<h1 id="hey-everyone!">Hey everyone!</h1>This LoRA is based on a curated dataset from 3D artist RudeFrog.Version 1 used 1,527 captioned images. There is major progress in style consistency, as well as significantly fewer artifacts.Great for creating very curvy women.<h3 id="v1">V1</h3>Stable release. Previous issues have been addressed, backed by much more extensive training. Significantly better NSFW support.<h3 id="v0.5">V0.5</h3>Images had noticeable grain, and the style wasn't as consistent across different prompts.Generation details &amp; parameters used for the images above:<hr /><h3 id="generation-details-and-parameters-used-for-the-images-above:">Generation details &amp; parameters used for the images above:</h3>Model:krea2_turbo_fp8_scaledCLIP:Clip Huihui-Qwen3-VL-4B-Thinking-abliterated-fp8_scaled (reduces image grain/banding)Steps: <code>8-10</code>Sampler: <code>euler</code>Scheduler: <code>simple</code>Recommended LoRa strength: 1 is optimal; 0.7 to reduce thickness.P.S. The goal wasn't to replicate an exact 1:1 3D style, but rather to push toward realism while retaining the key aesthetic traits of the 3D artist. If you like the style, please drop a like—it lets me know there's interest in more LoRAs like this!

Files with SHA256: 68b40e62b538fa1f0c8ab6a659a2e92e3d1b9ebb32a2924a30330d13bb7cb349