Krea opens the weights: Raw for fine-tuning, Turbo for fast 2K generation
- Source
- x.com
- Date
today, we release the open weights of Krea 2. welcome Krea 2 Raw and Krea 2 Turbo, an undistilled model from mid-training meant to be fine-tuned, and a fast distilled version with a wide aesthetic diversity. read the details below 👇
we are releasing two versions of our foundation image model: Krea 2 Raw and Krea 2 Turbo. combined, they bring unprecedented opportunities for fine-tuning, post-training, and working with diverse aesthetics. the core idea is: train on Raw, generate with Turbo.
Krea 2 Raw is an undistilled model from mid-training stage that we created while training Krea 2 Medium. The lack of distillation, fine-tuning, or post-training make it an ideal model to build on.
Krea 2 Turbo is a distilled and post-trained version of Krea 2 Medium. This model can produce imagery in a wide range of aesthetics fast at 2k native resolution, and it can run in consumer hardware.
Context
Krea released two downloadable versions of its image model for different jobs. Krea 2 Raw is an intermediate training checkpoint intended for developers who want to adapt the model. Krea 2 Turbo has undergone additional training and for faster image generation. Release thread.
That distinction determines where customization work begins. Raw exposes a model before the later stages that shape its finished behavior, while Turbo targets generating images efficiently. Krea's technical report describes distillation as an optional late stage in its training pipeline. The two releases therefore offer different starting points for model builders and people running generation workloads.
- distillation — Training a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.
- open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
- fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
Checking sign-in…
Loading comments…