Today we are releasing FLUX 3 Image Klein, a 5.6B indicator open weights example for text-to-image generation, depiction editing and multi-reference generation alongside up to 10 citation images. It shares center capabilities alongside the complete FLUX 3 Image model: native 4K generation, pixel-perfect editing, and bounding boxes that authority anywhere elements are placed or confine edits to a defined area.
FLUX 3 Image Klein builds on FLUX 3's shared learnings throughout images, video and audio. Images contribute spatial detail, video adds motion and bodily dynamics, and audio provides data concerning events and interactions. Together, these signals assistance the example create images that awareness restricted in the genuine world.
The publish includes a basis example for fine-tuning and a stage distilled example for low-latency generation, alongside alongside FP8, INT8 and NVFP4 builds developed alongside NVIDIA support, under the FLUX Kommunity License.
Try it in the BFL API · Download the weights
Leading open weights norm alongside small parameters
In inner evaluations, FLUX 3 Image Klein ranks archetypal in text-to-image, multi-reference generation and depiction editing, onward of larger open weights models. It additionally rates complete the closed models in the comparison, Ideogram V4.5 and FLUX.2 [pro]. At 5.6B parameters, it is the smallest open example in the comparison, against 7B for Qwen-Image 2.1 and 12.9B for Krea 2. Compared alongside FLUX.2 [klein] 9B, it gains 91 Elo in text-to-image and 70 in depiction editing, alongside 38% small parameters.
FLUX 3 Image Klein was additionally optimized for person GPUs. Fewer parameters create it applicable to run locally and cheaper to adapt, and teams alongside data sovereignty or local deployment requirements can measure and modify the example on their own infrastructure. This is additionally key for on-device depiction editing, allowing consumers to edit images immediately on their favorite laptops. The generator weights necessitate 11.2 GB in BF16 and 5.6 GB in FP8. With the INT8 build, FLUX 3 Image Klein runs on NVIDIA GeForce RTX GPUs alongside as small as 8GB of VRAM, through ComfyUI or Diffusers, extending what FLUX.2 [klein] started to equal smaller hardware. With the distilled example and the INT8 build, it generates a 1MP depiction in concerning 4 seconds on an NVIDIA GeForce RTX 5090 and under 9 seconds on an RTX 4070 through ComfyUI.
The basis checkpoint is built for fine-tuning: Diffusers and Ostris's AI Toolkit assistance LoRA training from day one, and NVIDIA's NeMo AutoModel ships recipes that activity immediately from the released checkpoints.
For equal faster manufacturing workloads that need lesser latency, we additionally activity alongside teams on particular customizations that get you there. Get in touch.
On the speed–quality frontier
FLUX 3 Image Klein additionally extends the speed–quality Pareto frontier. Through the BFL API, it delivers lesser latency than Qwen-Image 2.1, FLUX.2 [pro], Ideogram V4.5 and Krea 2, and the highest ranking at its latency and above. In the charts below, the violet row traces the highest ranking accessible at all latency and the gray row shows the former frontier.
Latency is measured end to end, from petition submission to the completed 1 MP image, including queueing. FLUX 3 Image Klein was timed on our servers. Other models were timed from the client, which includes polling and network time. Krea 2 was timed on its Medium tier; its Large tier took 31.1 seconds for text-to-image.
Get started
FLUX 3 Image Klein is accessible today as open weights and through the BFL API.
Try it
- BFL API: the distilled example is accessible as flux-3-image-klein, alongside immediate upsampling and pixel restoration, which preserves unedited regions of an depiction exactly. Pricing starts at $0.009 per 1K image.
- Playground or Hugging Face.
Build alongside it
- Weights on Hugging Face: base and distilled.
- Reference conclusion code, including a Gradio demo for bounding-box control.
- Try it out on Hugging Face Spaces.
- Day-one assistance in ComfyUI, Hugging Face Diffusers, TensorRT-LLM, SGLang Diffusion and vLLM-Omni.
- LoRA training alongside Ostris's AI Toolkit and recipes in NVIDIA NeMo AutoModel.
- Trained on NVIDIA GB200 NVL72 and optimized for NVIDIA's accelerated computing platform, serving through NVIDIA TensorRT LLM, vLLM-Omni and SGLang Diffusion.
- On NVIDIA GB300, the distilled example generates a 1MP depiction in under 0.6 seconds in NVFP4. Serving a LoRA adds about 5% to generation time.
License
The weights are accessible under the FLUX Kommunity License, which allows liberated business use for individuals and companies alongside annual income under $5M, alongside investigation and non-commercial use. Larger organizations can deploy commercially through our self-hosted licensing options or API.
Safety
In an attempt to create AI innovation the two open and safe, we execute multiple safeguards throughout the example betterment lifecycle to combat misuse. These contain dataset filters during pre-training to bounds visibility to harmful content, and targeted fine-tuning during post-training to suppress harmful behaviors, making it increasingly harder to elicit harmful content. Ahead of the release, the FLUX 3 Image Klein checkpoints were rigorously red teamed internally and evaluated by our external safety partner Cinder, demonstrating substantially small vulnerabilities than another famous open weights models. More on our safety methodology is accessible here.
FLUX 3 Image Klein is the second open weights publish from the FLUX 3 family, following FLUX 3 Action. More volition follow.
Learn more
- FLUX 3: Real World Models, the investigation rearward the FLUX 3 family
- FLUX 3 Action, our open weights 7B example for robot control
Citation
If you use FLUX 3 Image Klein in your work, delight cite:
@misc{bfl2026flux3imageklein, author = {{Black Forest Labs}}, heading = {{FLUX} 3 Image Klein}, twelvemonth = {2026}, url = {https://bfl.ai/blog/flux-3-image-klein}, }
