rzgar/r_drive

This repo holds several Qwen-Image LoRAs that I initially shared on CivitAI.

• Ewa_Qwen and V3b are character LoRAs (All LoRA files are trained on both SFW and NSFW data; trigger word: Eva_Qwen)
• Eva_Qwen V3 is compatible with Qwen-Image-Edit 2511 and trained on a large diverse dataset with high F2P accuracy. Example prompt: (in a kneeling position she is performing .... Example NSFW prompts in the workflow Qwen-Image_Edit_2511_Eva_V3.json)
• Eva_Gothic is trained exclusively on an anime dataset, 80% SFW (trigger words: anime style, Eva_Gothic)

Qwen-NSFW-Beta 1, 2, 3, META4 and 5 (V5 was trained on the same dataset but with a different prompting style than the rest e.g., Gender:"", Act:"". Style:"". Cloth:"". Description:""):

• These LoRAs were part of an experiment I did at the time of Qwen-Image's first release. It was an attempt to figure out accurate genitalia rendering and find a sweet spot for LR and steps per image.
• META4 was merged using a linear method with weights perfectly tuned after extensive tests. It is also compatible with Qwen-Image-Edit; at 0.76, 0.8, or 0.86 strength it gives better genitalia presentation than what Qwen-Image-Edit renders by default.