Spatial extension for video, powered by RunComfy's CLI and Wan 2-7 edit-video. You feed it a vertical clip and a prompt like "extend the canvas to 16:9 by adding matching environment on the sides" and it gives you horizontal output with the original action centered. Works well for aspect ratio swaps (TikTok to YouTube, square to wide) and quick social cuts. Seam quality is acceptable for most internal and social use but not hero delivery. For that, the skill points you at dedicated ComfyUI outpainting workflows like the LTX 2-3 graph, which the CLI can't reach directly. Honest take: prompt-driven spatial extension is still early and you'll notice the seams on close inspection, but it's fast and beats manual rotoscoping for non-critical work.
npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill video-outpainting