<h3 id="reignsoles-v3.0-(anima-base-1.0):">ReignSoles V3.0 (Anima Base 1.0):</h3><p></p><p>Finally, version 3.0 is out. In this version, I massively up sized the dataset (from ~172 images to 417). Around half of the dataset was tagged with a combination of natural language and danbooru style tags, and the other half was just tags. It is now a LoKR now as well, a new advanced type of LoRA that boasts higher performance at a much smaller size. It's compatible with ComfyUI, and it should be compatible with other supporting UI's like SD Web UI Forge Neo. <br /><br />Similar to prior versions, the tags "perfect feet" and "detailed soles" were added to each image with the highest standard of quality I was looking for in both foot shape (perfect feet) and detail (detailed soles). This model was trained in a way where images were resized to fit into a common image size / resolution, with the most common resolution by far being 832x1216. I recommend using this resolution alongside the model for the best results. </p><h3 id="reignsoles-v2.0-(anima-preview-2):"><br />ReignSoles V2.0 (Anima Preview 2):</h3><p></p><p>This is an upgrade in almost every sense of the word. Images in the dataset were further selected for and refined, more images were included in the dataset, improved captioning, improved natural captions, etc. It was trained on Anima Preview 2 rather than the original Anima Preview. Furthermore, the LoRA was trained using tag dropout to ensure even simple prompts could generate quality images. Similar to 1.0, each image was manually captioned with a natural language description along with Danbooru style tags. This Lora should also be capable of realism to a limited degree, as it was trained on a subset of real life images of feet as well, each again personally tagged and manually captioned. Finally, I lowered the size of the LoRA to increase it's compatibility with other Anima-based models aside from the original Anima Preview. Lastly, I added in the images from my other illustrous LoRA into the dataset. This allows it to generate feet in the 'foot pussy' position as well to a high degree of accuracy.</p><p></p><p>The tag "perfect feet" was used on the images in training which had feet I subjectively liked the most. And the tag "detailed soles" was used when the sole of the foot had visible depth and wrinkles. The tags "real feet", "real soles", and "realistic" were added to every real life image, so use those tags exclusively when prompting for realism.</p><p></p><h3 id="reignsoles-v1.0-(anima-preview):">ReignSoles V1.0 (Anima Preview):</h3><p></p><p>This LoRA enhances the quality of the soles of women's feet for use with Anima in ComfyUI. It was trained on 109 high quality hand-selected images of women's feet from Danbooru. Said images were all manually captioned with a natural language description along with the accompanied Danbooru tags.</p><p></p><p>Using tags such as "perfect feet" or "detailed soles," you can expect to increase the quality and realism of the generated feet. In addition, it enhances text to image accuracy when describing the position and pose of the feet relative to the other parts of the body / image.</p>

ReignSolesAnima

<p>New and improved with a vastly expanded dataset (417 images) along with more training time. 162 epochs. Trained on Anima Base 1.0. This version of the model is technically a LoKR, which from testing I found to have higher quality results at a smaller size than a traditional LoRA at the cost of increased training time per epoch. </p>

Files with SHA256: 26d29a4ccd4b4f6753e7c27db9cfe2228ea9eb23652678ea49755f80bcd33819