Here's a new model I've pushed to Replicate. It uses gpt-5 with tool-use and ffmpeg to run prompt based video edits: https://replicate.com/fofr/smart-ffmpeg The basics work: concatenate, trim, reverse, re-encode, watermarking, cropping, stitching... Weird and complex stuff works too: 2x2 grid of videos, flip every third frame, make this video trippy, stitch videos of different aspect ratios and frame rates