FLUX.2 is designed for real-world creative workflows, not just demos or party tricks. It produces high-quality images while maintaining consistency of character and style across multiple reference images, following structured cues, reading and writing complex text, following brand guidelines, and reliably handling lighting, layout, and logos. FLUX.2 can edit images up to 4 megapixels while preserving detail and coherence.
Black Forest Labs: Open Core
We believe that visual intelligence should not be shaped by just a few, but by researchers, creatives, and developers everywhere. That’s why we combine frontier capabilities with open research and open innovation, releasing powerful, observable, and composable open-weight models to the community, as well as robust, production-ready endpoints for teams that need scale, reliability, and customization.
When we launched Black Forest Labs in 2024, we set out to make open innovation sustainable, building on our experience in developing some of the world’s most popular open models. We have combined open models such as FLUX.1 [dev]- The most popular open image models globally – with professional-grade models like FLUX.1 Kontext [pro]That powers teams from Adobe to Meta and beyond. Our open source approach inspires experimentation, invites inquiry, reduces costs, and ensures we can continue to share open technology from the Black Forest and the Gulf with the world.
From Flux.1 to Flux.2
Precision, efficiency, control, extreme realism – where Flux.1 showed the potential of media models as powerful creative tools, Flux.2 shows how frontier capabilities can transform production workflows. By fundamentally changing the economics of generation, Flux.2 will become an essential part of our creative infrastructure.
output versatility:FLUX.2 is capable of generating highly detailed, photoreal images including infographics with complex typography, all at resolutions up to 4MP.
What’s new
- multi-context support: Reference up to 10 images together with the best character/product/style consistency available today.
- Image detail and photorealism: More details, clearer textures, and more stable lighting, suitable for use cases like product shots, visualizations, and photography.
- text renderingLegible: Complex typography, infographics, memes, and UI mockups with fine text now work reliably in production.
- advanced quick follow: Better following complex, structured instructions, including multi-part prompts and structured constraints.
- world knowledge: More based on real-world knowledge, lighting, and spatial reasoning, resulting in more consistent scenes with expected behavior.
- High resolution and flexible input/output ratio: Image editing at resolutions up to 4MP.
All variants of FLUX.2 offer text and image editing from multiple contexts in one model.
now available
The FLUX.2 family of models covers a spectrum of products, ranging from fully managed, production-ready APIs to open-ended checkpoints developers can run themselves. The overview graph below shows how FLUX.2 [pro]flux.2 [flex]flux.2 [dev]and flux.2 [klein] Balance performance and control
- flux.2 [pro], State-of-the-art image quality that rivals the best discontinued models, matches other models for quick adherence and visual fidelity, as well as generating images faster and at lower cost. No compromise between speed and quality. → Now available through BFL Playgrounds, BFL API, and our launch partners.
- flux.2 [flex]: Take control of model parameters such as the number of steps and guidance scale, giving developers complete control over quality, prompt adherence, and speed. This model is excellent at rendering text and fine details. → Now available through bfl.ai/play, the BFL API, and our launch partners.
- flux.2 [dev], 32B Open-weight model, derived from the FLUX.2 base model. The most powerful open-weighted image creation and editing model available today, combining text-to-image synthesis and image editing with multiple input images in a single checkpoint. flux.2 [dev] Weights are available on Hugging Face and can now be used locally using our reference estimation code. On consumer grade GPUs like GeForce RTX GPUs you can use an optimized fp8 reference implementation of FLUX.2 [dev]Created in collaboration with NVIDIA and ComfyUI. You can also sample Flux.2 [dev] Via API endpoints on FAL, Replicate, Runware, Verda, TogetherAI, Cloudflare, DeepInfra. For commercial licenses, visit our website.
- flux.2 [klein] ,coming soon, Open-source, Apache 2.0 model, shape-distilled from the Flux.2 base model. More powerful and developer-friendly than comparable models of similar size Trained from Scratch, with many of the same capabilities as its teacher model. join beta
- Flux.2 – VAE: A new variational autoencoder for latent representations that provides an optimized trade-off between learnability, quality, and compression rate. This model provides the basis for all FLUX.2 flow backbones, and an in-depth report describing its technical properties is available here. FLUX.2 – VAE is available on HF under the Apache 2.0 license.
Creating a Design with Variable Steps: flux.2 [flex] Typography provides a “step” parameter, balancing accuracy and latency. From left to right: 6 steps, 20 steps, 50 steps.
Controlling Image Details with Variable Steps: flux.2 [flex] Provides a “phase” parameter, reducing image detail and latency. From left to right: 6 steps, 20 steps, 50 steps.
The FLUX.2 model family delivers state-of-the-art image production quality at extremely competitive prices, offering the best value across performance levels.
For open-weight image models, FLUX.2 [dev] Sets a new standard, achieving leading performance in text-to-image generation, single-reference editing and multi-reference editing, consistently outperforming all open-source alternatives by a significant margin.
Whether open or closed, we are committed to responsible development of these models and services before, during, and after each release.
how it works
FLUX.2 builds on a latent flow matching architecture, and combines image creation and editing into a single architecture. This model combines the Mistral-3 24b parameter vision-language model with a rectified flow transformer. VLM brings real-world knowledge and contextual understanding, while Transformer captures spatial relationships, material properties, and structural logic that earlier architectures could not represent.
FLUX.2 now offers multi-context support, including the ability to combine up to 10 images into a new output, output resolution up to 4MP, significantly improved instant follow and world knowledge, and significantly improved typography. We retrained the latent space of the model to achieve better learning ability and higher image quality at the same time, which is a step towards solving the “learning ability-quality-compression” problem. Technical details can be found in the FLUX.2 VAE blog post.
more resources:
in new
We are building the infrastructure for visual intelligence, technology that changes the way we see and understand the world. FLUX.2 is a step closer to multimodal models that integrate perception, generation, memory, and reasoning in an open and transparent manner.
Join us on this journey. We are hiring in Freiburg (HQ) and San Francisco. View open roles,
<a href