Black Forest Labs open-sources FLUX.3 Image Klein: 2s per image on H100, layout control
ariG23498 · x · 2026-10-17
Black Forest Labs has released FLUX.3 Image Klein, its latest open image generation model, supported in Hugging Face Diffusers from day one. A single model handles text-to-image, multi-reference image-to-image, and new bounding-box layout control — you write boxes into the prompt to place elements or move parts of a reference image.
Key facts:
- Two open checkpoints: distilled flux-3-image-klein (16 steps, no CFG, 2s for 1024×1024 on an H100) and base flux-3-image-klein-base (50 steps, with CFG)
- NVFP4 and FP8 weights load via fromsinglefile for smaller GPUs
- LoRA training scripts included for text-to-image and image editing
- Ships as a Modular Diffusers pipeline, so blocks can be popped out or replaced
- Uses Qwen3-VL as text encoder; a live HF Space is available
More from Multimodal
- Lava lamp test: Gemini 4 nails it for $0.18 in 2 min vs Opus 5.5's $1.27 and 9.5 min — CodeByPoonam · 2026-10-17
- Nuwa.skill (34.8k stars) releases promo animation fully generated by Opus 5.5 — AlchainHust · 2026-10-17
- Midjourney v8.2 Style Code sref 2563363531 with Full Parameter Combo Shared — azed_ai · 2026-10-17
- T2I head-to-head: Moss vs Flux3 Klein vs Krea2, same prompt and seed — CutLongjumping8 · 2026-10-17
- Midjourney style code --sref 1181489133 nails black-and-white slow-shutter aesthetics — umesh_ai · 2026-10-17
- Two-step animation workflow: Opus 5.5 for storyboards, Seedance 2.5 for final video — ring_hyacinth · 2026-10-17