This LoRA is based on screen captures from Strike Witches and is trained on keyframes of Lynette Bishop. The strength of the LoRA should be anywhere between 0.6 and 0.8. The recommended tags are: lynette bishop, 1girl, ahoge, single braid. Training dataset available at https://huggingface.co/datasets/Deathspike/strike-witches-501st. Join the Citrus Discord for a wonderful community focused on Sudachi and related models.
Description
FAQ
Comments (4)
Would you consider using the same dataset to create a purely style lora? I've been using this at low weights and it pairs incredibly with AkkaiMix.
Is that what you used on your image with her? It's gorgeous! That said, I currently have no plans to experiment with style LoRAs. I'm focusing on characters and, after that, some clothes. But the training dataset and settings I created for this model is available for free for non-commercial usage (cc-by-nc-sa-4.0) so feel free to try it yourself. Link is in the model description.
@Deathspike Yup, its a really good model, although it has a strong style, but you can mute that to a degree. I can probably reproduce your settings fairly well since that data is all saved into the file by the looks of it, just be a case of tagging out the features of Lynette. Did you use the dataset as is or do additional tagging? Also, what model did you train on.
@SpiritOfCulture Yes, all the settings are in the settings.json under lynette bishop. Those are the full reproducible settings for kohya_ss. The entire dataset and settings are available in the repository. If you were to train using those inputs, you would get an identical model as I uploaded here, assuming you changed the paths, of course. Nothing is hidden or changed for the release here. The model I trained on is also in the settings, which points to anything-v3 which can be found here: 万象熔炉 | Anything V5/Ink - V3.0 | Stable Diffusion Checkpoint | Civitai
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.