Hedra Character 3 is documented as an audio-driven video system for a talking or singing character.
Sound is the driver, not a late optional layer
Unlike a general text-to-video prompt that invents speech on the fly, the developer form requires a start frame and audio. Its audio input accepts a duration from half a second up to 600 seconds, while the product model page lists a ten-minute maximum. This makes it conceptually suitable for one-presenter explainers, though a long technical allowance is not proof of consistent performance across every minute.
Input validation before a render
Use only a portrait that you are entitled to animate and speech material with appropriate licenses and identity consent. The displayed API form lists a start-frame upload maximum around 10.4 MB and audio up to about 104.8 MB. It permits multiple audio inputs with speaker-position annotations, which matters when more than one person appears. Validate the chosen image format, aspect ratio and audio duration before uploading; separate acoustic cleanup from the image generation stage so review can attribute a defect to its cause.
A two-speaker example
Imagine a historical dialogue where two museum staff members introduce an exhibit. Secure signed use permission for both likenesses and the recorded voice tracks. Choose a frame with enough separation to designate speaker positions, link each audio track to the correct person and generate a brief approval segment first. Inspect whether mouth movement follows the correct speaker and whether the silent presenter makes unintended expressions. One fixed portrait cannot guarantee credible hand motions or camera variety.
Version, outputs and price units
The model specifications list several aspect ratios and 540p, 720p and 1080p output options. Hedra’s first-party model pages also show per-second cost examples for these variants, but those are not the same thing as a formal enterprise invoice or a promise of region availability. Store the date checked and currency with the quote; AI 10 owns a comparable pricing registry. The apparent length of uploaded audio and the generated video duration can diverge when a request supplies an explicit target.
Review and finishing
Record input consent IDs, approved script, source audio and output task ID. Review lip synchronization on consonants, speaker changes, eyes and unusual phonemes; measure subjective continuity with a documented human rubric rather than claiming numerical accuracy. Use a standard editor for title graphics, brand color and cutaways. Any accidental resemblance to an unconsenting person requires intervention, not a simple prompt tweak.
Relevant alternatives and trade-offs
HeyGen Avatar V prioritizes a digital twin with motion reference; Synthesia focuses on scripted enterprise presenter workflows. Compare them only after defining whether the job begins with speech, a motion reference or an approved template.