>>58108917
>>58111901
Right here! The future is now thanks to software! Presenting an extended cut of an infamous frame! My mark in the history of /dg/!
Though I'll just tell you now, it is kinda janky, you could tell in the first few seconds that my prompting wasn't the most consistent and some scenes feel too motion-heavy or inconsistent like the direction of the sky. I probably could've saved a lot of time doing the clothes bit in one prompt, but I thought it would be more satisfying in pieces.
For real though, thank you to that other diffusion anon for the inspiration. To other anons who have the specs to do this and are interested,
I'll leave my thoughts on a beginner who started a few days ago with their minimum specs requirement. If you're not gonna change any parameters in /ldg/'s workflow, then prompt management, batch count, and patience are important. In the dozens of failed clips I have done and hours wasted, some outputs match your ideas, but majority of the time they do their own thing which leads into time used on rephrasing prompts.
Some of my horrors were frames where the outcome has bits of the background fused into hair or clothes. If even one frame has something that's not supposed to be there, the model could generate something for that even if your prompt has nothing on that. For example, a slight reveal of hair turns into a complete head with a janky face. Also some crucial info, always follow the recommended resolution, this one deviated from that, but I'm not sure if it would look better/takes less time if I did follow it.
