Black Forest Labs Open-Sources FLUX.3 Image Klein

Around October 15, Black Forest Labs open-sourced the image model FLUX.3 Image Klein—5.6B parameters, open weights—claiming frontier-model quality while running on a single consumer GPU (as low as 8GB VRAM), with native 4K output, bounding box control, and pixel-precise editing. The company says that in internal evals, it scores above current open-weight leader Qwen Image 2.1 on text-to-image and multi-reference generation at lower cost, sitting on the quality-speed Pareto frontier.

Confirmed

• Unified architecture: one model handles text-to-image, single-reference editing, and multi-reference composition (up to 10 reference images), with JSON bounding box layout control.
• Release contents: Base full undistilled weights (suited for fine-tuning, LoRA training, and research) plus a Distilled version (16 steps, low latency), with FP8, INT8, and NVFP4 quantized variants.
• Ecosystem support: weights are on Hugging Face (black-forest-labs/flux-3-image-klein) with Day-0 support from the Diffusers team (@RisingSayak, @multimodalart, etc.), with modular implementations of both distilled and undistilled pipelines; native ComfyUI support, plus a single-file flux-3-image-klein-comfy release in diffusion-single-file format (BF16, int8convro… and other specs) for direct use.
• Performance: trained on NVIDIA GB200 NVL72 and co-optimized with NVIDIA; 0.55s per image on GB300 (NVFP4), 4s on RTX 5090 (INT8); official per-GPU timing tables are published, along with enterprise licensing info.
• Community traction: the model and its Gradio Space hit Hugging Face trending; the Space carries an mcp-server tag, suggesting it may be callable via MCP.

Why it matters

• The open-source camp has been led by the Qwen Image series; FLUX.3 Klein's "frontier quality on consumer hardware" positioning directly challenges that, with internal scores beating Qwen Image 2.1.
• Native 4K, multi-reference composition, and pixel-level editing pack capabilities that previously required multiple models and pipelines into one 5.6B model, lowering the barrier to local deployment and fine-tuning, while Day-0 Diffusers/ComfyUI support lets developers start immediately.

2026-10-14 ~ 2026-10-16 · 34 related posts

• Black Forest Labs ships ComfyUI single-file release of FLUX.3 Klein image model — black-forest-labs · 2026-10-14
• [source] BFL launches FLUX.3 Image Klein: open-weights 5.6B model with native 4K on one consumer GPU — bfl_ai · 2026-10-15
• [source] FLUX 3 Image Klein ships open weights, distilled and quantized versions runnable on 8GB VRAM laptops — bfl_ai · 2026-10-15
• Black Forest Labs launches FLUX 3 Image Klein, a 6B model that outscores Qwen Image 2.1 in internal evals — bfl_ai · 2026-10-15
• BFL publishes per-GPU benchmarks for FLUX 3 Klein, from 0.59s on DGX Station to 50s on RTX 4070 — bfl_ai · 2026-10-15
• Black Forest Labs open-sources FLUX 3 Image Klein: 5.6B weights live on Hugging Face — bfl_ai · 2026-10-15
• Black Forest Labs open-weights FLUX 3 Image Klein: 4K generation and pixel-perfect editing on consumer GPUs — hyxon4 · 2026-10-15
• FLUX.3 Image Klein 6B lands on Hugging Face with base and distilled checkpoints — multimodalart · 2026-10-15
• Black Forest Labs finally releases model weights for download — MaziyarPanahi · 2026-10-15
• FLUX 3 Image Klein ships open weights: 5.6B params, native 4K, runs on a single consumer GPU — robrombach · 2026-10-15
• FLUX.3 Image Klein ships open in Diffusers: layout control, 2s per image on H100 — ariG23498 · 2026-10-15
• FLUX.3 Klein lands on Hugging Face with Day-0 Diffusers support — Disastrous-Work-1632 · 2026-10-15
• Flux 3 Image Klein lands on Hugging Face with ComfyUI support via Kijai's commit — TheSlateGray · 2026-10-15
• Diffusers adds full Flux.3 Image Klein support: FP8/NVFP4 quantized inference and bbox layout control — RisingSayak · 2026-10-15
• Diffusers Adds Native LoRA Training Scripts and FP8/NVFP4 Quantized Inference — RisingSayak · 2026-10-15
• FLUX 3 Image launches with pixel-perfect edits and open 5.6B Klein weights on HF — ariG23498 · 2026-10-16
• Flux 3 Klein open weights drop: 5.6B params, native 4K, day-0 diffusers support — linoy_tsaban · 2026-10-16
• FLUX 3 Image Klein launches at 5.6B params, runs on most RTX cards via quants — robrombach · 2026-10-16
• Flux.3 Image Klein launches with day-one full support in Diffusers — MaziyarPanahi · 2026-10-16
• FLUX 3 Image Klein (5.6B, open weights) lands in ComfyUI with native 4K — crystal_alpine · 2026-10-16
• FLUX.3 Image Klein weights published on Hugging Face — wapswaps · 2026-10-16
• FLUX 3 image-klein trends #1 on Hugging Face as image-to-image model — black-forest-labs · 2026-10-16
• Black Forest Labs' FLUX 3 image model Klein trends on Hugging Face — black-forest-labs · 2026-10-16

11 near-duplicate retellings: bfl_ai · bfl_ai · pbaylies · ariG23498 · linoy_tsaban · bfl_ai · JGByvygyrfg · testingcatalog · lmoroney · lmoroney · bfl_ai