Rearranging a fake 70s album cover with MOSS-Image layout boxes
retained text
I made up a 70s folk band just so I'd have something to push around with the layout JSON in MOSS-Image. Every person, the station wagon, the dog and the band name get their own box and a short caption, and the scene caption stays the same in all four. 1. First layout: the three of them on the left, the wagon parked behind them on the right, the name across the sky. 2. Dragged the wagon to the left and moved the band over to the right. 3. One more box for a big shaggy dog lying in the grass. 4. Swapped the title for an italic serif and squeezed it into the top left corner. All four are seed 2. I ran three seeds for each layout and everything landed in its box on all twelve, band name spelled right every time. The last image is the fourth layout with my boxes drawn on top. For the look it's MOSS-Image-Base with the realism recipe from the other threads (ER-SDE at 1.1 noise, cfg 3, 25 steps, SLG 22 and 24) at 1440 square, and the JSON goes straight into a regular CLIPTextEncode. No custom nodes. A back cover with a full tracklist is next on my list.
document fetched 2026-11-19 00:42 · canonical https://reddit.com/r/comfyui/comments/1xz8bds/rearranging_a_fake_70s_album_cover_with_mossimage