Flux-2-Klein
Model Information
Model Version
Black Forest Klein series models. This collection includes the 4 most commonly used ones:
- Flux-2-Klein-9B
- Flux-2-Klein-9B-FP8
- Flux-2-Klein-4B
- Flux-2-Klein-4B-FP8
Just choose the model version based on the cover image . On the RunningHub platform, you don't need to worry about compute power. If your points are sufficient, it is recommended to use 9B, otherwise use 4B.
If there are more commonly used models in the same series, they will be updated here simultaneously.
Parameter Suggestions
CFG Value recommendation:
Keep it at 1.0. If adjustment is needed, the range is 1.0~2.0 (preferably within 2.0). Going higher than this range will cause image corruption in the generated results.
Steps (Sampling Steps) recommendation:
- For the 9B model, recommended steps are 15~25
- For the 4B model, recommended steps are 8~20
Of course, both models can use the 4-step ultra-fast sampling (the official reference uses 4 steps), and the results are also very good. If the result is not ideal, then use the recommended steps above.
Model Selection
Regarding why the base model is not included:
There is a difference in image generation between the Klein base model and the 9B and 4B distilled models in the collection.
The base model is mostly used for training, and its image generation quality and efficiency are not as good as the distilled models. Therefore, among the Klein series models, the distilled models in the collection are the most commonly used for daily tasks.
If you are using it locally and have insufficient compute power, it is recommended to choose the 4B model.
(PS: You definitely need to download the models yourself; the download links for the models are at the end)
Official link to the Black-Forest-Labs repository: https://huggingface.co/black-forest-labs
Model Version
Black Forest Klein series models. This collection includes the 4 most commonly used ones:
- Flux-2-Klein-9B
- Flux-2-Klein-9B-FP8
- Flux-2-Klein-4B
- Flux-2-Klein-4B-FP8
Just choose the model version based on the cover image . On the RunningHub platform, you don't need to worry about compute power. If your points are sufficient, it is recommended to use 9B, otherwise use 4B.
If there are more commonly used models in the same series, they will be updated here simultaneously.
Parameter Suggestions
CFG Value recommendation:
Keep it at 1.0. If adjustment is needed, the range is 1.0~2.0 (preferably within 2.0). Going higher than this range will cause image corruption in the generated results.
Steps (Sampling Steps) recommendation:
- For the 9B model, recommended steps are 15~25
- For the 4B model, recommended steps are 8~20
Of course, both models can use the 4-step ultra-fast sampling (the official reference uses 4 steps), and the results are also very good. If the result is not ideal, then use the recommended steps above.
Model Selection
Regarding why the base model is not included:
There is a difference in image generation between the Klein base model and the 9B and 4B distilled models in the collection.
The base model is mostly used for training, and its image generation quality and efficiency are not as good as the distilled models. Therefore, among the Klein series models, the distilled models in the collection are the most commonly used for daily tasks.
If you are using it locally and have insufficient compute power, it is recommended to choose the 4B model.
(PS: You definitely need to download the models yourself; the download links for the models are at the end)
Official link to the Black-Forest-Labs repository: https://huggingface.co/black-forest-labs