unsloth/Krea-2-Turbo-FP8

This is an FP8 / INT8 quantized version of Krea-2-Turbo.

• Optimized for efficient inference with reduced memory footprint. Same-seed LPIPS vs the bf16 model (lower is better): 0.223 INT8, 0.172 FP8.

---

Samples

Prompt: "cute sloth typing on a computer"

INT8	INT8
FP8	FP8

---

Krea 2 Text-to-Image Model

Inference with the official codebase

1. Setup the official Krea 2 codebase
2. Download turbo.safetensors in this repo
3. export OSS_TURBO=<path-to-turbo.safetensors> Run inference:

   uv run inference.py "a fox walking in the snow" \
    --checkpoint oss_turbo --steps 8 --cfg 0.0 --mu 1.15 --width 2048 --height 2048
   

Inference with diffusers

Install diffusers from source (for Krea2Pipeline):

pip install git+https://github.com/huggingface/diffusers.git

import torch
from diffusers import Krea2Pipeline

pipe = Krea2Pipeline.from_pretrained("krea/Krea-2-Turbo", torch_dtype=torch.bfloat16).to("cuda")
image = pipe("a fox in the snow", num_inference_steps=8, guidance_scale=0.0).images[0]
image.save("krea2.png")

Inference with SGLang

Install SGLang from source (https://github.com/sgl-project/sglang)

From the CLI:

sglang generate --model-path krea/Krea-2-Turbo \
    --prompt "a red fox sitting in fresh snow, golden hour, photorealistic" \
    --num-inference-steps 8 --height 1024 --width 1024 --save-output

See the full SGLang Krea 2 Cookbook here

Model Overview

• Model Name: Krea 2
• Version: v1.0
• Release Date: June 22, 2026
• Model Type: Text-to-image diffusion model
• Architecture: Diffusion Transformer with 12 billion parameters
• License: Krea 2 Community License
• Release Format: Open-weight release and Krea-hosted product integrations
• Model Developer: Krea.ai, Inc.

Model Family and Release Checkpoints

This model card covers the Krea 2 model family, including the following release checkpoints:

• Krea 2 Raw: Base release checkpoint, prior to additional post-training and fine-tuning.
• Krea 2 Turbo: Post-trained release checkpoint with additional fine-tuning and distillation.

Capabilities and Intended use

Krea 2 is a text-to-image diffusion model that generates images from natural-language text descriptions. The model is designed to support creative, commercial, developer, and research use cases, including image generation, concepting, design exploration, visual production workflows, and integration into applications and creative tools.

Out-of-Scope Uses

This model is not intended or designed for uses that violate applicable law or regulations, infringe or misappropriate third-party rights, generate or facilitate unlawful or harmful content (including CSAM, NCII, harassment or defamation), or support fully automated decision-making that adversely affects legal rights of individuals. This summary is non-exhaustive. Use of Krea 2 is subject to the Krea 2 Community License Agreement and must comply with the Acceptable Use Policy. In the event of any conflict, the Krea Acceptable Use Policy and Krea 2 Community License control.

Training Data

This model was developed using a combination of publicly available data, data licensed from third-party providers, and synthetic data generated through proprietary methods. The training data includes images and their associated captions or text descriptions.

Prior to training, data was filtered to remove certain categories of harmful content and reduce low-quality, duplicative, or irrelevant data. Krea also used curated and synthetic training data selected to improve prompt following, visual quality, and alignment with intended use cases.

Safety Measures

We implemented safety measures across the full model development lifecycle. We applied targeted fine-tuning techniques to reduce the model's susceptibility to generating harmful content in response to both direct and adversarial prompts, and we conducted multiple rounds of internal and external safety evaluation before release.

For Krea's hosted products incorporating Krea 2, we deploy input and output classifiers using a combination of proprietary and third-party detection tools to flag or block policy-violating prompts and generated images.

Because this is also an open-weights release, Krea does not control downstream deployment of the model. Under the Krea 2 Community License, deployers are required to implement content filtering measures or equivalent review processes to prevent the generation or distribution of unlawful or policy-violating content appropriate to their use case. Deployers who fail to implement required safeguards are in breach of the license. See the license for details.

