Trained on 1024 x 1024, so if you want 16 frames, you need to generate at that, but it still gives good results lower, just shorter gifs.

There's a couple python scripts

"python frame_grid_to_gif_python.py" to start the one with a gui.

Or place the one with webui in the name into your webui scripts folder.

LOOK AT IT FIRST. I made it, so I know there's nothing malicious in it, but DO NOT TRUST strangers. Read and understand the code before using it!

I split a bunch of ?????????? gifs into grids of their individual frames, and trained this lora on them at 1024 by 1024, and this is the result.

You need to convert the images to a gif. I included a simple python script, gui and all, courtesy of chat gpt, that will make the conversion for you.

Not great, but not too terrible. I think it needs better training, and a more curated set. If you're interested in me improving this, then send me some mega links to 1 second long repeating gifs. This was trained on a bit over a hundred images.

The words this understands are:

frame grid, pov

frame grid, in profile

frame grid, offset pov

It's better at frame grid, pov, than it is the others; I might split this into separate loras.

NOTE: The second gif uses my Mizumizuni lora at 0.4. Grab that if you want similar results, but in general, this does better when you combine it with another lora.