Pick Veo when dialogue is the job

Pick Veo when dialogue is the job. Veo 3.1 is the dialogue row in this catalog: native audio on, official 4K tier, clips of 4 / 6 / 8 seconds, 40 credits. The line and the mouth are born in the same pass. If this week’s clip is a sentence a person has to say, do not send that brief to a silent row and hope.

Press and model trackers have claimed Veo is strong at dialogue, including high sample-rate talk on some write-ups. Treat those as third-party claims. What we will stand on here is the catalog: Veo 3.1 is the sounded, short, cinematic talking take. Write the line in the prompt. Do not invent a kilohertz number the card does not print.

This is not a Flow review. Flow is a Veo surface. This is the weekly radio inside a multi-model studio: when the job is speech, pick the dialogue row.

What “dialogue is the job” means

The deliverable is a spoken sentence (or a two-person beat) that has to be in the file, not added later, and eight seconds is enough. A hook that is a joke. A presenter line. A product sentence you are willing to let the model voice. A scene where the sound of the room and the voice should agree by construction.

Then Veo 3.1. Prompt the words in quotes. Prompt the room. Prompt whether anyone else speaks. Leaving sound out means the model chooses. A talking shot with a surprise bed is how you fight the take in the edit.

The ladder is 4, 6, or 8 seconds. There is no 12. There is no 30. If the monologue is twenty seconds, Veo is not the whole job. You split beats, you extend a keeper on VEO 3.1 Extend (7 seconds, 720p, 40 credits, audio on), or you pick a longer sounded row (Seedance 2.5 lists 4–30 seconds) and accept that you left the dialogue row. Extending a miss continues a miss. Generate the line until the last frame is a keeper, then hop only if you need more of the same take.

4K is listed on the generate card. Use it when the plate needs it. Do not pick Veo “for 4K” on a mute product orbit you will score yourself. That is a motion job. Turbo exists.

Silent versus sounded on Veo is a rate toggle in the cost arithmetic: the family has been documented as 20 credits a second silent and 40 with native audio, times 4 / 6 / 8. If the job is dialogue, silent Veo teaches you nothing about the line. Pay the sounded rate on purpose, or pick a silent row and a voice pass on purpose. Do not split the difference.

When not to pick Veo

The clip must be silent because you own the VO. Kling 3 Turbo is silent, 1080p, 3–15s, 12 credits. Generate motion. Add the line in an editor or a TTS row. You keep the words exact. You lose model-native lips unless you lipsync later.

The line must be word-perfect and legal. Native audio improvises. If a claim, a drug, a price, or a disclaimer has to be exact, record or synthesize the VO you control. Veo is the wrong place to gamble a regulated sentence.

The SKU must hold. Start from a still on a product row. Veo T2V will invent a bottle. A talking presenter beside a real pack still is a different edit (composite, cutaway). Do not ask Veo to typeset the carton and speak.

You need Kling physics or a 15-second sounded move. Kling O3 Pro is audio on, 3–15s, 70 credits. Dialogue can live there. It is not the dialogue row in the sense of “pick this first when the brief is a short talking take.” It is the Kling door when motion and sound both matter past eight seconds.

You need 30 seconds as one thought. Seedance 2.5. Not five Veo clips of different faces cut as one speaker.

You do not have a line yet. You are testing blocking. Silent cheap drafts. Come back to Veo when the sentence exists.

Flow users: staying in Flow because “Veo is dialogue” is fine if the slate is only Veo talking shots. The moment you need captions, a SKU hold, or a 30-second pass, you were never only on the dialogue row. Keep Veo. Leave the surface.

How to run the talking take this week

Write the line. If it does not fit in 8 seconds at a human pace, cut the line. Do not ask Veo to speak 40 words.
Set 9:16 if the destination is a feed. Dialogue in 16:9 cropped for TikTok is a mouth you will crop off.
Pick Veo 3.1, audio on. Prompt the line, the speaker (wide enough to see the mouth), the room.
Generate with a retry cap. Listen on a phone. If the words are wrong, decide: regenerate, or keep picture and replace audio (an edit, not a toggle).
Picture lock. Caption the speech (auto-captions, not overlay). Overlay only the claim they did not say.
If you need more of the same take and the last frame is a keeper, extend once. If the idea changed, new generate.

Hero takes stay on Veo once you like a face. Failover to O3 Pro for “better dialogue” is a new actor. Drafts may shop. The talking hero may not.

The AI video generator will run Veo next to silent rows. The radio is yours. Quotes in a Turbo prompt will not make Turbo talk.

The Monday radio

Brief is a short spoken line → Veo 3.1, sounded, 4 / 6 / 8, aspect of the destination.
Brief is exact legal VO → silent plate + controlled voice.
Brief is pack identity → Seedance from a still, not Veo adjectives.
Brief is long Kling motion with sound → O3 Pro, not Veo’s 8-second ladder.

When dialogue is the job and eight seconds is the take, sign up and pick Veo 3.1 before you write the rest of the prompt. The line is the row.

FAQ

Is Veo the only native-audio row?

No. Seedance 2.0 / 2.5 and Kling O3 Pro also flag audio on. Veo is the dialogue row in the sense of short cinematic talking takes at 4 / 6 / 8 seconds with a 4K listed generate. Pick the others when length or Kling motion is the constraint.

Should I quote 48 kHz to a client?

Not from this catalog. The Veo 3.1 card lists native audio, durations, 4K, 40 credits. If you cite a sample-rate claim, cite the third-party article and date. Do not print it as a Versely spec.

Can I lipsync a Veo take to a new line?

You can run a lipsync family on a picture. That is a new job and a new credit. If you already know the line must be yours, skip native audio and generate silent (or record first). Do not pay Veo for a line you intend to throw away.