Black Forest Labs Releases Open FLUX.2 Image Decoder
This new component is part of a novel transformer-based architecture for text-to-image generation, released under a permissive Apache 2.0 license.

Black Forest Labs has released a key component of its new image generation architecture, the FLUX.2 small decoder. Available under a permissive Apache 2.0 license, this decoder is designed for high-performance text-to-image and image editing tasks, signaling a new contender in the open-source creative AI space.
The release is significant not just for the tool itself, but for the underlying architecture it represents. Unlike established latent diffusion models like Stable Diffusion that rely on a U-Net, the FLUX architecture uses a transformer-based design. The goal is to process text and image information in a more unified, parallel manner, which could lead to improvements in speed and coherence.
This release contains only the "small" decoder model. As detailed in the official repository, the decoder is one part of a larger system that includes a main transformer for processing prompts and generating image latents. The full model suite, including a larger variant, has not yet been published, making this a component-first preview of the new technology.
By open-sourcing this foundational piece, Black Forest Labs is giving developers an early look at a different approach to image synthesis. The move away from U-Nets towards diffusion transformers mirrors trends in the field, and the arrival of FLUX.2 suggests the open-source ecosystem will soon have another powerful and distinct architecture to build upon.
Sources
- Visit
black-forest-labs/FLUX.2-small-decoder
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.