---
license: other
license_name: modified-mit
license_link: https://huggingface.co/AiArtLab/sdxs-2b/blob/main/LICENSE
pipeline_tag: text-to-image
widget:
  - text: sdxs-2b
    output:
      url: media/refined.jpg
---

# Simple Diffusion XS (sdxs-2b alpha version)

*XS Size, Excess Quality* 

Train status: 4xRTX5090 / we need more gold / [support us please..](https://huggingface.co/AiArtLab/sdxs-2b#donations)

At AiArtLab, we strive to create a free, compact and fast model that can be trained on consumer graphics cards.

- CosmosTransformer3D: 2b parameters
- Qwen3.5:             0.8b parameters
- Qwen vae:            16ch8x
- Resolution:          Default 768x1152, trained from 576 to 1152 with step 64
- Limitations:         trained on small datasets ~2.5kk, focused on art / illustrations / anime and photo
- Captions:            danbooru, natural (short/medium), trained with 250 max toks


### Key points

<img src="media/start.png" height="256"/>

 - Apr 2026: Started research of Cosmos Transformer
 - May 2026: 1st preview (Initial model weights)
 - May 2026: finetuned Qwen3.5-0.8B
 - May 2026: hybrid EDM/Karras scheduler
 - May 2026: img2img pipeline and txt2video pipeline (not trained)
 - May 2026: [telegram bot](https://t.me/inkimpbot)

### Random samples
![promo](media/result_grid.jpg)

### Text 2 image

```python
#!pip install -U torch torchvision
#!pip install -U diffusers accelerate transformers

import torch
from diffusers import DiffusionPipeline

device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float32

pipe_id = "AiArtLab/sdxs-2b"
pipe = DiffusionPipeline.from_pretrained(
    pipe_id,
    torch_dtype=dtype,
    trust_remote_code=True
).to(device)

refined = "A blonde-haired Red Eyes girl with a hair ribbon, half-updo, and tsurime stands solo in a flower field holding a bouquet with a serene smile, wearing green overalls, a white shirt, rolled-up sleeves, and a straw hat with a flower while looking at the viewer under volumetric and natural lighting with a Dutch angle."

negative_prompt = "low quality, bad quality, blurry, sketch, sepia, text, bad anatomy, bad proportions, bad hands, missing fingers, child drawing"

output = pipe(
    prompt=refined,
    negative_prompt=negative_prompt,
)

image = output.images[0]
image.show()
```

### Prompt refiner
```python
refined = pipe.refine_prompts("1girl")
print(refined[0])
```

## Donations
Donated: 25$ 

Thanks for your support!

 - Euro / dollars: sorry, we can’t accept payments via Patreon or Ko-fi. Please register on [vast.ai](https://cloud.vast.ai/billing/) (this is the GPU provider we use for training the model), top up any amount, and transfer: Transfer Money / User / Email: vadim-kulibaba@yandex.ru.
 - RUB: [donate in rub](https://www.tbank.ru/cf/90ensBQqpJj)
 - DOGE: DEw2DR8C7BnF8GgcrfTzUjSnGkuMeJhg83
 - BTC: 3JHv9Hb8kEW8zMAccdgCdZGfrHeMhH1rpN
 - USDT
   - Ethereum / Polygon / BNB SmartChain: 0xD4388B6698dFaE1460E72099D4F208aaCA4f6E6C
   - Tron: TD7ey4h9igPGdcrcBcnZaz56R5tNgRZNvV
   - Solana: MMYFJeYEtYHrSNFHChytJDHbEDniXrnAxPNLhJ1LbkB

## Contacts
Please contact with us if you may provide some GPU's or money on training

 - telegram [recoilme](https://t.me/recoilme) *prefered way
 - mail at aiartlab.org (slow response)

## Citation
```bibtex
@misc{sdxs,
  title={Simple Diffusion XS-2b},
  author={recoilme and AiArtLab Team},
  url={https://huggingface.co/AiArtLab/sdxs-2b},
  year={2026}
}
```