P-JiT 21K

P-JiT 21K

Generates images from text prompts with P-JiT 21K, a diffusion transformer that works directly on pixels. Includes a prompt expander for short prompts. For image generation researchers and developers.

Made by hugging-apps 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: P-JiT 21K is a demo of a text to image model that works in pixel space with a diffusion transformer. Its tagline says it was trained for 21 thousand dollars. You type a prompt and get an image.

The Space uses the authors' inference code, and its example prompts and seeds are samples from the authors' gallery. A short prompt can be expanded first by a built-in prompt upsampler.

Who it suits: People who want to try a low-budget research image model, and developers who compare image generators.

Industry | Who uses it | How they use it | Example
Research | Image generation researchers | Look at output from a pixel-space model | Run one of the gallery prompts
Software | Developers | Judge the model before integrating it | Try short and long prompts
Design | Visual artists | Explore what the model draws | Type a short idea and let the upsampler expand it

You can skip it if you need a finished image tool with editing features, since this Space is a demo.

Where to use it:

• A web browser, on the Space's page at Hugging Face, with nothing to install
• Windows, macOS or Linux with Python, from a copy you download
• An MCP client, since the Space carries an MCP tag, though the source gives no setup for it

Use case | Who | What they get back
Generate an image | Anyone | An image for your prompt
Expand a short prompt | Developers | A longer prompt written by the text encoder
Replay a gallery sample | Newcomers | The authors' example prompt and seed

How to set it up:

• Python 3.12 on your computer
• The Python packages listed in requirements.txt
• The huggingface_hub package, which gives you the hf command

1. Install the huggingface_hub package.
2. Download the Space with the hf command.
3. Move into the downloaded folder.
4. Install the pinned Gradio version.
5. Install the packages from requirements.txt.
6. Start the app with Python.

pip install -U huggingface_hub
hf download hugging-apps/pjit-21k --repo-type space --local-dir pjit-21k
cd pjit-21k
pip install gradio==6.31.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space's page in your browser, or start your own copy with the command below. Type a prompt or pick an example, and run it to get the image.

python app.py

Example: You pick one of the gallery prompts and its seed, then run it. The demo returns an image generated from that prompt, which you can compare with the authors' sample.

For newcomers and experts: If you are new to image models, start with a gallery example so you know what a good result looks like. Then change a few words and run it again.

If you know the field, the model denoises in pixel space instead of a learned latent space. In the original setup, prompt expansion sends the prompt to Claude via OpenRouter. This Space swaps that step for Qwen3.5-4B, which is already loaded as the text encoder, so no outside service is needed.

You type a prompt or pick an example
Qwen3.5-4B can expand a short prompt
The diffusion transformer denoises in pixel space
The image shows on the page

Good to know:

• License: the Space lists apache-2, and the README says the vendored inference code is Apache-2.0
• Cost to run: none named, since the prompt upsampler runs locally instead of calling a paid service
• Data it reaches: the prompts you type, plus the model weights and the text encoder
• Not for: reproducing the authors' hosted upsampler, which the Space replaces with a smaller local model

On Hugging Face, 13 November 2026

• Likes: 2
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/hugging-apps/pjit-21k
• Files: https://huggingface.co/spaces/hugging-apps/pjit-21k/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.