How to Generate AI Videos Using Multiple Character References
Open YB Video at YB.Digital’s video tool and start a new generation. Prepare a clear reference image for each character, preferably with the face visible, unobstructed, and facing the camera. Separate images work well when each person needs a distinct identity, while a reference sheet can be useful if it labels each character clearly.
Upload the images in the character references area. Add each reference as its own entry rather than combining several people into one unlabeled image. Give every entry a recognizable name, such as “Mara” or “the detective,” so you can connect the image to the right role in your prompt.
Build your scene prompt around named characters. State who appears, what each person looks like, and what they do. For example, specify that Mara stands beside the red door while the detective enters from the hallway. Avoid vague instructions such as “the woman” when two women are in the scene; names and distinct actions reduce ambiguity.
Assign each uploaded image to its matching character using the tool’s character or reference controls. Confirm that the selected reference corresponds to the name in your prompt before generating. A useful setup pairs Mara’s portrait with Mara’s scene role and the detective’s portrait with the detective’s role, rather than assigning both images to the same character.
| Scene element | What to specify | Example detail |
|---|---|---|
| Identity | Character name and reference | Mara, short dark hair |
| Position | Where the character stands | Beside the red door |
| Action | Distinct movement or behavior | The detective enters the hall |
| Appearance | Clothing or visual detail | Gray coat, blue scarf |
Set the scene composition with camera framing, setting, and interaction. Describe whether the characters face each other, cross the room, or remain in separate parts of the frame. Clear blocking gives the generator useful spatial cues, especially in a two-shot where both faces need to remain visible.
Generate a draft and inspect character consistency across the clip. Check that each person retains the intended face, clothing, and screen position, then refine the prompt if identities appear blended or a character performs the wrong action. Shorter scenes with one clear action per character are easier to diagnose than prompts with several simultaneous events.
For a cleaner next pass, make targeted edits rather than rewriting every instruction:
- Use different names for characters with similar appearances.
- Keep each character’s clothing description consistent across prompts.
- State which character speaks or moves first.
- Describe foreground and background positions separately.
YB Video from YB.Digital supports workflows where several recognizable characters share a generated scene. With references assigned to the correct roles and actions stated by name, you can build a sequence featuring, for example, two colleagues meeting at a train platform without treating them as one blended subject.