Back in the day, the biggest headache when drawing a set of story illustrations was keeping things consistent. You’d generate one stunning image, but the next panel would have the characters in completely different positions and the background layout all messed up. The only fix was endless inpainting and local rework to force it together.
Lately I tried MJ’s dynamic composition feature, and it actually seems to remember the relative positions and spatial relationships of characters across a set of images. The transitions between multiple shots are way smoother now, and the emotional flow can carry through too. For us doing doujin storyboarding, this is way more useful than just stacking keywords.
Of course, it’s not perfect—complex multi-character scenes still drift—but at least I don’t have to start from scratch aligning every single panel. Has anyone here used it to run a sequence longer than four panels? Curious where the limit is.