We conducted multiple rounds of internal and external safety evaluations before release, including adversarial testing designed to assess the model's resilience to attempts to elicit harmful or policy-violating outputs. Testing covered sexually explicit content, non-consensual intimate imagery, child-safety risks, and other high-risk content categories. Based on these evaluations, the release checkpoints demonstrated high resilience against violative inputs across the tested risk categories.

Krea maintains reporting channels for harmful, illegal, or policy-violating outputs at safety@krea.ai. Reports involving potential CSAM are escalated to NCMEC as required by law. Krea reserves the right to update model weights or revoke access in response to identified misuse patterns.

Risks and Limitations

Krea 2 is a new technology and there are risks associated with its use. Testing conducted to date has not covered, nor could it cover, all possible scenarios. The model's potential outputs cannot be predicted in advance and may, in some instances, produce inaccurate, objectionable, or otherwise undesirable outputs.

This model is not intended to provide factual information. The model may fail to generate output that matches the prompt, and prompt following may be influenced by prompt style, specificity, language, and phrasing.

Before deploying any application using this model, developers should perform safety testing and tuning tailored to their specific application and must implement safeguards required by the Krea 2 Community License.

License and Outputs

Krea does not claim copyright or other intellectual property rights over content generated by users of this model. Users are solely responsible for their outputs and any subsequent use of those outputs. As with other generative tools, the nature of a user's inputs influences the outputs produced, and prompts may produce images that implicate third-party rights. Users are solely responsible for assessing and addressing those risks. See the Krea 2 Community License for more information.

Reporting

To report harmful, illegal, or policy-violating outputs generated by this model:

• Email: safety@krea.ai

Reports involving potential CSAM will be escalated to the National Center for Missing & Exploited Children (NCMEC) as required by law.

Contact

• General inquiries: support@krea.ai
• Safety and abuse: safety@krea.ai
• Model card version: 1.0
• Last updated: June 22, 2026

---

Quantized transformer checkpoints (this repo)

This repo adds pre-quantized diffusion transformer checkpoints for krea/Krea-2-Turbo, built with torchao dynamic activation quantization from the dense bf16 transformer. The official model card above is unchanged from the source repo.

License and modifications

Krea 2 is licensed under the Krea 2 Community License Agreement. For more information, visit https://krea.ai/krea-2-licensing. A copy of the Agreement is distributed in this repo as LICENSE.pdf, and the attribution notice is in NOTICE. Use of these weights is subject to that Agreement and to the Krea Acceptable Use Policy; by downloading them you agree to both.

This is a modified version of Krea 2 Turbo, not an official Krea product and not endorsed by Krea. The modifications are the quantization and pre-cast described on this page; the weights are otherwise those of krea/Krea-2-Turbo.

Files:

• Krea-2-Turbo-INT8.pt (13.1 GB)
• Krea-2-Turbo-FP8.pt (12.8 GB)

Details:

• int8: Int8DynamicActivationInt8WeightConfig (per-token activation, per-channel weight, torch._int_mm).
• fp8: Float8DynamicActivationFloat8WeightConfig with PerRow granularity (e4m3, torch._scaled_mm). The loader must floor the dynamic activation scale (activation_value_lb=1e-12 on torchao 0.13+) so all-zero activation token rows cannot produce a zero scale.
• Loading a checkpoint is bit-identical to quantizing the dense bf16 transformer on the fly; the checkpoint skips the dense load and quantize step.
• Validated against same-seed dense bf16 renders (SSIM, LPIPS-vgg, CLIP delta, non-finite and black-frame checks) on torch 2.12.1 and torchao 0.17.

Samples

Prompt: "cute sloth typing on a computer" (1024x1024, family default steps/guidance, seeds 0-2).

int8

fp8

Pre-cast fp8 text encoder (this repo)

Krea-2-Turbo-text_encoder-FP8.pt (4.8 GB) is the pipeline's text encoder (Qwen3VLModel, Qwen3-VL-4B, the text_encoder subfolder of krea/Krea-2-Turbo) with the layerwise fp8 storage cast Unsloth Studio applies at load time, saved pre-cast:

• Loading it is bit-identical to downloading the dense encoder and casting on load (verified tensor for tensor: 713 tensors, 460 cast to fp8 storage).
• Cuts the text-encoder download from 8.9 GB to 4.8 GB.
• Plain-tensor state dict: loads with torch.load(weights_only=True). Metadata records scheme fp8, component text_encoder, base krea/Krea-2-Turbo.