CPU3DAI tools for 3D, video and audio

Black Forest Labs releases flux-3-image-klein on HuggingFace

On October 13, 2026, the model black-forest-labs/flux-3-image-klein appeared on HuggingFace with the image-to-image task. In the first days, the model gained 3,241 downloads and 271 likes. The code has been released as open source.

What it means

The model contains 5.5 billion parameters. The largest weights file — model.safetensors — takes up 10.3 GB, while the entire repository weighs 40.3 GB. All weights are published in the safetensors format, and it is expected to be run via diffusers. Running it locally will require a GPU with enough VRAM for the main weights file of 10.3 GB — this is without accounting for the overhead of the image-to-image pipeline itself. The full 40.3 GB repository apparently includes several weight variants, of which not the entire volume is needed for operation. The developer did not specify a license on the model page. In the same days, flux-3-image-klein-base and flux-3-image-klein-pup were released in the lineup. There are no details about them on the main model page. You can follow the release of new tools in the CPU3D news feed.
flux-3-image-klein on HuggingFace Input image image-to-image task flux-3-image-klein black-forest-labs 5.5 billion parameters running via diffusers Output image generation result model.safetensors 10.3 GB Entire repository 40.3 GB, safetensors GPU VRAM at least 10.3 GB Overhead image-to-image pipeline 3241 downloads in the first days 271 likes "likes" Also released in the lineup: flux-3-image-klein-base · flux-3-image-klein-pup License not specified by the developer
How the method works. The diagram is drawn based on this news note.

See also