Tier 2 Restricted open-weight
What the publisher claims: open weights, open-weights
The one-liner: Top-of-chart open image weights with a revenue threshold written into the license.
The six questions
Base and distilled weights, quantized variants, and the text encoder are downloadable from the publisher's repository.
The FLUX Kommunity License grants non-commercial use and lets only sub-$5M-revenue Qualifying Users generate commercial output; other commercial use requires a separate license.
Runs on consumer hardware — 11.2GB in BF16, about 8GB with the INT8 build — through the reference code, Diffusers, or ComfyUI.
The release describes safety filters, post-training, and evaluation, but publishes no training corpus or recipe.
Limited: nothing published supports rebuilding an equivalent model.
BFL accepts outside contributions on the inference repos while the flagship remains API-only.
License
FLUX Kommunity License v1.0 — license source ↗
- Non-commercial use only, except that Qualifying Users may generate commercial output if they have less than US$5M in gross annualized revenue
- Acceptable Use Policy and trademark conditions apply
- The Qwen3-VL-4B-Instruct text encoder inside the package is separately Apache-2.0
What is actually released
- base checkpoint for fine-tuning
- step-distilled checkpoint
- BF16, FP8, INT8 and NVFP4 builds
- Qwen3-VL-4B-Instruct text encoder
- image autoencoder
- reference inference code
- prompt upsampler
Why it matters
The clearest current example of 'open weights' plus a hard commercial ceiling: the license names a dollar figure, so the openness question becomes a business-model question.
Sources
- https://bfl.ai/blog/flux-3-image-klein
- https://huggingface.co/black-forest-labs/flux-3-image-klein-base
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.