Genereal description
I thought it would be awesome to create a template for gifs. This lora should help with that. It should create some almost identical captions of a girl who are performing a fellatio.
Enjoy)
You can try out this Lora online using Dazzle.ai
Feel free to ask any questions here in comments, or in my discord channel.
Also I am making games with AI arts. They will be free in the future, but if you want to participate in making or have an early access - you can support me on patreon)
Also eq2bcknatagzqbylcjob (many thanks to him) mad a really cool script that will help with splitting the image into frames.
And DigitalDreamer (many thanks to him) made a script that may help to make gifs inside AUTOMATIC1111. Just place it inside scripts folder and restart UI. You'll find it in the UI down bellow in "scripts" dropdown. Use 2 when you made 2x2frames, and 3 for 3x3frames. It will make a gif and put it in your outputs folder.
tags to trigger
4bj, pov, 2x2 frames, 3x3frames
Optional tags
fellatio, blowjob, cum, cum in mouth, cum from nose
How to use
If you can't get anything good - here is the tutorial.
Versions
v4OT
*OT - stands for over trained.
A lot of people are strugling with the frames set, so I made it overfitted. Hope this helps.
Use lower weight of the Lora if you want to stick closer to the model's original style.
v3
Retrained, borders are now gone. Now it can generate stable grids with 768 and even 1024. If you are strugling with receiving a grid - use 2x2frames tag (3x3frames is kinda working to, but meh).
If you are still unable to generate grid - use controlNet for guidence.
I tried to make a gif with new approach - you can inpaint a part of the picture, like one of the frames, and use the same seed, but high denoising strength (I used even 1). Use 'latent noice' for more variability or 'original', if you want something close to the original frame.
Could be better, but it's ok, I think)
v2
Retrained, works better. Now works with character Loras. I tried to use it with my Viv character - results is in the examples.
*Tends to create photorealistic pics
v1
Basic version
Description
base model
FAQ
Comments (12)
I don't get how it works. It makes image like yours? 4 images at once with different stages of DT? Wouldn't it bee too small to make 512x512 if there will be 4 images in it?
Doesn't seem to work. Maybe I'm missing something but after 80+ generation over 4 different models I never once had the 4 squares pattern. Though it is pretty consistent at doing good pov BJ.
same Sd version?
How to make GIF，please
Create an image with this lora, which looks like the example (four frames at one picture), upscale it for better quality, split it to the frames, then use gimp or any other editor to make a gif from it)
@aDDont @aDDont thank you，
Ask a basic question. How split it to the frames
@sqdhl You can use any photo/picture editor. Even paint if you are using windows (select a frame and crop it, the save as different files). Or you can use something like https://pixlr.com or any other online editor, then just crop it and save to you computer.
@aDDont Thank you very much.
https://github.com/sj-data/stable-diffusion-webui-4koma-gif2grid I have created a Automatic 1111 script to split gifs to grids and grids to gifs inside the Web UI.
@sjdata867 I know you didn't create this model, and thanks so much for posting your extension! What about generating a 4x4 or higher? Even say 8x8...or whatever, really...I know about the "Seed Travel" script, but it doesn't exactly do what this model achieves (a decent change between the frames). I just think the apparent limitation of 2x2ish could be increased, no? If this effect could be generated using a script or extension, it would be so much more versatile. I figured I'd throw that at you, since you obviously have a clue about coding/scripting; any ideas on that?
@phreakheaven I should have asked DD before posting here but the script is a free useful tool. So DD's Grid LoRas are capable of generating 3x3 and 4x4 grids. (as well my extension can work with up to 4x4). You lose quality the higher you go obviously. So here's my workflow: Canny controlnet is a great way to force the kind of grid you want. What I do is use my script to split a Gif that is somewhat like what I want to create in the dimensions I want. I use the Start settings of control net at one and the end at 10-20 steps, unless you want your end grid to look like your start grid, then just leave at 1-100.
I always generate at 512 and then use the "upsample" option in generation to upsample to 1024. I do believe this Lora can handle 738x738 so experiment. No matter what you generate at my script will currently resize each frame to 512, so if you have a 1024x1024 4x4 grid, the frames would naturally be 1024/4=256, they will be resized using PIL so you may want to use extra upsampling.
TL:DR: Control net to get what you want, upsampling to get the quality.