This DoRA was trained on 600k images of hyper sized anime characters. It focus mainly on breasts/ass/belly/thighs/fat. This dataset is a subset of the larger hyperfusion dataset, but filtered down to body shape/size related images only. The full dataset would have taken more than a year to train on SDXL, lol.Recomendations:<ul><li>Dora/LoRA strength: 1.0 (DoRA's work in most WebUI's by now)</li><li>Resolution: ~1024px</li><li>samplers: any sampler PonyXL supports</li><li>in v10 you can push the lora weight more than in v9, so do that if your concepts are not working as well as you like.</li></ul>Uploaded 1.4 million custom tags used in hyperfusion <a target="_blank" rel="ugc" href="https://huggingface.co/datasets/thojm/hyperfusion_classified_tags_export">here </a>for integrating into your own datasetsv10 Noob_vpred Release 2025/07/29:<ul><li>Did you guys think I disappeared? Nope, just hopelessly training a model with a frozen text encoder for 7 months.</li><li>This new DoRA has the same concepts you are used to by now, but with a few new concepts as usual. Also 200k more images than v9.</li><li>This version is trained on NoobAI_Vpred, so there is no guarantee it will work with anything else. Especially not on non-v_pred models.</li><li>Wanted to try training with the Text Encoder frozen one last time. Also decided to stick to it no matter how long it took. And now I can definitively say I will be including TE in future models just for the sake of time. It works, but its way too slow for my setup.</li><li>Use the tag list in v9 for now, until I get around to building the new one with the small number of new concepts.</li><li>This one should handle concepts a little better than v9_sdxl, and is less prone to exploding gradients as well.</li></ul>v9 Pony Release:<ul><li>This model has been training for over 2 months now, but since Flux dropped, I decided to release what I have so far to free up a GPU. Technically it should have trained for longer, but I'm impatient, and some of you are probably tired of waiting anyway.</li><li>The tags are mostly the same as the last v8 release for SD1, with a few new additions like blob content for example. See the tag.csv for more in "Training Data".</li><li>Pony is a little tricky to train on, so I was experimenting a lot with this model. Because of this you should try to keep the DoRA strength near 1.0. Anything above 1.1 tends to explode. (weight regularization like scale_weight_norms is critical for training on pony, fyi)</li><li>To keep training time reasonable I trained at 768x768 resolution initially, and had planned on finishing up training with 1024px resolution, but then Flux happened. The results still seem reasonable.</li></ul> I put plans and progress here every now and then.<h2 id="changelog-article-link"><a target="_blank" rel="ugc" href="https://civarchive.com/articles/3320">Changelog Article Link</a></h2>

hyperfusion SDXL DoRA 600k images

Trained on 400k (hyper focused) images extracted from the larger hyperfusion dataset. The big takeaway from training this version is that scale_weight_norms is critical to keep the model stable on Pony. If you remove it the model will quickly become unusable over the course of a few epochs. Results may differ for smaller datasets. Training Notes:<ul><li>~400k images</li><li>LR 3e-4</li><li>Unet only training (It might be slower but SDXL really doesn't need TE training in my experience)</li><li>batch 4</li><li>GA 16</li><li>dim 16</li><li>alpha 8</li><li>c_dim 8 (technically a DoRA LoCon)</li><li>c_alpha 4</li><li>optimizer: started with Adafactor, switched to Adamw8Bit half way.</li><li>scheduler: cosine</li><li>base model Pony</li><li>flip aug</li><li>375 token length</li><li>bucketing at 768 max 1024</li><li>tag drop chance 0.1</li><li>network dropout 0.25</li><li>tag shuffling</li><li>--min_snr_gamma 4</li><li>--ip_noise_gamma 0.02</li><li>--scale_weight_norms 1</li><li>custom code to drop out 75% of tags 5% of the time to hopefully improve short tag length results</li><li>15% CogVLM captions inserted 30% of the time.</li><li>over 60 days training time</li></ul>

Files with SHA256: a64aa5fbe50e3b239cdf330de79c18119ceb50f389b765c589337f65b9f564c6