# Baseline concept training [Contents](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/README.md) · [Next: Training the assistant: v1 and v2](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/experiments/training-assistants.md) At 500 updates, ordinary 512px training produced a recognisable Domokun in the beach scene, with unusual geometry. At 1,000 updates, both original scene prompts returned people and the unrelated fox acquired a large toothed mouth. | Setting | Value | | --- | --- | | Training | Domokun only, 512px, 1,000 updates | | Learning rate | `1e-4` | | Assistant | None | ## Training progression 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
Checkpoint / methodBeach (original prompt)Scarf (original prompt)FoxPortrait
BaseBase, 512pxBase, 512pxBase, 512pxBase, 512px
Update 500Update 500, 512pxUpdate 500, 512pxUpdate 500, 512pxUpdate 500, 512px
Update 1,000Update 1,000, 512pxUpdate 1,000, 512pxUpdate 1,000, 512pxUpdate 1,000, 512px
## Prompt and guidance comparison With the character noun and CFG 4, the final checkpoint produces Domokun-like imagery with facial-placement and shape errors; the fox is conventional. 512px output · 40 inference steps · true two-pass CFG 4 · CUDA-generator seeds shown below · Details-fixed VAE (madebyollin) · full-frame decoding. Seed: 123. ### Final checkpoint, CFG 4, seed 123
Checkpoint / methodBeach (original prompt)BeachRed scarfFoxPortrait
Basedomo_beachcharacter_beachcharacter_scarfprior_foxprior_portrait
Unassisted 512, update 1,000domo_beachcharacter_beachcharacter_scarfprior_foxprior_portrait
## Reproduce [Training configuration](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/recipes/unassisted-512/config.json) · [Datasets](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/recipes/unassisted-512/dataloader.json) · [Prompts](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/recipes/unassisted-512/prompts.json) [500 updates](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/checkpoints/unassisted-512/step-500/pytorch_lora_weights.safetensors) · [1,000 updates](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/resolve/main/checkpoints/unassisted-512/step-1000/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 Prompt wording and CFG change together in the second comparison, so their individual effects remain unresolved. [Contents](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/README.md) · [Next: Training the assistant: v1 and v2](https://huggingface.co/SimpleTuner/Qwen-Image-2.1-LoRA-experiments/blob/main/experiments/training-assistants.md)