Here is how to set up group photos, prepare dialogue audio and direct who does what.
Setting up a photo for multi-person AI video
Everyone who should move needs a visible face, ideally similar in size and not overlapping. Two to four people side by side in a horizontal frame is the easiest starting point — think podcast table, interview couch or stage line-up. Even lighting across all faces avoids one person looking flatter than the others.
If people are at very different distances from the camera, the closer faces dominate. Crop so the speakers are the focus.
Preparing dialogue audio
Upload one mixed audio file containing the whole conversation, in the order it should play. Leave small gaps between speakers so the turn-taking reads clearly, and avoid crosstalk where two people speak at once for long stretches.
Podcast and interview recordings work well as they are — trim the segment you want, normalise the volume and upload it as MP3 or WAV, up to five minutes.
Directing who speaks and who reacts
Describe people by position in the prompt; keep it short:
- “Left host speaks first, right host nods and laughs, then answers.”
- “Woman on the right leads the verse, man on the left joins on the chorus.”
- “Interviewer on the left asks, guest on the right gestures while explaining, static wide shot.”
- “Whole family waves at the end, gentle push-in.”
Formats that suddenly become possible
Video podcasts from a single still, sketch comedy with two characters, interview teasers, duets and choir clips, family greetings and classroom role-plays. Each used to require filming several people together; now it needs one photo and a recording.
Common issues in group scenes
If the wrong person appears to speak, make the positions in the prompt unambiguous — “the person on the left” rather than a name the model can’t see — and check that the speakers’ faces are roughly the same size in the frame. If everyone moves too much, ask for “listeners stay mostly still” so the speaker stands out.
Very large groups spread detail thinly across many small faces. For a choir or team shot, crop to the front row and keep the clip short.
Writing dialogue for two voices
Short exchanges work best: one person asks, the other answers in a sentence or two, and they trade lines quickly. Long monologues in a group photo leave the listener with little to do; break them up with reactions or follow-up questions.
FAQ
How many people can be in one video?
Groups work, but two to four clearly visible faces give the most reliable results.
Do I need separate audio for each person?
No. Upload one mixed file of the conversation and describe who speaks in the prompt.
Can people react while others speak?
Yes. Listeners nod, smile and glance at the speaker when you ask for it in the prompt.
Did OmniHuman 1.5 support several people?
Version 1.5 used a mask to choose one speaker. Multi-person performances are new in 2.0.
Can I use a family photo?
Yes, with everyone’s permission. Clear, front-facing faces work best.
How long can the scene be?
Up to 5 minutes in one take.
Can each person have a different voice?
Yes. Record or mix the dialogue with each person’s voice into one audio file, in the order they speak.
Will listeners stay completely still?
No — they make small natural movements. Ask for reactions such as nodding or smiling if you want more.
Does the group photo need to be horizontal?
Not strictly, but a wide frame gives each face more room. Vertical group shots work best with two people stacked closely.