Excited to see that
@krea_ai shipped UniFusion design with Krea 2! More frontier models now recognize that multimodal semantics are never encoded in just a single layer.
In the paper we showed that not only is Layerwise Attention Pooling better than all other conditioning methods, VLM hidden layers are perfectly capable of *reconstructing* images, thereby enabling editing without VAE inputs whatsoever.
Glad to see UniFusion proven effective at scale. Congrats to
@viccpoes and team, it’s a great model 🚀
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 👇