Do you want to support me? Consider being a patron on Patreon to vote on train polls and access stuff early. ❤️
Lite models are less trained versions of the Full models but yet usable ones. Full models will be on Patreon and after 1-2 week span I'll release Full models on our Discord server.
Trained with 100 images and 768x768.
Use 512x and hires fix for better results
Tags you might need are
1girl, solo, pussy, from below, nude, pubic hair, looking down, clitoris, ass, anus
Use the LoRA weight around 0.8 and 1 for the best results
Discord server feel free to drop and say hi ^^
Description
FAQ
Comments (12)
lol cant even say "this doesnt work reliably for me" anymore cuz its the lite model... But yeah this is not compatible with character loras for me
I tested on couple characters most of them is working. Having trouble with characters trained on dynamic poses like Makoto but still they working ok even in lite version 🤔 Contact me on discord if you want, can find the problem
Its much easier to describe the body type you want and inpaint a character face afterwards, esppecially for these kind of lora's where nothing is interacting with the face
1. Generate the image first using this + a checkpoint for real people. Try to describe the character you're looking for to the best of your ability. You may need to prompt in bigger eyes if you are trying to emulate anime.
2. Change it over to an anime checkpoint such as Anything v5 (if you're hoping to create an anime character like this)
3. In paint the face with the specific character face you are looking for at full strength.
4. In paint the hair to be the hair of that character. You may have better results if you draw in the flat color general shape of the hair as an in painting mask if you're struggling due to the ambitious angle.
5. Once you have the hair and face correct, send it to image to image and use the entire prompt that describes the character, the scene, all trigger words, the character lora at full strength, this lora for the angle at full strength, and add these to the positive prompt: anime style, cel shaded, flat shading. For negatives, use EasyNegative (download it on civitai), soft shading, realistic, realism, photograph, photo, photography, photo realism. Run it through with something now lower than .2 and probably generally no higher than .4 for the de noising strength. Keep testing it until it works.
6. If any part is bad, fix it with inpainting. You will usually need to go back in and inpaint the face one more time. This time set it to upscale it and go for 768x768.
These tricks have worked repeatedly for other loras, and have made my anime pictures better because when you use images of real people for a lora, the proportions and anatomy have a higher chance of coming out correct for once in your stylized picture.
@xSatouMatsuzakax yeah that is actually pretty good advice in general. i still dont like doing that tho because i like the body proportions to look like the character too(abs, legs, boob size,etc) and when you try to inpaint those too, at that point thats way too much to do
@veryfirstquestion800 The bodies turn out more like the characters if you're using a character lora that has trigger words that work outside of using the lora. If it doesn't spawn in the right proportions, run it through several more times until you need it. Use more negative prompt for body types you aren't looking for, use lower weights for loras that add those things on principle.
@xSatouMatsuzakax again, i understand that is a thing that can be done but that is like saying "instead of putting your food into your microwave to warm you can actually go outside, collect wood,make a campfire and then put your food on top of it" it can be done but you can also just use a microwave. I am no expert but i assume these incompatibility issues either come from the tags/model or alpha/dimensions before a model is trained and can probably be improved.
@veryfirstquestion800 Who are you trying to spawn in? If the chest is too larger, you won't be able to see her face, but you'll have the effect of the general size they are. If they need to be way smaller, you can use the squeezer lora, and then go back in and in paint the face on full strength with the character lora and you'll have smaller breasts, but a mature face.
This is very consistent for realistic results for me.
Awesome angle
If you decide to do a follow up, consider changing 'looking down' as tag, maybe use 'looking downward', because the former also means 'expressing a depressed mood'
looking down is general tag for booru. for "down" booru uses "sad" tag
Details
Files
MS_Real_POVPussyFromBelow_Lite.safetensors
Mirrors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
MS_Real_POVPussyFromBelow_Lite.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.




