Flux.2 Dev
Developer-Grade Control
The 32-billion parameter open-weights model from Black Forest Labs. Built for developers and researchers who need full parameter control—adjust step count, guidance scale, and train custom LoRAs. Unifies image generation and editing in a single architecture with exceptional text rendering and multi-reference support.
Please log in to get free credits.
Reference Images (Optional, up to 4)
Prompt
What is Flux.2 Dev?
Flux.2 Dev is the developer-focused variant of Black Forest Labs' Flux.2 family. Combining 32 billion parameters with open weights, it gives you complete control over generation parameters while delivering production-quality results with improved prompt adherence and typography.
How to Use Flux.2 Dev
Developer-friendly workflow with parameter-level control
Add Reference Images
Upload up to 4 reference images for multi-reference generation. Maintain consistent characters, products, or artistic styles across outputs using advanced reference blending.
Craft Your Prompt
Write detailed text instructions. Dev excels at complex compositions, structured layouts, UI mockups, and legible fine text—perfect for infographic-style assets and typography-heavy designs.
Fine-Tune Parameters
Control resolution (up to 4MP), aspect ratio, and output format. Experiment with generation settings to balance quality, prompt alignment, and performance for your workflow.
Generate & Iterate
Generate production-ready images with repeatable parameters. Download in WebP, JPG, or PNG, then iterate with adjusted settings for systematic prompt testing.
Key Features of Flux.2 Dev
4MP High-Resolution Output
Generate images up to 4 megapixels with enhanced texture detail, realistic materials, and professional lighting. Ideal for print-ready assets, product photography, and detailed visual content.
Multi-Reference Generation
Combine up to 4 reference images with intelligent blending. Win rate of 59.8% in single-reference and 63.6% in multi-reference editing—leading among open-weight models for consistent character and style generation.
Advanced Text Rendering
Generate legible fine text, structured layouts, and UI elements with superior reliability. Reworked typography system handles infographics, mockups, and text-heavy compositions that challenge other models.
Unified Architecture
Single checkpoint for both generation and editing powered by Mistral 3 24B vision-language model plus rectified flow transformer. No separate pipelines—one model does it all.
Full Parameter Control
Adjust step count, guidance scale, and custom parameters for fine-grained quality balancing. Build repeatable workflows and systematic prompt testing for production environments.
FP8 Optimized Performance
40% less VRAM with FP8 quantization and 40% faster on RTX GPUs. Run on 18-24GB cards or offload to system RAM for 8GB configurations. Open weights enable local deployment.
Trusted by Developers & Researchers
"Full parameter control with open weights changed my research workflow. I can iterate systematically and reproduce results exactly—critical for academic work."
"Trained a custom LoRA for our brand style guide. Now we get consistent product renders every time. The text rendering for UI mockups is unmatched."
"Deployed locally on our infrastructure with complete data privacy. FP8 optimization makes it viable on our existing GPU fleet. Perfect for production experiments."
Frequently
Asked
Questions
What is Flux.2 Dev?
Flux.2 Dev is a 32-billion parameter open-weights image generation model from Black Forest Labs. Built for developers and researchers, it unifies text-to-image generation and image editing in a single checkpoint with full parameter control, LoRA training support, and exceptional prompt adherence.
How is Flux.2 Dev different from Flux.2 Pro?
Flux.2 Pro is the highest-performance closed-weights model available only via API, optimized for production with predictable latency. Flux.2 Dev offers open weights with full parameter control—adjust step count, guidance scale, and train custom LoRAs. Dev leads open-weight alternatives with 66.6% win rate in text-to-image generation. Choose Pro for production consistency, Dev for experimentation and customization.
Can I use Flux.2 Dev commercially?
Flux.2 Dev is released under a non-commercial license for personal, scientific, and research use. For commercial applications, you need to obtain a commercial license from Black Forest Labs. The open weights enable local deployment and complete data privacy for authorized use cases.
What are the system requirements?
The full model requires 80GB+ VRAM when running both the Mistral 3 24B text encoder and DiT together. FP8 quantization reduces VRAM by 40% and improves RTX GPU performance by 40%, making it viable on 18-24GB cards. With CPU offloading, you can run it on 8GB GPUs with sufficient system RAM.
Does Flux.2 Dev support LoRA training?
Yes! Flux.2 Dev fully supports custom LoRA fine-tuning for specialized styles, brand consistency, and domain-specific imagery. The open-weights architecture enables building repeatable workflows, systematic prompt testing, and training custom adaptations for production environments.
What resolution can Flux.2 Dev generate?
Flux.2 Dev generates images up to 4 megapixels (4MP) with enhanced texture detail, realistic materials, and professional lighting. Supports flexible aspect ratios from 1:1 to 21:9, with resolution ranges from 256-1440 pixels per dimension—ideal for both web and print applications.
How many reference images can I use?
Flux.2 Dev supports up to 4 reference images with intelligent blending. It achieves 59.8% win rate in single-reference editing and 63.6% in multi-reference editing, leading among open-weight models for maintaining character identity, product appearance, and style consistency across outputs.
Can Flux.2 Dev render text in images?
Yes! Flux.2 Dev features reworked text rendering that generates legible fine text, structured layouts, UI elements, and infographic-style assets with superior reliability. It excels at typography-heavy compositions, mockups, and complex visual layouts that challenge other generation models.