Lip-Sync deep-dive: Act-One vs LivePortrait on the SAME original character, same audio, side by side. Full breakdown.

by @jordansmith1950 in c/character-animation · · Lip-Sync

Drove the identical 8-second line onto my OC (a stylized sci-fi mechanic, semi-realistic) with both Act-One and LivePortrait so you can see the real tradeoff instead of marketing clips.

Act-One:
Performance quality: excellent. Because I acted the line into my webcam, her eyebrows, micro-squints, and a little scoff at the end all came through. It FEELS acted.
Identity hold: very good, Act-One preserved her design well.
Cost/friction: paid, cloud, but genuinely one-take easy. Record yourself, upload, done.
Weakness: it took MY head movement too, so I had to perform stiller than feels natural to keep her 'in frame'.

LivePortrait:
Performance quality: good lip tracking, but the upper face was flatter — less emotional transfer. The scoff didn't land.
Identity hold: some jaw-stitching artifacts on the stylized face, needed cleanup.
Cost/friction: free, local, but 40 minutes of parameter wrangling for this one clip.
Strength: total control and zero per-clip cost once dialed in. Great for VOLUME.

Verdict: Act-One for hero emotional shots, LivePortrait for background characters / high volume where you just need mouths moving. I use both in one project. Side-by-side in comments, watch the end scoff specifically.

923 points · 6 comments

Comments

  • @hanasmith_art · · 71 points

    using BOTH in one project based on whether a shot is hero or background is the mature workflow. not every character needs a performance. the barista in the back just needs a moving mouth. spend your Act-One budget on the faces that carry emotion.

  • @avatanakaworks · · 99 points

    'the scoff didn't land' on LivePortrait vs landing on Act-One is the whole ballgame for emotional work. the upper-face flatness of audio/landmark-driven tools is their fundamental limit — they sync the mouth and approximate the rest. Act-One transfers an actual PERFORMANCE because a human performed it.

  • @noahmiller · · 88 points

    the 'I had to perform stiller to keep her in frame' note is such a real and underdiscussed Act-One limitation. it maps your whole head, so you become a constrained puppeteer. great actors who move a lot actually struggle with it.

  • @masondavisworks424 · · 54 points

    what's your identity stack here — LoRA, InstantID, both? the face holds better than most and I'm trying to reverse-engineer the pipeline.