Prompt language and spoken lines
Describe the scene in English and put the character’s words in quotes in the language they should speak. Template, example and common mistakes.
Video models were trained mostly on English descriptions, so they follow the scene, motion and camera most precisely from an English prompt. Other languages work too, but small requirements (“static camera”, “text stays unchanged”) get lost more often.
The rule
- Describe the scene in English: who is in the frame, what they do, how the camera moves.
- Put the character’s words in quotes, in the language they should speak. The model says them word for word, so the quotes must hold exactly the text you want to hear.
- Name the language: says in Spanish: "…". Without it the model may speak with an accent or switch to English.
- Don’t translate the line itself: an English line in quotes means the character speaks English.
Example
The camera is static. The teddy bear smiles, looks at the camera and says in Spanish: "¡Feliz cumpleaños, Ana! ¡Esto es para ti!" Then it slowly picks up the bouquet lying next to it and holds it out to the camera. Keep all text on the card unchanged.
Tips
- Duration. 1–2 short sentences sound natural in 8 seconds. If the character also has to do something, choose the longest duration: at 4 seconds the action looks abrupt.
- Names — tricky names can be hinted with a stress mark or spelled the way they are pronounced.
- Animating a photo or a card — add it as the “First frame” and say what must stay unchanged. Build the action from what is already in the frame and ask for slow, gentle movements.
- Only models with sound can speak: Veo 3.1, Veo 3.1 Lite, Seedance 2.0, Kling 2.6 Pro, Seedance 1.5 Pro, Veo 3.1 Fast, FLUX 3 Video, Seedance 2.5, Wan 3.0, Gemini Omni Flash.
A longer greeting
If the text is longer than 1–2 sentences, voice it with ElevenLabs v3, then animate the photo to that track with OmniHuman 1.5 or Kling Avatar 2.0 — see Talking avatar.