Ali G Car Dance AI: before-and-after test & frame audit
By Sendare AI Studio · Updated October 30, 2026
Most AI video generators show you the original TikTok clip and hope you assume your photo will look just as sharp. Here is the actual 480p output from our production Seedance 2.5 pipeline (`task_01M66X90RKFMYYMPBDM1BZE669`), measured frame by frame against both the uploaded photo and the Gordon Ramsay reference clip.
Left: the input photo shown on the Ali G Car Dance page. Right: the same second of the 9-second 720p example rendered from it. Nothing in the clip is stock footage or someone else's result.
Make your own Ali G Car Dance video
One photo, 9s 720p MP4 with synced soundtrack, no watermark. From $6.90 one-time.
What did the ArcFace face-embedding test measure across the 5 shots?
We ran InsightFace (`buffalo_l` ArcFace) on the input portrait, the Gordon Ramsay reference frames, and our rendered 480p keyframes (baseline similarity between the input photo and Gordon Ramsay is 0.0047 — two completely different faces). Shot 1 (1.0s front windscreen): 0.7230 vs input photo, 0.0106 vs Gordon (margin +0.7123). Shot 3 (4.2s front windscreen nod): 0.3134 vs input photo, 0.2166 vs Gordon (margin +0.0968). Shot 4 (6.5s side groove): 0.3937 vs input photo, 0.1695 vs Gordon (margin +0.2243). Shot 5 (8.5s closing windscreen): 0.2957 vs input photo, 0.1809 vs Gordon (margin +0.1148).
Where does the swap soften, and why?
In Shot 2 (2.0–3.3s, the passenger-window side cut where the driver waves his left arm), the head turns rapidly in profile and motion blur lowers the embedding score (0.1892 vs input photo). As soon as the camera cuts back to the front windscreen at 3.3s, the dark wavy hair, beard, and grey crew-neck sweater lock back onto the input portrait.
Did Gordon Ramsay's white chef jacket bleed into the render?
No. Across all 217 frames (9.04 seconds at 24 fps), the driver wears the charcoal grey crew-neck sweater from the uploaded photo underneath the blue four-point racing harness straps.
