Black Forest Labs Releases New FLUX.2 Image Model
The new 9-billion-parameter text-to-image model uses a novel architecture that operates directly on pixels for faster, more efficient generation.
German AI research company Black Forest Labs has released FLUX.2 klein, a new 9-billion-parameter text-to-image model. The release marks the debut of the FLUX architecture, which is designed for high-speed, high-quality image generation and editing. The weights and code are available now on Hugging Face.
A Different Approach to Diffusion
Unlike many popular image models such as Stable Diffusion that use a VAE to work in a compressed latent space, FLUX.2 operates directly on pixels. The architecture is based on Flow Matching, a newer technique for generative modeling. It also processes text and image embeddings in parallel, a design choice intended to improve efficiency and the model's understanding of prompts.
The result is a model capable of generating 1024x1024 images in under a second on consumer-grade GPUs. Beyond standard text-to-image prompting, FLUX.2 is also equipped for a variety of image editing tasks, including:
- Inpainting and outpainting
- Style transfer
- Image-guided generation
This 9B model, named 'klein' (German for 'small'), is the first of two planned releases; a larger 50B 'groß' model is also in development. The model is released under a custom FLUX.2 Community License, which allows for creative and small-scale commercial use but requires a separate commercial license for entities with over $1 million in annual revenue.
Sources
- Visit
black-forest-labs/FLUX.2-klein-9B
Hugging Face
0 comments
No comments yet. Be the first to weigh in.
More in Text → Image
Ideogram 4.0: A 9.3B Open-Weight Text-to-Image Model
The new 9.3 billion parameter model uses a Diffusion Transformer architecture and excels at rendering coherent text within generated images.
ByteDance Releases Lance, a Unified Generative AI Model
The 3-billion-parameter model handles image and video generation, editing, and understanding from a single set of weights under a permissive license.
SenseTime Releases 8B 'Any-to-Any' Infographic Model
The new 8B-parameter SenseNova U1 model from SenseTime is designed for complex multimodal tasks, including the in-conversation generation and editing of infographics.