Instructions to use unsloth/FLUX.2-klein-9B-ComfyUI with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use unsloth/FLUX.2-klein-9B-ComfyUI with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
| license: apache-2.0 | |
| license_link: https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE | |
| base_model: | |
| - Qwen/Qwen3-8B | |
| tags: | |
| - text-encoder | |
| - diffusion-single-file | |
| - comfyui | |
| - flux | |
| # FLUX.2 klein 9B conditioner (Qwen3-8B), single file (mirror) | |
| The single-file Qwen3-8B conditioner that FLUX.2 klein 9B needs under | |
| [stable-diffusion.cpp](https://github.com/leejet/stable-diffusion.cpp), mirrored from | |
| [Comfy-Org/vae-text-encorder-for-flux-klein-9b](https://huggingface.co/Comfy-Org/vae-text-encorder-for-flux-klein-9b). | |
| klein 9B pairs with Qwen3-8B while klein 4B pairs with Qwen3-4B, and handing a klein 9B | |
| checkpoint the 4B encoder fails deep inside sd-cli, so Unsloth Studio picks this file per variant | |
| on the no-GPU path. It used to come straight from the community repack, which carries no model | |
| card and no licence metadata at all; a rename or a takedown there turns every CPU klein-9B load | |
| into a 404, so it is mirrored here. **The weights are unmodified:** byte for byte the file of the | |
| same name in the source repo. | |
| | File | Size | Upstream model | Licence | | |
| |---|---|---|---| | |
| | `split_files/text_encoders/qwen_3_8b.safetensors` | 15.26 GiB | [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) | Apache-2.0 | | |
| Only this file is mirrored. The source repo also ships a FLUX.2 VAE and quantised conditioners; | |
| Studio does not use them from there, so they are not here. | |
| ## Licence: Apache-2.0, and here is the evidence | |
| This file ships inside `black-forest-labs/FLUX.2-klein-9B`, which is under the FLUX | |
| Non-Commercial License, so the obvious move would be to label it non-commercial too. That would | |
| be wrong: **the weights are stock Qwen3-8B**, with nothing of Black Forest Labs's in them. | |
| All 399 tensors were compared against `Qwen/Qwen3-8B` by byte hash of the raw tensor buffer, key | |
| for key: | |
| ``` | |
| BIT-IDENTICAL 399 DIFFERENT 0 | |
| ``` | |
| Same key names, same dtypes, same shapes, no key in one file missing from the other. The FLUX | |
| Non-Commercial License itself provides at section 9 that third-party components stay subject to | |
| their own licensors' terms, so Alibaba's Apache-2.0 grant is what governs these bytes. Full text | |
| in [`LICENSE`](https://huggingface.co/unsloth/FLUX.2-klein-9B-ComfyUI/blob/main/LICENSE), | |
| attribution and the statement of changes in | |
| [`NOTICE`](https://huggingface.co/unsloth/FLUX.2-klein-9B-ComfyUI/blob/main/NOTICE). | |
| **This says nothing about FLUX.2 klein 9B itself.** That model is under the FLUX Non-Commercial | |
| License, and running it is governed by those terms no matter what licence its conditioner carries. | |
| Not an official Alibaba Cloud, Black Forest Labs or ComfyUI product, and not endorsed by any of | |
| them. | |