune like Kromoss)

• speed: fast / quality / auto

• versions per page

• page size

• seed

Speed

About 5 s per page version on an RTX 5090 with the 8-step turbo LoRA. A 30-page doujin takes about 8 minutes. Title pages automatically use the slower 50-step mode, which spells titles correctly.

Models (put them in your normal ComfyUI folders; any sub-folder / file name works)

• MOSS-Image Base (or a MOSS fine-tune) → models/diffusion_models

• Qwen3.5-9B text encoder of MOSS-Image → models/text_encoders

• FLUX.2 VAE → models/vae

• 8-step DMD turbo LoRA (optional, much faster) → models/loras

The nodes find the files by name (auto). If yours are named differently, pick them in the dropdowns.

Downloads (right-click → save into the folder in brackets):

• MOSS-Image Base: bf16, 24 GB+ VRAM or int8, smaller GPUs (models/diffusion_models)

• Qwen3.5-9B text encoder (models/text_encoders)

• FLUX.2 VAE (models/vae)

• 8-step turbo LoRA (models/loras)

Install

Extract the zip into ComfyUI/custom_nodes/, restart ComfyUI and press F5. The workflows are under Templates. Needs a ComfyUI version with MOSS-Image support. The gibberish-bubble check needs opencv: if it is missing (e.g. a fresh ComfyUI portable), the pack installs opencv-python-headless by itself on the first start.

Tips

• Write each character once in full (hair, eyes, body, outfit) and use the same text in every page.

• Keep bubbles short (up to about 12 words); 3–4 panels per page look best.

• Black-and-white style LoRAs: no colour words. Describe the characters by shape (hair length, glasses, body, outfit type), not by colour ("red hair", "black blouse"). Colour words override the ink look.

• For explicit panels, start the line with "Explicit uncensored:" and name what is visible. MOSS-Base is uncensored.

• Famous characters (FF7, Fate, Re:Zero …) work without a LoRA. Niche or original characters need a character LoRA; make one with the MOSS LoRA Trainer kit.

For advanced users: the zip also contains the building-block nodes and 4 advanced workflows in example_workflows/advanced/:

• free JSON page layouts

• double-page spreads

• fixed-camera pages

• stray-bubble cleanup

Full guide: Whole doujins with MOSS-Image. License: MIT.