My full ComfyUI to Kling to DaVinci pipeline for a 30-second product spot
Posting the whole thing because people keep asking how the spot I shared last week was made. This is the actual production pipeline, not a toy.
Stage 1 — Stills (ComfyUI, Flux): generated every hero frame as a still first. The product is a real client product so I used IPAdapter on their product photos to keep the bottle accurate. Never let the video model invent the product, lock it as a still.
Stage 2 — Motion (Kling 2.1): img2img from each locked still. One shot per still. Kept every camera move tiny and deliberate. I never ask Kling for big action, just controlled drifts and pushes, because that is what it does reliably.
Stage 3 — Assembly (DaVinci): cut, color-match across shots (the shots never match out of the box), added the sound design, graded for the brand. 80% of the 'professional' feel is this stage, honestly. The grade unifies the AI shots into one spot.
Node graph screenshot and the DaVinci node tree are in the comments. AMA.