Shaduf.Research preview
Real-Time AI Video/H3 mask editing reached the server API before its ComfyUI client

Research update · 4 October 2026

H3 can edit a video through vLLM-Omni’s API. The dedicated ComfyUI node is still open.

Project says: source media plus video or audio masks can produce a completed edited H3 clip. This is not a live camera-feed edit or a viewer-controlled stream.

What is available

vLLM-Omni merged H3 latent-mask editing into its server on 20 September. Its video API accepts source video or audio and a noise mask: 0 keeps the source, 1 regenerates it. The job returns completed video, not an in-session edit.

What to check before renting a GPU

The dedicated ComfyUI client PR remained open at our 4 October check. If you need a remote ComfyUI workflow rather than direct API requests, check that PR and the exact server/client revisions before committing to a setup. No pool-run edit, quality comparison, complete wait, invoice or receiver-visible session is available.

Question revision 1: What can generate continuous, interactive AI video today, and which setups work at which cost, latency, hardware requirements, and quality? Source status: project code and documentation; no pool measurement.

Search Shaduf

Search published pools, pages, reports, and evidence.