Everything below uses OmniHuman 2.0, which takes one still image plus an audio track and generates a full performance: lip sync, expression, gestures and optional camera movement.
Preparing images for photo to talking video
Treat the source photo like a casting decision. Shoot or generate one hero image per presenter with soft, even light, a neutral background and the hands visible if you want gestures. Save it as a master file and reuse it for every episode — identity stays stable when the input stays the same.
The aspect ratio of the photo sets the aspect ratio of the video. Crop vertical (around 9:16) for Reels, Shorts and TikTok, square for feeds, and horizontal (16:9) for YouTube and presentations. Ratios between 0.4 and 2.5 are accepted.
Producing the audio track
Audio drives timing, so finish it before rendering. Record in a quiet room, trim silences at both ends and normalise loudness. Scripts read at a natural pace look better than rushed ones. If you use synthetic voices, export MP3 or WAV and upload it like any other recording.
Split long scripts only where a real cut makes sense — the model can handle a full five-minute take, which avoids continuity jumps between segments.
Producing a weekly batch
A simple routine that keeps quality and cost under control:
- 1Lock the master photoSame image, same crop for every episode of a series.
- 2Write and record all scriptsBatch-record on one day for a consistent voice.
- 3Render a 5-second testCheck framing and motion before committing full-length credits.
- 4Render the full takesQueue several videos; up to three render at once per account.
- 5Edit and captionAdd subtitles, music and logos in your usual editor.
Budgeting by the second
Cost follows duration: 10 credits per second. The Pro plan’s 1,200 weekly credits cover about two minutes of finished video — for example eight 15-second ads or two one-minute explainers. Plan scripts to length and you will know the cost before you hit render.
Quality checklist before publishing
Review each clip on a phone at full screen before it goes live. Look at three moments: the first second, where viewers decide to keep watching; any line with fast consonants, where sync errors show; and the final second, where the clip should end on a natural expression rather than mid-gesture.
- Trim the last half-second if the presenter freezes after the audio ends.
- Burn in captions — most short-form viewers watch muted.
- Keep the original audio as the master; don’t re-compress it twice.
- Name files with the script version so A/B results stay traceable.
FAQ
How do I keep the same face across many videos?
Reuse one master photo with the same crop for every render. Identity comes from the image.
Which aspect ratio should I use?
Match the platform: vertical for short-form social, square for feeds, 16:9 for YouTube. The output follows your photo’s ratio.
Can I render several videos at once?
Yes, up to three renders can run at the same time per account.
Is there a length limit per clip?
Each clip can be 4 seconds to 5 minutes long.
Can I add subtitles?
Add them in your editor after downloading the MP4; the model renders video without burned-in text.
What resolution are the videos?
OmniHuman 2.0 outputs 720p, which suits social platforms and presentations.
Can several team members share one presenter photo?
Yes, as long as the person in the photo agreed to that use. Reusing one master image keeps the presenter identical across everyone’s videos.
Does the photo need a plain background?
No, but a calm, uncluttered background keeps attention on the presenter and makes edges around the hair cleaner when they move.