More than one person in a thumbnail
The Avatar field on the dashboard is multi-select. Pick two, three or more saved faces and the generator composes them into a single frame together — which is what collab videos, podcasts, versus formats and reaction content all need.
The picker shows the first few selected as a stack, with a count for the rest, and a Clear selection control to start over.
Using it
- Save each person as their own avatar first. A guest needs to exist as a face before they can be in a frame.
- On the dashboard, open the Avatar field and select every person who should appear.
- Describe the relationship in the scene description. This matters far more with several people: "two hosts facing each other across a podcast desk" produces something very different from "two people standing side by side".
- Pick a reaction if the emotion should be shared. If the two people need different emotions, set one here and adjust the other in the editor.
- Generate. Both variations will place the same set of people.
Two is the sweet spot
A 16:9 thumbnail viewed at sidebar size has room for two faces that read clearly. Three is tight. Four is a group photo nobody can identify at 168 pixels wide. If you genuinely need four people, consider one hero face plus the others smaller in the background.
Formats that need several faces
| Format | Composition that works |
|---|---|
| Versus / debate | Two faces, one per side, facing each other, split background |
| Podcast / interview | Both across a desk, host slightly larger or nearer |
| Collab | Side by side, same scale, shared reaction |
| Reaction | Reactor large in the foreground, subject smaller behind |
| Guest series | Fixed host position, guest swapped per episode |
The efficient way to run a guest series
If you publish interviews or podcasts, do not generate a new thumbnail per episode. Build one frame and reuse it:
- Generate one strong two-person layout with your face and a placeholder guest.
- For each new episode, save the guest's photo as an avatar.
- Open the original thumbnail in the editor and swap the guest's face in (Pro and up).
- Replace the guest's name in the image text.
- Download. Two edits per episode instead of a full generation, and every episode is visibly part of one series.
This is the single biggest time saving available in the product for interview channels, and it is why Pro's unlimited saved avatars matter more than the credit allowance for that use case.
When it goes wrong
- Faces blended together. The scene description was too vague about their relationship. Say explicitly where each person is: "on the left", "on the right", "behind".
- One person dominates. Use the avatar actions in the editor to resize, or pop one out and place them precisely.
- Both faces too small. Cut a person. Three heads at sidebar size are three unrecognisable blobs.
- Identity drift on one of them. Swap that specific face back in from your saved avatars.
Permission
Everyone in a thumbnail should have agreed to be there. For guests, ask for a photo rather than pulling one from their channel — you get a better source image and an unambiguous yes at the same time.