This lora was trained off of google banana pro images. I know the watermark is still there, I can mask them out, this was a first run through as I haven't used Qwen on AI Toolkit yet.Weakness, her breasts are burned in by her bikini, using my NSFW female body enhancer V0.3 lora will help a ton.Training Program: AI Toolkit (Onetrainer sucks for Qwen, this is so much better)training resolution 768 (The images were 768 anyways and it worked out very well with RAM)rank 32images 18Trained float8Low VRAM enabled <pre><code>noise_scheduler: "flowmatch"
 optimizer: "adamw8bit"
 timestep_type: "weighted"
 content_or_style: "balanced"
 optimizer_params:
 weight_decay: 0.0001</code></pre>Trigger word is mandatoryJ0s13tips: using "wolf" can alter the imagedon't use female, the word female is tied to it being human, it does her somewhat well with her human side.I didn't put much love into this as AI Toolkit produced a usable result in under 2 hours. This lora needs the 1 step save treatment. Default training settings are perfect (Disable Quant)Follow them on their Fur Affinity account- (<a target="_blank" rel="ugc" href="https://www.furaffinity.net/user/spacepoptart">https://www.furaffinity.net/user/spacepoptart</a>)X or twitter (<a target="_blank" rel="ugc" href="https://x.com/cosmicW00F">https://x.com/cosmicW00F</a>)

Josie Realistic (Qwen) Google Banana Pro Dataset

trained resolution 1536All in one (original, modem, banana pro), doesn't really work but what ever. captioned as "original design of j0s13, modern j0s13"

Files with SHA256: ccc13c77fb447abfe1f9ed5488f9f0e60779377e42d81e1a13e3295ffe9e37dc