Writing lines Veo 3.1 can speak
Write the line in quotes, say who speaks it, and keep it short enough to fit the clip. Six steps to a spoken line, then the models that speak.
1. Pick a model that speaks
Of the six video models in this guide, Veo 3.1 speaks a line you write in quotes, lip-synced, in its own sound. Seedance 2.0, H3, Seedance 2.5 and Wan 3.0 add sound of their own too, but Thisbe lists only Veo 3.1 as speaking a line you write. Kling 3.0 Pro makes its films without sound in Thisbe.
On Auto, Thisbe picks the model for the job and says why. To be sure your line is spoken, open Models in the chat and pick one that speaks.
2. Write the line in quotes
Put the exact words inside quotation marks, as the character should say them. Words in quotes are spoken; everything outside them describes the shot.
“Close on the baker behind her counter. She looks up and says, ‘We’re out of sourdough, sorry.’”
3. Say who speaks
Name the speaker by what the camera sees: the baker, the man in the grey coat, the girl on the left. With two people in the shot, say which one speaks.
Describe the rest of the sound in sentences of its own, apart from the line: the hum of the room, a bell over the door, the rain outside.
4. Keep it short enough to fit
Read aloud, a line runs at about 183 words a minute, or three words a second. At that pace, Veo 3.1's longest clip, 8 seconds, holds at most 24 words; its 4- and 6-second clips hold 12 and 18.
Write fewer words than that: a shot needs a moment before the line and one after it. If a line runs long, cut it, or give the rest to the next shot.
5. Check the line in the Prompt
The Director writes the Prompt from what you said. The Prompt is exactly what runs, and it reaches the model as written, so your line should stand in it in quotes, word for word.
If a word is off, change it in the Prompt before anything runs.
6. Generate, then listen
Generate on the Director's card carries the price, and no render is charged until you press it. The line itself adds nothing to it: a clip costs what its model, its length and its size cost.
The take lands in the chat and on your board. If the line comes out wrong, check its quotes and its speaker in the Prompt before you try again.
A conversation, shot by shot
A clip of a few seconds holds one line well. Give each line of a conversation its own shot: the baker's line in one, the customer's answer in the next, each with its own speaker, each priced on its own card.
For a whole story told this way, see Short Films.
The models that speak
| Model | Lengths | Longest line | First take | Credits |
|---|---|---|---|---|
| Veo 3.1 Lite | 4s, 6s or 8s | 24 words | 4s at 1080p | 27 |
| Veo 3.1 Fast | 4s, 6s or 8s | 24 words | 4s at 1080p | 49 |
| Veo 3.1 Standard | 4s, 6s or 8s | 24 words | 4s at 1080p | 192 |