When you create a Space or an Omniagent, you can upload a custom image or video to generate a personalized avatar. The system produces an animated, lip-synced digital companion — but it needs the right input to work well.
Two avatar models: v1 and v2
| v1 (default) | v2 | |
|---|---|---|
| Input | Still photo | Video (recommended) or still photo |
| Generation speed | Fast | Up to several hours |
| Quality | Good photorealism | Highest quality |
| Lip-sync | Good | Best accuracy |
| Facial hair | Standard | Better results |
What works — photo requirements (v1 and v2)
- Clear, front-facing photo of a real person — headshots, portraits, casual photos with a clear face
- Cartoons or illustrated characters with recognizable human-like facial features (eyes, nose, mouth, lips)
- 16:9 aspect ratio, person framed waist up, looking into the camera
- Even, front-facing lighting — avoid harsh shadows or backlighting
- Sufficient resolution — sharp and not pixelated
- Single person — one face per image
- Clean or neutral background preferred
What works — video requirements (v2 only)
- 1–5 minutes of continuous natural speech — conversational tone produces better results than reading a script
- Head and shoulders framing, centered, with space above the head
- Even, front-facing lighting
- Clean, uncluttered background
- Eye-level, stable camera (tripod or resting on a surface — avoid handheld)
- Clear audio — use a quiet room so lip movements align correctly
What does not work
- Faces without recognizable features — robot faces, blank or featureless characters, heavily abstract art
- Images with no face — logos, landscapes, product photos, abstract images
- Partially obscured faces — sunglasses, masks, heavy face paint, helmets
- Religious or cultural face coverings (e.g. niqab, full veil): These will not generate a standard avatar because the system requires the mouth and lips to be visible for lip-sync animation. Submit a support ticket for alternative approaches — a looping animated image without lip-sync is available as a workaround.
How will I know if my image failed?
Avatar generation is asynchronous. If your input fails validation, you will not see an error immediately. The next time you open the Space editor or test in preview mode, you will be prompted to re-upload or choose a pre-generated option. This message is only visible to you as the editor — end users never see it.
Image artefacts (green screen blurring)
Occasional blurring or halo artefacts can appear around generated avatars. Try re-uploading with a photo that has a plain, light-coloured background.