Evaluate Video (Mock)

This node scores a video on five consistency dimensions - color stability, garment structure stability, face identity, anatomy quality, temporal flicker - and every single run returns a perfect pass. That's not a flattering verdict on your work. It's because the mock evaluator doesn't look at the video at all. It reads the video_record, emits a fixed VideoEvaluation (94, 92, 95, 93, 96, decision pass), and moves on. Deterministic, free, and spectacularly uninformative about actual video quality.

Which makes it sound useless, and honestly as a quality tool it is. But as a contract tool it's quietly doing a real job. The mock video provider produces fake motion, and this mock judge produces fake scores, and together they let the entire video branch of the pipeline - generate, evaluate, export, manifest - run end to end at zero cost. When a real video provider lands (milestone 4), this node's output type (AFS_VIDEO_EVALUATION) is the shape that real QA will have to fill, so anything you build downstream of it today won't need rewiring. That's the whole point of the mock slice: build the plumbing against honest placeholders.

Inputs are video (the native VIDEO from Image to Video (Mock)) and video_record (its provenance). The node needs both - the video itself is technically unused by the mock scorer, but the socket exists because a real evaluator will need it, and the contract is the contract. Outputs are video_evaluation (the AFS_VIDEO_EVALUATION object) and video_evaluation_json. The evaluation carries a weighted total_score too (25% color, 25% structure, 20% face identity, 15% anatomy, 15% flicker), which mirrors how the image side weights its dimensions.

Wire it like this: Image to Video (Mock) → this node → Export Run. That chain is what makes the exported manifest a complete record - video metadata and its evaluation together. If you leave the evaluation stage out, AFS_ExportRun has an empty socket and the run isn't complete, so in practice you always include it in the mock slice.

Nothing special to install: it's part of the zero-cost "Mock MVP" category under AI Fashion Studio / Mock MVP, needs no API key, no models, no environment variables. Pack install is the usual Manager search ("AI Fashion Studio") or clone into custom_nodes plus restart.

The one trap is psychological, not mechanical. When a mock evaluator hands you a 94 and a pass, the temptation is to treat it as validation of your video and stop worrying. Resist. The README is explicit that mock outputs exist to verify plumbing and UX, not to predict what real providers produce. Read the number the way you'd read a placeholder in a wireframe: it's there so the layout works, not because it means anything. When the real video QA ships, that's when the scores earn the right to be taken seriously.