Krea-2-Turbo is a local AI image generation model that produces high-quality images in about 3 seconds. It can be easily uncensored via prompt rebalancing and also supports image editing through masks, despite being marketed as text-to-image only.
I've been super impressed with Krea-2-Turbo. It can generate high quality images in ~3 seconds. The quality is quite good compared to other local AI image gen models. Now, I don't want to make you watch or click a you tube video, so I'll just give these clear instructions on how you can uncensor this model with simple prompt adherence re-balancing (NOTE: LoRAs work better for this FYI): Install SGLang diffusion: uv pip install 'sglang[diffusion]' --prerelease=allow Just give Codex or Claude Code this prompt: "add a rebalancer parameter to sglang diffusion and support it in the /v1/images/generations endpoint. This should allow me to pass in a post-prompt rebalancing / conditioning parameter like "1, 1, 1, 1, 1, 1, 1, 2.5, 5.0, 1.1, 4.0, 1.0" with a multiplier of "2" for the krea-2-turbo image model located at {INSERT PATH TO MODEL HERE}" ETA: BF16 model weights (OG model card) on Huggingface: https://huggingface.co/krea/Krea-2-Turbo GGUF (4bit is ~8gb): https://huggingface.co/vantagewithai/Krea-2-Turbo-GGUF And that's all you need to generate any image you want without any restrictions or blockers. You can also use reference images as a style/template to generate a ton of consistent images. OR or or or or or (oro!)... Use masks to actually edit images in this supposedly text-to-image-only model Full walk through: https://youtu.be/_kv2dZbD4II
Krea released Krea 2 Turbo, a 12-billion parameter text-to-image diffusion model, available open-weight on Hugging Face with support for multiple inference libraries.
Krea 2 is a 12-billion parameter text-to-image diffusion model released open-weight on Hugging Face, with Raw (base) and Turbo (post-trained) checkpoints available.
Krea 2 is a series of foundation models for creative image generation, built with a large-scale data infrastructure and multi-stage training pipeline. It introduces a prompt expander and style-reference system to improve steerability and enable creative exploration.
Alibaba’s 6B-parameter Z-Image-Turbo text-to-image model, further compressed by PrunaAI, generates 1024×1024 photorealistic images with bilingual text in <1s on 8 diffusion steps.