Sharing a little more of how I actually create these. No workflows at all, I do each step by tweaking and testing.

1. Made the image first. Prompt and LoRAs included.

2. Zoomed in on the face and ramen as 2 new images.
   Minimal prompt tweaking from the original (denoise 0.15), to have detailed reference images of that.

3. Experimented with video prompts at a lower resolution, and decided H3 "background music" prompting wasn't good enough for me.

4. Generated the YuE2 track with 60 second hard cutoff.

5. Ran a few filters (reverb, low pass, decreased gain) to make it sound muffled.
   Now thinking oh poo I forgot to make it mono.

6. Used that as audio reference for the final video prompt.