flux2_klein_4b_training_adapter

FLUX.2-klein-4B Training Adapter

Copyright

This model, kigland/flux2_klein_4b_training_adapter, is a model trained by KigML (KIGLAND Machine Learning Group). It is released under the Apache License 2.0.
You may use, modify, and distribute this model in accordance with the terms of the license.
Users must also comply with the base model's license (e.g., FLUX.2 Klein 4B) when using this adapter.

Description

Following content are mainly derived from ostris/zimage_turbo_training_adapter.

This is a training adapter designed to be used for fine-tuning FLUX.2-klein-4B. It can also be used as a general de-distillation LoRA for inference to remove the "Turbo" from "FLUX.2-klein-4B" (degenerate to FLUX.2-klein-base-4B).

Why is it needed?

When you train directly on a step distilled model, the distillation breaks down very quickly. This results in losing the step distillation in an unpredictable way. A de-distill training adapter slows this process down significantly allowing you to do short training runs while preserving the step distillation (speed).

What is the catch?

This is really just a hack to significantly slow down the distillation when fine-tuning a distilled model. The distillation will still be broken down over time. What that means is, this adapter will work great for shorter runs such as styles, concepts, and characters. However, doing a long training run will likely lead to the distillation breaking down to a point where artifacts will be produced when the adapter is removed.

How was it made?

We sampled a 5,000 (text-to-image generation) and 5,000 (text-to-text edition) dataset via WINDop/OpenGPT-4o-Image (arxiv:2509.24900). Then we trained a LoRA on those images at a low learning rate (1e-5). This allowed the distillation to break down while preserving the model's existing knowledge.

How does it work?

Since this adapter has broken down the distillation, if you train a LoRA on top of it, the distillation will no longer break down in your new LoRA, since this adapter has de-distilled the model. Your LoRA will now only learn the subject you are training. When it comes time to run inference / sampling, we remove this training adapter which leaves your new information on the distilled model allowing your new information to run at distilled speeds. Attached, is an example of a short training run on a character with and without this adapter.

Credits

• ostris/zimage_turbo_training_adapter
• WINDop/OpenGPT-4o-Image
• black-forest-labs/FLUX.2-klein-4B
• DiffSynth-Studio: Training Magic for Z-Image Turbo LoRA: How to Keep Turbo Image-Generation Ability (Content in Mandarin/Chinese)