Flux POV Handjob

This LORA was trained at 896x1152 pixels, purely because the dataset images I collected were portrait. Trained using ai-toolkit with an RTX3090 for 6 hours on 42 high resolution images. Preparing the images was a time consuming process as, apart from rigorous manual captioning, most of them had the studio logo in the bottom right covering the hands and sometime the penises. These had to be either cloned away using Photoshop or inpainted out using flux. This hard work payed off as I think the LORA is flexible and allows Flux's realism and flexibility to shine.

Prompting

Check my generation images but generally make sure you include 'POV kneeling and holding the viewers ?????'. Describe what you want to see and this LORA should oblige (within reason). Text adherence is OK but not great. Hand fidelity again is OK but not great. Not always 1 shot, but after a handful of generations you should get something like.

EXTRA TIP: If you prompt 'leaning forward and sticking her tongue out' you might get lucky ;)