╚══〔 <a target="_blank" rel="ugc" href="https://x.com/x_n15g">𝕏</a> ┊ [SD 1.5] ┊ <s>SDXL</s> ┊ <s>Pony</s> 〕══╝An update as of early 2024: I've largely abadoned this model as it never came close to performing as well as training against the NAI leak. In the end, despite intentions to move away from using the leaked model, it's far too useful for producing widely-portable LoRA and nothing I've tried since has come close.F3E/Forge is a LoRA training model I created for both generatong training sets and have also adopted as the new base model for my LoRA. I've decided to start open-sourcing my LoRA and wanted to have a base model that other's could use to recreate the training, and that wasn't directly based on the NAI leak.The model has a good understanding of anatomy, and has some NSFW concepts mixed in as well to allow the best flexibility when training. LoRA produced using this checkpoint perform best with anime/2d models, but also perform reasonably well with many realistic models as well. While not intented for generation, it can produce some very nice output and has the <a target="_blank" rel="ugc" href="https://huggingface.co/hakurei/waifu-diffusion-v1-4/tree/main/vae">WD anime2</a> VAE baked-in.I lost track of the actual recipe over several merges and adjustments, but the three merged models are:<ul><li>Foundation - <a target="_blank" rel="ugc" href="https://civarchive.com/models/23900/anylora-checkpoint">AnyLora</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/9409/or-anything-v5ink">AnythingV5</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/12606?modelVersionId=100675">MeinaHentai</a></li></ul>

Files with SHA256: 6a3edbbfafe147add3adf1dd1ffe768d3a6db39e885dcea1b2f2193fff8378f0