My read
- Best talking mouth: Wan 3.0 (17.5 credits a shot). Wan 2.7 is close behind, livelier and cheaper (12.5).
- The version in the ad now is the weakest: Gus's mouth barely moves while he speaks.
- All four kept Evan's exact voice line, so the sound in the ad doesn't change.
- If you like it, I redo Gus's 4 talking shots in Wan 3.0 (about 70 credits) and re-render the ad.
All four at once, on the same voice line
Top left: in the ad now. Top right: MiniMax H3. Bottom left: Wan 2.7. Bottom right: Wan 3.0.
In the ad now: Kling 3.0 pro + Sync Lipsync 3
Kling animates the shot silently, then Sync redraws the mouth to the line.
- about 26 credits
- our exact voice line
- face moves 0.8× more while talking than in the pauses
- His mouth barely moves while he talks: frame by frame it stays mostly shut through the line and opens wider after it ends.
- The look and framing are what's in the ad now.
Two frames a second
MiniMax H3 (2K)
One step: picture and lip-sync together. Takes the still only as a reference image.
- 10 credits
- our exact voice line
- face moves 2.1× more while talking than in the pauses
- Moves his mouth with the words, but holds the same toothy grin the whole time.
- Starts wider than our picture (more street, smaller Gus), then pushes in; it doesn't start on our frame.
- Sharpest file (1440×2560) and the cheapest.
Two frames a second
Wan 2.7 (1080p)
One step, and it starts on our exact frame.
- 12.5 credits
- our exact voice line
- face moves 1.9× more while talking than in the pauses
- Mouth opens and closes with the words; the head tilts and the ears flap, so he's the most animated.
- Starts on our exact frame. The ears flop wide near the end.
Two frames a second
Wan 3.0 (1080p)
One step. Takes the still only as a reference image.
- 17.5 credits
- our exact voice line
- face moves 1.2× more while talking than in the pauses
- Clearest talking mouth: distinct open and closed shapes on the syllables, and it rests in the pauses.
- Matched our picture almost exactly, although it only got it as a reference.
- Calmer head movement than Wan 2.7.
Two frames a second
