Black Forest Labs: FLUX 3 Image Klein API

FLUX 3 Image Klein is an open-weights image generation and editing model from Black Forest Labs. Its image transformer has 5.6 billion parameters and pairs with Qwen3-VL-4B-Instruct as a separate text encoder. The model accepts text and image inputs and produces image outputs. It supports text-to-image generation, single-reference editing, multi-reference composition with up to 10 reference images, prompt-based bounding-box layout control on a 0–1000 grid coordinate system, and text rendering within generated images. A prompt upsampler is required, as short prompts fed directly tend to produce generic results. The model ships in two variants. Klein Base is intended for fine-tuning, LoRA training, and research, using 50-step sampling with a guidance scale of 3.5. Klein distilled is built for direct generation and editing with 16-step sampling and no guidance, and is the checkpoint behind the Black Forest Labs API. Weights occupy about 11.2GB in BF16 and about 5.6GB in FP8. An INT8 quantized version runs on NVIDIA GeForce RTX GPUs with as little as 8GB of VRAM, such as the RTX 4070. With the distilled model and INT8 build, generating a 1MP image takes about 4 seconds on an RTX 5090 and about 9 seconds on an RTX 4070. FLUX 3 Image Klein is well suited for image generation on local or owned infrastructure, image editing on consumer devices, fine-tuning and LoRA training on the base checkpoint, and rapid iteration on thumbnails and design drafts. Integration is available through ComfyUI or Diffusers, with official workflow files provided. The model is released under the FLUX Kommunity License v1.0, which permits non-commercial use and offers a commercial output exception for eligible users. Deployment is also possible via the NVIDIA NIM platform, supported globally. Downloadable base checkpoints are provided for fine-tuning research. Before release, the model underwent internal red-team testing and evaluation by external safety partner Cinder.

Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices. Billed based on the actual generation cost. See your bill for the final amount.

It accepts text and image inputs, and produces image outputs.

It accepts up to 10 reference images for multi-reference composition.

Klein Base is designed for fine-tuning, LoRA training, and research, using 50-step sampling with a guidance scale of 3.5. The distilled version is for direct generation and editing, using 16-step sampling without guidance, and is the checkpoint behind the Black Forest Labs API.

The INT8 quantized version can run on NVIDIA GeForce RTX GPUs with as little as 8GB of VRAM, such as the RTX 4070.

It can be integrated through ComfyUI or Diffusers, and official workflow files are provided.

Yes, the model must be used with a prompt upsampler, as short prompts fed directly produce generic outputs.