# Learning-rate experiments

[Contents](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/README.md) · [Previous: Multi-scale training and REPA](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/experiments/multiscale-repa.md) · [Next: Longer training: regression and recovery](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/experiments/longer-training.md)

The assisted 512px run at `3e-4` severely degraded all four validation prompts by update 500. The unrelated fox and portrait became Domokun-like imagery.

| Setting | Value |
| --- | --- |
| Data / training resolution | Domokun only / 512px |
| Assistant | Frozen v1; disabled for validation |
| Learning rate / norm clipping | `3e-4` / 1.0 |
| Checkpoint | 500 updates |

CPU-generator seed 42 · 40 inference steps · CFG 1 · Details-fixed VAE (madebyollin) · full-frame decoding. Output: 512px. Columns show prompts; rows show checkpoints or methods. Click a sample to view it at full resolution.

### 512px output

<div style="overflow-x: auto;">
<table>
<thead><tr><th scope="col">Checkpoint / method</th><th scope="col">Beach (original prompt)</th><th scope="col">Scarf (original prompt)</th><th scope="col">Fox</th><th scope="col">Portrait</th></tr></thead>
<tbody>
<tr><th scope="row">Base</th><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/3f600b644723d5b7f98b4c7f.png"><img alt="Base, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/3f600b644723d5b7f98b4c7f.png" style="min-width: 160px; max-width: none;" width="160"/></a></td><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/99ef08c05b1dec71fb423658.png"><img alt="Base, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/99ef08c05b1dec71fb423658.png" style="min-width: 160px; max-width: none;" width="160"/></a></td><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/c150cc084b9dae047545cc64.png"><img alt="Base, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/c150cc084b9dae047545cc64.png" style="min-width: 160px; max-width: none;" width="160"/></a></td><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/998d84865ba20d52b1f69b71.png"><img alt="Base, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/998d84865ba20d52b1f69b71.png" style="min-width: 160px; max-width: none;" width="160"/></a></td></tr>
<tr><th scope="row">Concept, 500 updates</th><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/3c1f1007ed231e9b6f466cff.png"><img alt="Concept, 500 updates, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/3c1f1007ed231e9b6f466cff.png" style="min-width: 160px; max-width: none;" width="160"/></a></td><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/e8b897b76cf666cd6af1f625.png"><img alt="Concept, 500 updates, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/e8b897b76cf666cd6af1f625.png" style="min-width: 160px; max-width: none;" width="160"/></a></td><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/a89103de6565995196cd9cde.png"><img alt="Concept, 500 updates, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/a89103de6565995196cd9cde.png" style="min-width: 160px; max-width: none;" width="160"/></a></td><td><a href="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/7151e47276f75e04e3929d7f.png"><img alt="Concept, 500 updates, 512px" loading="lazy" src="https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/assets/fixed-vae/7151e47276f75e04e3929d7f.png" style="min-width: 160px; max-width: none;" width="160"/></a></td></tr>
</tbody>
</table>
</div>


## Reproduce

[Training configuration](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/recipes/assisted-v1-high-lr/config.json) · [Datasets](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/recipes/assisted-v1-high-lr/dataloader.json) · [Prompts](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/recipes/assisted-v1-high-lr/prompts.json) · [500 updates](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/checkpoints/assisted-v1-high-lr/step-500/pytorch_lora_weights.safetensors) · [Shared training and evaluation settings](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/reference/methods.md)

## Limitations

This separate run illustrates degradation at the higher learning rate; the available comparison does not isolate learning rate as its cause.

[Contents](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/README.md) · [Previous: Multi-scale training and REPA](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/experiments/multiscale-repa.md) · [Next: Longer training: regression and recovery](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/experiments/longer-training.md)