FLUX.2 [klein] Base 4B — MLX 4-bit transformer (Pixai add-on)

The undistilled Base transformer of black-forest-labs/FLUX.2-klein-base-4B, 4-bit MLX (group size 64), byte-identical to mflux-community/flux-2-klein-base-4b-mflux-q4 at revision f69aa0b43285d0db24695ba68faabbdb1a0ad016. Not an official Black Forest Labs release.

This is an add-on. It contains only transformer/ (2.18 GB). The Qwen3-4B text encoder, VAE and tokenizer are identical to FLUX.2 klein 4B and come from nForce101/flux2-klein-4b-q4.

Use it for slower, higher-quality runs: about 28–50 steps with classifier-free guidance (≈4) and a negative prompt, i.e. two transformer passes per step. The distilled klein remains the fast 4–12-step option.

pixai-model.json lists every file with its size and SHA-256.

License

Apache License 2.0 — see LICENSE.md and NOTICE.

Downloads last month: -; Downloads are not tracked for this model. How to track

MLX

Hardware compatibility

Quantized

Model tree for nForce101/flux2-klein-base-4b-q4

Base model

black-forest-labs/FLUX.2-klein-base-4B

Quantized

(16)

this model