---
license: other
license_name: krea-2-community
license_link: https://www.krea.ai/krea-2-licensing
pipeline_tag: text-to-image
library_name: diffusers
tags:
  - text-to-image
  - image-generation
  - diffusion
  - flow-matching
  - dit
  - krea
  - turbo
base_model: krea/Krea-2-Turbo
base_model_relation: finetune
---

# Krea 2 (K2) Turbo - Diffusers

Diffusers-format conversion of the Krea 2 **Turbo** checkpoint from [Krea](https://krea.ai).
Turbo is an 8-step distilled model built for fast, high-quality text-to-image generation, and
is the checkpoint recommended for inference.

LoRAs trained on the undistilled
[Krea-2-Base-Diffusers](https://huggingface.co/CalamitousFelicitousness/Krea-2-Base-Diffusers)
apply directly to Turbo, so the recommended workflow is to train on Base and run on Turbo.

## Model Summary

Krea 2 is a latent-diffusion image model trained from scratch with an emphasis on aesthetics
and stylistic control. The architecture is a single-stream multimodal diffusion transformer.

- **Transformer**: single-stream DiT, 12.9B parameters, 28 blocks at width 6144. Grouped-query
  attention, a learned output gate, per-head QK normalization, and a 3-axis rotary embedding.
  A text-fusion stage inside the transformer collapses twelve text-encoder hidden-state layers
  into one conditioning stream.
- **Text encoder**: [Qwen/Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct),
  tapped at twelve intermediate layers (text-only conditioning).
- **VAE**: the Qwen-Image autoencoder (`AutoencoderKLQwenImage`, f8, 16 latent channels).
- **Sampler**: flow matching with a fixed timestep shift.

Weights are stored in their original mixed precision (bf16 matmuls, fp32 norms and modulations).
The transformer config carries `is_distilled: true`, so guidance is disabled automatically.

## Recommended Settings

Turbo is distilled for few-step sampling and runs without classifier-free guidance.

| Setting | Value |
| ------- | ----- |
| Steps | 8 |
| Guidance (CFG) | 0 (disabled) |
| Resolution | 1024 x 1024 up to 2048 x 2048 |

The timestep shift is pinned (`mu = 1.15`), matching the distillation schedule.

## Prompting

Natural-language prompts are recommended. Long, detailed descriptions yield the best results,
though strong images are produced from short prompts as well. For text rendering, the words to
be rendered are wrapped in quotes. An optional prompt-expansion system prompt is available in
the upstream [krea-2-oss](https://github.com/krea-ai) repository.

## License

The weights are released under the [Krea 2 community license](https://www.krea.ai/krea-2-licensing).

## Citation

```bibtex
@misc{krea2,
  title  = {Krea 2},
  author = {Krea},
  year   = {2026},
  url    = {https://www.krea.ai/krea-2}
}
```