FLUX.2 is built for real creative production—not just impressive demos. It delivers high-quality images with consistent characters and styles across multiple reference inputs, accurately follows structured prompts, handles complex text both visually and semantically, respects brand requirements, and reliably manages lighting, layout, and logo placement. It also supports image editing at resolutions up to 4 megapixels while maintaining clarity and coherence.
From FLUX.1 to FLUX.2
Where FLUX.1 demonstrated the potential of media models as creative tools—offering precision, control, efficiency, and realism—FLUX.2 shows how frontier-level capability can reshape entire production pipelines. By dramatically improving the economics of generation, FLUX.2 is poised to become a foundational component of modern creative workflows.
What’s New
- Multi-Reference Support
- Use up to 10 reference images at once, achieving the strongest character, product, and style consistency available today.
- Improved Detail & Photorealism
- Sharper textures, richer details, and more stable, realistic lighting suitable for product photography, visualization, and high-fidelity imagery.
- Advanced Text Rendering
- Complex typography, infographics, memes, and UI mockups now produce clear, readable fine text that works reliably in real production scenarios.
- Stronger Prompt Alignment
- Better handling of complex, structured instructions—including multi-step prompts and precise compositional rules.
- Expanded World Knowledge
- More grounded understanding of real-world behavior, lighting, and spatial logic, leading to coherent, believable scenes.
- Higher Resolution & Flexible I/O
- Supports image editing at resolutions up to 4 megapixels, with more freedom in aspect ratios for both input and output.
Available Now
The FLUX.2 lineup spans a full range of model offerings—from fully managed, production-ready APIs to open-weight checkpoints that developers can run locally. The comparison chart below illustrates how FLUX.2 [pro], FLUX.2 [flex], FLUX.2 [dev], and FLUX.2 [klein] differ in terms of performance and control.
FLUX.2 [pro]
Delivers state-of-the-art image quality that competes with top closed-source systems, combining strong prompt adherence with exceptional visual fidelity. It produces images faster and at lower cost, offering speed and quality without compromise.
→ Available now via the BFL Playground, the BFL API, and our launch partners.
FLUX.2 [flex]
Provides granular control over generation parameters—such as step count and guidance scale—letting developers fine-tune quality, accuracy, and speed. It is especially strong at producing crisp text and intricate details.
→ Available now at bfl.ai/play, through the BFL API, and via launch partners.
FLUX.2 [dev]
A 32B open-weight model built from the core FLUX.2 architecture. It is the most capable open-weight model for image generation and editing today, supporting text-to-image, multi-image reference input, and editing within a single checkpoint. The weights are published on Hugging Face and can be run locally using our reference inference code. On consumer-grade GPUs (e.g., GeForce RTX), an optimized fp8 implementation—developed with NVIDIA and ComfyUI—allows efficient local use.
You can also access FLUX.2 [dev] through APIs on FAL, Replicate, Runware, Verda, TogetherAI, Cloudflare, and DeepInfra.
For commercial licensing, visit our website.
FLUX.2 [klein] (coming soon)
An Apache 2.0 open-source model distilled from the FLUX.2 base. It offers stronger performance and a more developer-friendly design compared to similarly sized models trained from scratch, while retaining many of the abilities of its teacher model.
→ Sign up for the beta.
FLUX.2 – VAE
A new variational autoencoder engineered to balance learnability, visual quality, and compression efficiency. It serves as the foundation for all FLUX.2 flow backbones. A detailed technical report is available, and the FLUX.2 VAE is released on Hugging Face under the Apache 2.0 license.