---
license: mit
base_model: MiniT2I/MiniT2I
base_model_relation: finetune
pipeline_tag: text-to-image
tags:
- budgetany
- budgetpix
- adaptive-tokenization
- diffusion
- text-to-image
- pixel-space
---

# BudgetPix-MiniT2I

🌐 [Project page](https://budgetany.github.io/budgetpix/)  |  💻 [Code](https://github.com/BudgetAny/BudgetPix/tree/main/budgetpix/minit2i)  |  📄 [Paper](https://arxiv.org/abs/2610.12307)  |  🧩 [BudgetAny](https://huggingface.co/BudgetAny)

**One checkpoint, any token budget.** Text-to-image generation at 512 x 512 in pixel space with [MiniT2I](https://huggingface.co/MiniT2I/MiniT2I), fine-tuned with BudgetPix compute-adaptive tokenization. At inference you choose how many tokens the diffusion transformer runs on: an entropy-based detail map merges the flattest regions into coarse tokens first and gives the fine tokens to the most detailed ones. The same weights serve every budget.

> **Unofficial, open-source implementation**, taken from [budgetpix/BudgetPix-public](https://github.com/budgetpix/BudgetPix-public):
> a public reimplementation of BudgetPix: compute-adaptive tokenization for pixel-space image diffusion (JiT, MiniT2I,
> PixelDiT, pMF). This model is part of [BudgetAny](https://huggingface.co/BudgetAny), a community reimplementation of
> [BudgetPix](https://arxiv.org/abs/2610.12307) (Kara et al., 2026). It is not an official release.
> Thanks to the BudgetPix authors for the method.

## Files

| file | base model | dense tokens | parameters | size | SHA-256 |
|---|---|---|---|---|---|
| `budgetpix_minit2i_b_512.pt` | MiniT2I-B/16, 512 px | 1,024 | 260 M | 1.04 GB | `84517824ac2f823e…` |
| `budgetpix_minit2i_l_512.pt` | MiniT2I-L/16, 512 px | 1,024 | 915 M | 3.66 GB | `3868f0a4faa6ff48…` |

`SHA256SUMS` lists the full checksums. The text encoder (`google/flan-t5-large`) is downloaded from the Hub on first use.

## Usage

```bash
git clone https://github.com/BudgetAny/BudgetPix && cd BudgetPix    # setup: see README.md
hf download BudgetAny/BudgetPix-MiniT2I budgetpix_minit2i_l_512.pt config.json --local-dir weights
python -m budgetpix.minit2i.sample --ckpt weights/budgetpix_minit2i_l_512.pt --prompts "a corgi in a spacesuit" --budget 512 --cfg 6.0
```

`--budget` is the number of tokens per image (1,024 is the dense grid). Default guidance 5.0 (6.0 recommended for L, 2.5-5.0 for B at the dense grid; about 7.0 at 25 % of the grid).

## Citation

If you use this model or the code, please cite the BudgetPix paper:

```bibtex
@article{kara2026budgetpix,
  title={BudgetPix: Compute-Adaptive Tokenization for Pixel-Space Image Diffusion},
  author={Ozgur Kara and Yujia Chen and Daniel Watson and David Forsyth and James Matthew Rehg and Wen-Sheng Chu and Du Tran},
  journal={arXiv preprint arXiv:2610.12307},
  year={2026}
}
```

Built on [MiniT2I](https://github.com/Hope7Happiness/minit2i-torch) and [APT](https://github.com/rccchoudhury/apt).

## License

MIT, like the MiniT2I weights it derives from.