This is the middle section of 'Shinigami', the short film I'm currently making while researching. The workflow is from ComfyUI to Photoshop to Luma and then back to Photoshop and Premiere.

Don't be misled into thinking "AI" means you can ignore preparation, storyboarding, photography, writing, video editing, VFX, grading and scoring. Machines and genAI models on their own make a lot of mistakes that you don't see in the edited material you see online. You will always need to be passionate about film theory to get the best out of new tools.

As mentioned in other blog posts, 90% of the footage that I got out of video generators is unusable even though I gave them completed images to start with. Without a completed image to guide them the discard rate is closer to 100%. Most of the time they won't follow your instructions no matter how clear and simple you make them.