Lock a character in Imagine Video with reference images

Video 1.5 accepts image references (and voice references) so a face, a product, or a location holds from shot to shot. Up to seven reference images per generation.

Attach the references

1. Open grok.com/imagine.
2. Switch the session to video.
3. Add the stills that must stay true: one face, one outfit, one location. Each reference locks one thing.
4. If you have a speaking character, add the voice reference with the face still so both hold.
5. Write the action and the camera. Do not re-describe the face you already attached.
6. Generate. Play it. If the face drifts, drop a weaker reference and keep the two that actually match.

Image and voice references started in the US on SuperGrok Heavy and SuperGrok Plus, on grok.com/imagine and iOS, then rolled out. Text-to-video and native 1080p are generally available on grok.com/imagine, iOS, and Android even when reference slots are still missing on an account.

Prompt like a shot list

• Keep the character. Swap the scene. Or keep the scene and swap the character.
• Name only the change: walk, turn, weather, camera move.
• One clip, one action. Stack shots as separate generations from the same references.

Pitfalls

• Seven slots does not mean seven faces of the same person. Conflicting faces fight. Use the two or three that agree.
• Voice references hold a speaking voice. If the clip should be silent, say so and skip the voice file.
• The API takes reference_image_urls on grok-imagine-video-1.5. Voice reference on the API is listed as available on request.