Sana Manga Coloring

Sana Manga Coloring

Colors black-and-white manga pages with two Sana 600M models, one that adds color on top and one that repaints the whole page, for artists and fans working on their own pages.

Made by masterofaudio2077 on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Sana Manga Coloring takes a black-and-white manga page and adds color using a pair of Sana 600M models trained for the task. The Tagged model adds color and leaves the page almost as it was. The Painted model repaints the entire page and afterwards restores the original linework and lettering. One container runs the model on the CPU and serves the web interface.

Who it suits: Artists and readers who want color drafts of manga pages they own or draw.

Industry | Who uses it | How they use it | Example
Comics | Independent manga artists | Draft a colored version of a page | Testing a dusk mood on a finished page
Hobby art | Fans with their own black-and-white pages | Add color without painting by hand | Coloring a page they drew
Art education | Students | Compare the Tagged and Painted looks on one page | Running both models on the same page

If you need results in seconds, you can skip it, because the free CPU takes a minute or more for each page.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• Docker on Windows, macOS or Linux, for a copy you run yourself
• Scripts that call the app's job-based web API

Use case | Who | What they get back
Color with tags | Artists | The page with colors added on top of the drawing
Repaint a page | Illustrators | A page that looks closer to a finished illustration
Batch from a script | Developers | A job they can start, watch and fetch through the API

How to set it up: Online, nothing needs installing. A copy on your own computer needs Docker and a little Python.

• Docker
• Python and the pip tool, only for the download step
• Room for model weights, since the development notes say the backend downloads about 4 GB on its first start

1. Install the Hugging Face hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Build the Docker image.
5. Run the container with port 7860 open.

pip install -U huggingface_hub
hf download masterofaudio2077/Sana-Manga-Coloring --repo-type space --local-dir Sana-Manga-Coloring
cd Sana-Manga-Coloring
docker build -t sana-manga-coloring .
docker run -it -p 7860:7860 sana-manga-coloring

How to use it: Open the page and upload a PNG, JPEG or WebP page up to 15 MB. Pick the Tagged or the Painted model, set its options, and choose Draft or Standard speed. The page shows your place in the queue and the current step. Download the result within 15 minutes.

docker run -it -p 7860:7860 sana-manga-coloring

Tool | What it does
Tagged model | Adds color with five optional tags for mood, time of day, lighting, saturation and palette
Painted model | Repaints the page with a Look, Guidance and Ink setting, then restores the original lines and text
2x upscale | Sharpens the result to 1024 pixels with Real-CUGAN
Draft and Standard | Run 8 or 20 steps, trading speed for quality
Queue view | Shows queue position and step, and a busy server asks you to retry

Example: You upload a black-and-white page, keep the Tagged model and set saturation to vivid. After about a minute or two on the free hardware, you get the same page back with colors added over the drawing. The lines and text stay almost unchanged.

For newcomers and experts: The tool guesses colors for a page that has none. To start, upload a small page, keep the Tagged model and use Draft speed so you see a result sooner.

Both models color at 512 pixels on the short side, and the optional upscale brings the result to 1024. The Tagged model was trained on color pages that carry those five tags. The Painted model was trained on roughly 29,000 panels, each matched with a painted copy that a different AI produced. Only one model is loaded at a time, so after a switch the next page waits about half a minute for the new model.

Upload a black-and-white page
Pick the Tagged or the Painted model
Queue holds the job, then the model colors it
Optional 2x upscale, then download the PNG

Good to know:

• Limits: a page can be a PNG, JPEG or WebP file, 15 MB at most, one runs at a time with three queued, and output stops at 1536 pixels
• License: Apache 2.0, and the upscaler Real-CUGAN is listed as MIT
• Cost to run: none named, and the free cpu-basic tier gives you 2 vCPUs
• Data it reaches: the pages you upload, kept as finished jobs for 15 minutes, and the models named in its metadata
• Not for: fast results or large pages, since larger pages take longer

Title: Time to color one 672x512 page on the free CPU
Draft, 8 steps: 61 s
Standard, 20 steps: 112 s

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: docker
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/masterofaudio2077/Sana-Manga-Coloring
• Files: https://huggingface.co/spaces/masterofaudio2077/Sana-Manga-Coloring/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.