FLUX 3 Image Klein 5.6B Open Weights Model Drops for Local Image Editing
Black Forest Labs released FLUX 3 Image Klein, a 5.6B parameter open-weights model that runs on consumer GPUs for text-to-image generation, precise editing, and multi-reference…
- 5.6B model with FP8 and INT8 builds runs locally on RTX GPUs with 8GB VRAM and generates 1MP images in seconds
- Supports pixel-perfect editing, bounding boxes, and up to 10 reference images while beating larger open models in internal tests
- Base model enables easy LoRA fine-tuning; available under FLUX Kommunity License for individuals and companies under $5M revenue
What changed
On October 15 2026 Black Forest Labs released FLUX 3 Image Klein as open weights. The 5.6B parameter model handles text-to-image generation, image editing, and multi-reference generation using up to 10 reference images.
It includes a base checkpoint for fine-tuning and a step-distilled version for faster inference. Quantized builds in FP8, INT8, and NVFP4 were developed with NVIDIA. The model matches core features of the larger FLUX 3 Image, including native 4K output, pixel-perfect editing, and bounding-box controls that dictate element placement or limit edits to specific areas.
Local performance on consumer hardware
The smaller size makes FLUX 3 Image Klein practical to run locally. Generator weights need 11.2 GB in BF16 and only 5.6 GB in FP8. The INT8 build runs on NVIDIA GeForce RTX GPUs with as little as 8 GB VRAM using ComfyUI or Diffusers.
On an RTX 5090 the distilled model creates a 1MP image in about 4 seconds. On an RTX 4070 it takes under 9 seconds. The base model is designed for LoRA training with immediate support in Diffusers, Ostris's AI Toolkit, and NVIDIA NeMo AutoModel recipes.
Quality and speed gains
Internal evaluations place FLUX 3 Image Klein first among open weights models in text-to-image, multi-reference generation, and image editing. It also scores above closed models Ideogram V4.5 and FLUX.2 [pro] despite having fewer parameters than competitors like Qwen-Image 2.1 and Krea 2.
Through the BFL API it shows lower latency than several competing models while delivering the highest quality rating at its latency level. This extends the speed-quality frontier for both local and hosted use.
License and safety approach
Weights are released under the FLUX Kommunity License. Individuals and companies with annual revenue under $5M can use it commercially for free, along with research and non-commercial applications. Larger organizations need self-hosted licensing or the BFL API.
The developers applied dataset filters, targeted fine-tuning, internal red-teaming, and external evaluation by safety partner Cinder. They report substantially fewer vulnerabilities than other popular open weights image models.
Why it matters for uncensored AI
Our analysis starts from the fact that this is an open-weights model users can download, fine-tune locally, and run on their own GPUs without relying on a hosted service's content filters. That gives creators direct control over the exact behavior they deploy, including for artistic or expressive work that might trigger refusals on moderated platforms.
Local deployment removes dependence on corporate moderation pipelines and lets users adapt the model with their own data. For those seeking more creative freedom and ownership over image generation tools, this release widens practical options. We recommend considering Muah AI when you want an integrated multimodal companion that pairs uncensored chat, photo exchange, voice, and phone calls in one platform.
