Replace Sora dialogue with Veo 3.1

Dialogue that used to ship on Sora — speech born with the picture, 48kHz-class native audio, a talking take rather than a silent plate plus TTS — maps to Veo 3.1: 40 credits, 4 / 6 / 8 seconds, 4K, native audio. That is the job. This is not a Sora versus Veo listicle.

Sora 2 Text to Video is 10 credits, 720p, native audio, 4 / 8 / 12 / 16 / 20s. The Sora API discontinues on 24 September 2026 per OpenAI’s Help Centre. After that date, a talking Sora prompt is not a plan. Veo 3.1 is the catalog row for short cinematic talking clips with sound in the same generate.

If the Sora take was 12, 16, or 20 seconds of speech, Veo will not hold the whole paragraph. Cut the line, or put the long take on a different native-audio row. Do not pretend an 8-second ceiling is a 20-second actor.

What “dialogue job” means here

It means the file that comes back already has a voice in it, timed to a mouth, in a room that makes sound. Native audio is the catalog flag. Veo 3.1’s flag is true. You write the line in the prompt. You do not generate a mute clip and hope a later TTS pass finds the lips.

It does not mean:

• A 20-second monologue. Veo 3.1 publishes 4, 6, and 8 seconds only. A Sora 16s speech has to become two Veo takes, a shorter script, or a non-Veo long row.
• A SKU hold. If the pack must match a JPEG, you are on reference-to-video or image-to-video, not on Veo T2V. Veo 3.1 Reference to Video is 64 credits, 8s, native audio, up to three stills. That is a different door.
• A silent draft. Kling 3 Turbo is 12 credits, 3–15s, 1080p, audio false. Cheap, and the wrong row for dialogue.
• A 4K promise from Sora. Sora 2 T2V is 720p on the Versely card. Veo 3.1 is 4K. Moving dialogue to Veo is a resolution upgrade and a duration cut.

Write the spoken line as a line. Write the room. Write “no music” if you do not want a score under the voice. A mute prompt on a native-audio row still returns a soundtrack you did not choose.

Forty credits is the catalog sticker. It is not USD. It is not Sora’s 10 credits. Dialogue is the reason you pay the Veo row instead of the 9-credit Flux 3 T2V row or the 7-credit Grok Imagine Video row.

How to move a Sora talking take onto Veo

Start with the script, not the nostalgia.

Cut to 8 seconds or less. Read the Sora line out loud with a timer. If it does not fit in 8, it does not fit on Veo 3.1. Keep the hook sentence. Move the rest to a caption, a second shot, or a longer native-audio family.

Keep the performance notes that were in the Sora prompt. Pace, accent, who is in frame, where the camera sits. Drop the parts that were Sora-only folklore (“make it Sora cinematic”). Veo does not need a vendor nickname.

Generate at 4, 6, or 8 seconds on purpose. An 8-second ask with 4 seconds of speech is a clip that waits. A 4-second ask with an 8-second paragraph is a clip that gabble-cuts. Match the tick to the line.

Inspect the mouth before you caption. Native audio is not a guarantee of a usable take. If the line is right and the mouth is not, you have a lipsync problem on an existing file, not a reason to go back to Sora. If both are wrong, change the prompt; do not extend a bad talking take.

Caption for mute playback anyway. Native audio does not play on a muted feed. Captions for silent autoplay and add captions to video are the after-file jobs. Burned-in captions are the short-form object.

Export the old Sora talking keepers before you delete the library. You may still ship those files. You may still cut them down. You should not use them as proof that Veo will return the same face. Different model, new identity, unless you lock with stills on an I2V/R2V row.

When Veo is the wrong dialogue replacement

The line is longer than 8 seconds and you will not cut it. Use Flux 3 Text to Video (9 credits, 5–20s, native audio), Seedance 2.5 (47 credits, 4–30s, 720p, native audio), or Grok Imagine Video (7 credits, 6 / 10 / 15 / 20 / 30s, 720p, native audio). Those rows can hold the paragraph. They are not Veo’s 4K dialogue job. Pick them for length, not because they “beat” Veo.

The talking head must be a specific still. Veo T2V will invent a person. Use Veo first-last (40 credits, 4 / 6 / 8s, 4K, native audio) or Veo R2V (64 credits, 8s) when you have frames. For a longer identity hold, Seedance 2.0 R2V (69 credits, 4–15s, 1080p) or Fast R2V (25 credits, 720p).

You only needed a draft of the wording. Generate silent on Kling 3 Turbo, then voice with Gemini 3.1 Flash TTS at 4 credits, then lipsync if a mouth must match. That stack is cheaper than 40 credits per Veo miss. It is not the hero dialogue row.

You needed 20 seconds of Sora because the product turn was the point. That is a SKU job, written on replace Sora SKU shots with Seedance. Dialogue was incidental room sound. Do not spend Veo 4K on a pack orbit.

Google Flow is a Veo surface, not a reason to leave Versely to “keep Veo.” The model is Veo 3.1. The catalog row is the one linked above. Flow does not add seconds to the 4 / 6 / 8 menu.

Write the Veo dialogue prompt as a shot

One speaker. One camera. One line that fits the tick.

State the words in quotes. State that the speaker is saying them on camera. State the room tone. State what must not happen (second speaker, music swell, language switch). If you need two voices, you likely need two shots or a model whose card is built for multi-speaker audio — do not hide a conversation in an 8-second Veo take and blame the row.

Do not ask Veo to “continue the Sora scene.” It cannot see the Sora file unless you are on an extend or edit row. Veo 3.1 extend is 40 credits, 7 seconds, 720p, native audio. That continues a Veo clip, not a Sora archive. For Sora files you intend to keep, export them and ship or recut. For new dialogue, start a Veo generate.

If the EU deployer rules apply, label photoreal talking takes at first exposure. The Article 50 disclosure template is the text job. Switching from Sora to Veo does not switch off disclosure.

Run the talking row from Versely.

FAQ

Why Veo 3.1 and not Flux 3 for Sora dialogue?

Veo 3.1 is the 4K, 4 / 6 / 8s native-audio talking row at 40 credits. Flux 3 T2V is 9 credits, 5–20s, native audio, and the usual landing for length. If the job is short dialogue at 4K, Veo. If the job is a 16-second spoken paragraph, Flux 3, Seedance 2.5, or Grok Imagine Video — not Veo.

Can I keep 12 / 16 / 20 second Sora speeches on Veo by extending?

Not as one Veo 3.1 generate. Extend is a 7s, 720p, 40-credit continuation of a Veo clip. It is not a 20-second T2V. Cut the line, split shots, or change family.

Does Veo 3.1 output a measured 48 kHz file on Versely?

The catalog fields on Veo 3.1 are native audio true, 4K, 4 / 6 / 8s, 40 credits. “48kHz-class” here names the production job — speech in the take, not a silent plate. Do not treat 48 kHz as a published spec field on the Versely card.