Tessera Pixel Studio

Tessera Pixel Studio

Paints 16 by 16 pixel art from a text prompt and lets you watch the sprite appear step by step, then edit it. Built around a small diffusion model. Free for personal, research and education use only.

Made by thejagstudio 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: Tessera Pixel Studio makes small pixel art sprites from text. The engine is Tessera, a language model with 47.6M parameters that uses masked diffusion. It draws a sprite on a 16 by 16 grid and gives each pixel one of 512 learned palette colors. The Space adds a full studio with live generation, a gallery and an editor.

Who it suits: It suits game hobbyists, pixel artists and students who want sprites or want to see how a small diffusion model works, and researchers who study masked diffusion. The license allows non-commercial use only.

Industry | Who uses it | How they use it | Example
Game development | Hobbyist developers | Generate small sprites for personal projects | Make a batch of character sprites
Pixel art | Artists | Start from a generated sprite and edit it | Repaint part of a sprite with the Magic brush
Education | Teachers and students | Watch a diffusion model build an image step by step | Switch to the confidence view and discuss it
Research | ML researchers | Study a masked-diffusion model of images | Read the technical write-up

You can skip it if you need sprites for a paid game or app, since commercial use needs a separate license from the author.

Where to use it:

• A web browser: open the Space page on Hugging Face
• Windows, macOS or Linux: run a copy with Python 3.12

Use case | Who | What they get back
Generate sprites | Game hobbyists | One to sixteen sprites per batch from a prompt
Edit a sprite | Pixel artists | A sprite repainted where you marked it
Learn how it works | Students | A live view of the picture forming over the steps
Export | Anyone | A 512 pixel PNG of a chosen sprite

How to set it up: The online page needs nothing. A local copy needs Python 3.12 and the packages in requirements.txt.

• A web browser for the online demo
• Python 3.12 for a local copy
• The packages listed in requirements.txt

1. To use it online, open the Space page on Hugging Face.
2. To run a copy, install the Hugging Face hub package.
3. Download the Space into a local folder.
4. Move into the folder and install Gradio 6.29.1 and the listed packages.
5. Start the app.

pip install -U huggingface_hub
hf download thejagstudio/tessera-pixel-studio --repo-type space --local-dir tessera-pixel-studio
cd tessera-pixel-studio
pip install gradio==6.29.1
pip install -r requirements.txt
python app.py

How to use it: Start the app from its folder, or open the Space online. Type a prompt, choose how many sprites you want and generate. Watch them form, then open one in the detail viewer or the pixel editor to refine it.

python app.py

Tool | What it does
Live unmasking | Streams every step so you watch the sprite appear, with Ghost and Confidence views
Batches | Makes 1 to 16 sprites with a seed lock and sampler settings
Explore gallery | Shows 56 curated sprites that replay their own generation on hover
Detail viewer | Offers a step scrubber, palette breakdown, PNG export, variations and Remix
Pixel Editor | Paints with the model's colors and repaints marked pixels with the Magic brush
How it works | Opens a technical write-up of the model and its training

Example: You type a prompt for a small character and generate a batch of four. You watch the pixels fill in, pick the sprite you like, mark one patch with the Magic brush and let the model repaint it. Then you export it as a PNG.

For newcomers and experts: If you are new to this, generate one batch and open the Explore gallery. Hover over a sprite to see how it was formed, then try your own prompt.

If you know the field, the design does without a VAE, a latent space and autoregressive decoding. Generation begins with every pixel hidden, then reveals all 256 together over roughly 16 passes. Text conditioning comes from a frozen CLIP text encoder, and training used a pixel art dataset released by another author on one GPU.

You type a prompt
The canvas starts fully masked
About 16 parallel unmasking steps run
Each pixel gets a palette color
You edit or export the sprite

Good to know: The license is CC BY-NC 4.0, so the model, code and generated images are free for education, research and personal projects with credit to the author, Jagrat Patel. The author reports that generated sprites score about as well on a CLIP measure as real ones.

Title: CLIP score of sprites
Generated sprites: 25.24 score
Real sprites: 25.28 score

• License: CC BY-NC 4.0
• Cost to run: none named
• Data it reaches: the prompts you type, plus model weights loaded from Hugging Face
• Not for: paid products, since the model, code and its images need a license for commercial use

On Hugging Face, 13 November 2026

• Likes: 6
• Built with: gradio
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/thejagstudio/tessera-pixel-studio
• Files: https://huggingface.co/spaces/thejagstudio/tessera-pixel-studio/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.