Text-to-Image
MLX
Safetensors
English
apple-silicon
flux
flux-3
image-editing
image-to-image
quantized
Instructions to use VisualInference/FLUX-3-Image-Klein-MLX-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use VisualInference/FLUX-3-Image-Klein-MLX-8bit with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download VisualInference/FLUX-3-Image-Klein-MLX-8bit --local-dir FLUX-3-Image-Klein-MLX-8bit
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Card: pip install flux3-mlx (PyPI)
Browse files
README.md
CHANGED
|
@@ -29,7 +29,7 @@ The transformer is quantized to 8-bit; the output is validated against Black For
|
|
| 29 |
> An independent community conversion, not affiliated with or endorsed by Black Forest Labs.
|
| 30 |
|
| 31 |
```sh
|
| 32 |
-
pip install "flux3-mlx[pup]
|
| 33 |
flux3-mlx "a ceramicist's studio at golden hour, shelves of glazed bowls" --model 8bit --upsample --out studio.png
|
| 34 |
```
|
| 35 |
|
|
@@ -120,7 +120,9 @@ Quantization study across five cases (photo, typography, fashion, still life, ed
|
|
| 120 |
| `transformer/` | FLUX 3 Image Klein (distilled, flow map) DiT. Linear layers in the 43 transformer blocks are 8-bit (group size 64, affine). Modulation, time, input embeddings and final layer stay BF16. |
|
| 121 |
| `text_encoder/` | Text path of [Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) (byte-identical to BFL's bundled encoder), layers 0–31 only (Klein reads layers 4…32), 8-bit. Apache-2.0. |
|
| 122 |
| `vae/ae.safetensors` | The FLUX autoencoder, unchanged (fp32). |
|
|
|
|
| 123 |
| `LICENSE.md`, `NOTICE.md` | FLUX Kommunity License and Attribution Notice |
|
|
|
|
| 124 |
|
| 125 |
Source: `black-forest-labs/flux-3-image-klein` `model.safetensors` (sha256 `c3e9813c…`, matching BFL's manifest).
|
| 126 |
|
|
|
|
| 29 |
> An independent community conversion, not affiliated with or endorsed by Black Forest Labs.
|
| 30 |
|
| 31 |
```sh
|
| 32 |
+
pip install "flux3-mlx[pup]"
|
| 33 |
flux3-mlx "a ceramicist's studio at golden hour, shelves of glazed bowls" --model 8bit --upsample --out studio.png
|
| 34 |
```
|
| 35 |
|
|
|
|
| 120 |
| `transformer/` | FLUX 3 Image Klein (distilled, flow map) DiT. Linear layers in the 43 transformer blocks are 8-bit (group size 64, affine). Modulation, time, input embeddings and final layer stay BF16. |
|
| 121 |
| `text_encoder/` | Text path of [Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) (byte-identical to BFL's bundled encoder), layers 0–31 only (Klein reads layers 4…32), 8-bit. Apache-2.0. |
|
| 122 |
| `vae/ae.safetensors` | The FLUX autoencoder, unchanged (fp32). |
|
| 123 |
+
| `config.json` | Repo manifest for the `flux3-mlx` runner (format version, components, sampler defaults) |
|
| 124 |
| `LICENSE.md`, `NOTICE.md` | FLUX Kommunity License and Attribution Notice |
|
| 125 |
+
| `gallery/` | Example images with their prompts, seeds and timings |
|
| 126 |
|
| 127 |
Source: `black-forest-labs/flux-3-image-klein` `model.safetensors` (sha256 `c3e9813c…`, matching BFL's manifest).
|
| 128 |
|