FLUX.2-dev
- Name in the API
black-forest-labs/FLUX.2-dev- Base model
- black-forest-labs/FLUX.2-dev
- Weights we run
- The original weights, quantized to FP8 when they are loaded
- Type
- Image generation and editing
- Input → output
- text, reference images → image
- Availability
- Dedicated
- Hardware
- One NVIDIA H100 (94 GB)
- Limits
- Image size: at most 4,194,304 pixels (width × height), for example 2048×2048.
- Editing with more than one reference image: at most 1024×1024 pixels.
- Images per request (
n): only 1. Send one request per image.
- Steps (
num_inference_steps): at most 50, default 28.
- Endpoint
/v1/images/generations, /v1/images/edits- Example
- Generate an image, edit an image. Set
model to the name above.
Notes
- FLUX.2-dev is larger and slower than FLUX.2-klein-4B.
The two models are fallbacks for each other.