Catalog
FLUX.2-klein-4B: a smaller FLUX model for image gen and editing
Black Forest Labs' 4-billion-parameter model handles both text-to-image and image editing in a lighter package, which makes it worth a look if the full FLUX models have been too heavy for your setup.
What it is. A 4-billion-parameter image generation and editing model from Black Forest Labs, the team behind the FLUX line. It handles text-to-image and image-to-image tasks and ships in both diffusers and safetensors formats under an Apache 2.0 license.
Why it matters. The full FLUX models do excellent work but demand real hardware. Klein is Black Forest Labs’ attempt at a more accessible point on that same curve, something you can run on a capable but not extravagant GPU without giving up the model family’s core strengths.
What it improved on. The earlier FLUX releases were not designed with a lighter footprint in mind. Klein is a purpose-built smaller variant, which means the tradeoffs were made deliberately rather than just being a compressed afterthought.
Strengths. Lower parameter count means faster inference and less VRAM pressure. The Apache 2.0 license is permissive, so it fits into commercial workflows without the restrictions that come with some other open models. It slots into existing diffusers pipelines if you are already working in that ecosystem, and single-file safetensors support means it loads cleanly in tools like ComfyUI.
Weaknesses. At 4 billion parameters it will not match the output quality of the full FLUX models, and the metadata from the model card is sparse enough that it is hard to say exactly where the quality gap lands until you run it yourself. Image editing at this scale also tends to be less precise than dedicated inpainting or editing tools, so complex localized edits may disappoint.
More Models
New A fine-tuned LoRA on FLUX.2-klein that takes a black-and-white manga page and a reference image and colors the line art to match the palette and style of that reference.
LoRA for coloring manga pages by reference image
JUN 10
New Roblox released open weights for a diffusion model that turns text into 3D meshes, which puts game-ready asset generation in reach for anyone with the hardware to run it.
Roblox's open diffusion model for text-to-3D mesh
MAY 27
Nvidia's Lyra 2.0 takes one photo and hands back a 3D asset, which cuts out a lot of the manual work that usually sits between a reference image and something you can actually use in a scene.