Here’s the experimental release of a JAV concept (still in revision) LoRA, extremely common in JAV filmotography, to enhance ‘plot’ and fetishes (盗撮/映像/撮影) – i.e. filmed in action.
PS: I scrapped like 4 LoRA models in few weeks time, and decided to try training this a multi-concept model.
Due to the complexity of this LoRA, here’s some
Brief instruction for use:
a) POV – filming with a smartphone: (loRA str: 0.6 – 0.9)
prompt: pov, pov hands, h4nd_with_sm4rtphone, cellphone picture
This is probably the easiest to prompt and replicate, as long as your content is fit for POV
(sex positions POVs, environment, camera angles). Use “cellphone picture”, to create a smartphone with a screen (with a chance) to actually print a mini version of the scene/girl being filmed. (doesn’t always work, but when it does, its totally worth it).
b) POV – filming with a camera/camcorder (LoRA str: 0.6 – 0.9)
prompt: pov, pov hands, h4nd_with_camera
Same as the above, but replaces the camera with a camcorder or black camera.
c) 3rd person filming: (LoRA str: 0.8-1.0)
prompt: multiple boys, c4meraboys, camera/camcorder
Here we deploy the cameraboy (‘c4meraboy’), which is suppose to be a person standing somewhere, with a camera. In theory. In reality, if you push the LoRA strength higher than 0.8, he will start popping out, if there’s someone there with empty hands. the “camera/camcorder” term is optional, use it only if you have difficult prompting this.
[Still trying to refine this]
Recommended approach:
Sometimes, the easiest thing to do is just prompt them all at once/in combinations, and see what you got, then trim the prompts back out from there.
Best performance:
Despite the training images come from photorealistic footages and screencaps online, this LoRA seems to do quite well in anime/2.5D models.
Works slightly better in portrait than lanscape, but does do a decent job, if you go with wide-angle prompts.
Works very well with “view from above”, “view from below”, “view from side” camera angles.
Good synergy with crowd prompts: “surrounded by boys, multiple boys, 6+ boys, 3boys,”
Good synergy with POV hentai/sex position LoRAs.
This LoRA loves: the inside of a train cabin, enclosed spaces (vehicle interior), bedroom, toilets, basements. 😉
Weaknesses:
Warning:
This LoRA will not work well with other LoRAs that triggers hand gestures, holding something, pointing, even BJ etc. as the diffusion process will just pop in a camera/cellphone on his/her hand more than 50% of the time. The only way to address this is to
1) reduce weighting on other competing LoRAs, or
2) use controlNet to enfore the conditioning
Adetailer correction can sometimes undo the “picture” in the smartphone. No easy fix for that as of now. To fix.
As with many LoRAs, this LoRA have serious hand-issues. Fix them with inpaint, or good-hand LoRAs at img2img; else you will lose cameras/cellphones.