1sglang generate --model-path krea/Krea-2-Turbo \2 --prompt "a red fox sitting in fresh snow, golden hour, photorealistic"\3 --num-inference-steps 8 --height 1024 --width 1024 --save-output
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:
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.
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.
