FLUX 3 Image Klein(bfl.ai)
213 points by gmays 9 days ago | 58 comments
tl;dr: Black Forest Labs released FLUX 3 Image Klein, a 5.6B parameter open-weights text-to-image model supporting editing and multi-reference generation (up to 10 images) with 4K output and bounding-box controls. Despite being smaller than competitors like Qwen-Image 2.1 (7B) and Krea 2 (12.9B), it tops internal benchmarks and runs on consumer GPUs with as little as 8GB VRAM, generating 1MP images in ~4 seconds on an RTX 5090. It's available under the FLUX Kommunity License, which permits free commercial use for companies under $5M in revenue.
HN Discussion:
  • ↓Model struggles with generating people due to NSFW content removal from training data
  • ↓Censorship and over-restriction of image models is inefficient and societally regressive
  • ~Editing is solid but text details and body horror issues make competitors like Moss Image more appealing
  • ↑Amazement at how much world knowledge is compressed into such a small model
  • ↑Pricing is highly competitive compared to alternatives like Krea 3