I'm mainly working with Stable Diffusion, and it relies on trained 'checkpoint' or ckpt files, but I can't seem to find any reliable guidelines on how to create these 'from scratch'. They almost always suggest or recommend starting with an existing file?
Which I want to avoid, especially with all of the issues over copyright and trademarks using existing graphical content.
I'm simply not sure where to start, searching online for this tends to result in useless youtube video guides that simply point me to download other files, etc. I'd rather have an open source tool solution.
Is anyone else playing with these tools at the moment? Linkin Park's 'Lost' has somewhat inspired me to dabble with it.
In their music video, they've used cuts from footage of the band, and used the 'stable diffusion' type of AI generation as a filter, along with original animation art, and trained the filter on that animation art. So what you get is a mixture of original pieces, some AI weirdness and amazing effects:
I've played around with this myself on a video, it took about 1-2 minutes per frame to render it out and then recompile back into a video, but I'd want to train it on a checkpoint I'd had full control over instead.