Unsloth released a GGUF-quantized version of FLUX.2-klein-4B. It is an image-to-image model with 4 billion parameters. The model supports text-to-image and multi-reference editing in one architecture. It runs on consumer GPUs with as little as 13GB VRAM. Inference can complete in under one second. The model is open-weights under Apache 2.0 for commercial use. It uses Unsloth Dynamic 2.0 for state-of-the-art performance. ComfyUI-GGUF tooling is integrated for workflows. The base model comes from Black Forest Labs. The quantized version is available on Hugging Face.
What changed
FLUX.2-klein-4B was available only in its original, larger format.
What it unlocks
Running a 4B-parameter image model on consumer GPUs with sub-second generation and editing.
- 4B parameters
- 13GB VRAM minimum
- 431,507 downloads last month
- under 1 second inference
What you need to act on it
- a GPU with at least 13GB VRAM
- use of GGUF quantization
Sources