AI avatar lip-sync: Fast vs Quality with the same image and audio

01 / Avatar workflow

What stayed the same, and what changed?

Both runs used the same stored image and audio assets, with face framing, natural motion, expression scale 1 and face enhancement off. The narration is 10.397 seconds; both API records report 10.419-second outputs. Only the Fast/Quality route changed. Runs were made separately on 7 November 2026, not under a synchronized latency test.

There was one output per mode in this comparison and no best-of selection. Both MP4s fully decoded without video or audio errors, and local transcription recovered the same three-sentence script. Sampled frames were reviewed for facial continuity. We have not measured phoneme-level synchronization or performed an independent blind human assessment.

02 / Avatar workflow

Watch both original outputs

Play the clips separately with sound; simultaneous playback makes lip-sync harder to judge. Look at mouth closure, consonants, teeth, blinking and head motion rather than treating the upscale dimensions as proof of extra detail. The clips below are original provider outputs, not retouched or time-stretched versions.

03 / Avatar workflow

If the lip-sync looks wrong

First listen to the input itself. Remove accidental silence, distortion or overlapping speakers in your own editor. Use a face with an unobstructed mouth and avoid changing both the image and voice while comparing modes. Test a short passage with the same inputs before spending on a longer video.

Fast is the explicit lower-cost route; Quality remains the default. Neither guarantees perfect speech alignment for every face or recording. A different face, language, camera angle or speaking pace can change the result. For two visible people, assign a separate track to each side rather than sending a mixed conversation to a single speaker.

04 / Avatar workflow

Free access, costs and commercial use

Start without a subscription using free welcome or earned credits; paid credits are optional. If credits are insufficient and a fresh check confirms no surveys are available, eligible accounts may receive automatic lower-priority processing. It requires available free funds and must fit the existing daily subsidy limit. It does not promise unlimited free avatars of any duration.

Fast: 45 credits per billed second. Quality: 135 credits per billed second. Round audio duration up to whole seconds with a 3-second minimum. A 10-second avatar costs 450 credits in Fast or 1,350 in Quality; 30 seconds cost 1,350 or 4,050 credits respectively. Both include the 1080p-target upscale. Quality remains the default; Fast is an explicit lower-cost choice, not a free-processing switch. Creating the input image or voice separately has its own quote. The editor confirms the current price before submission.

Donatello permits commercial and monetized use of Avatar outputs under the Terms, including client explainers, lessons and marketing. You must have permission for the image, audio, voice and likeness, and comply with applicable provider terms. This is not a guarantee of copyright, exclusivity or freedom from third-party claims. Do not impersonate someone or present a synthetic presenter as a real testimonial.