gguf-flux2-klein-4b

FLUX.2 [klein] 4B diffusion weights in GGUF format for stable-diffusion.cpp, mirrored byte-identical from leejet/FLUX.2-klein-4B-GGUF for JoyFusion's on-device (Android) inference.

File	Quant	Size	SHA-256
flux-2-klein-4b-Q4_0.gguf	Q4_0	2,460,378,560	d1023499ef3f2f82ff7c50e6778495195c1b6cc34835741778868428111f9ff4
flux-2-klein-4b-Q8_0.gguf	Q8_0	4,300,629,440	0bba6951258ec8f92d51114a8fa13e66828297bfff58a738f52729b3ef66fa28

Shared components live in JoyFusionAI/gguf-components: text encoder Qwen3-4B-{Q4_0,Q8_0}.gguf (base Qwen3-4B, not the Instruct-2507 variant) and VAE flux2-vae.safetensors.

Recommended: 4 steps, CFG 1.0, Euler, --diffusion-fa. Supports text-to-image and reference-image editing.

License

FLUX.2 [klein] 4B, including its text encoder and VAE, is released by Black Forest Labs under the Apache License 2.0 (see LICENSE.md). The GGUF conversion was done by the stable-diffusion.cpp project; the files here are unmodified copies.