Sora 2 to Veo 3.1 migration

Veo 3.1 replaces Sora 2 on native audio, 4/6/8s clips and 4K. Rewrite Sora prose as timestamp blocks. The Sora API (sora-2, sora-2-pro, snapshots) shuts on 24 September 2026; the app and web already ended 26 April 2026. If the job is a short cinematic take with always-on sound and a 4K ceiling, the landing row is Veo 3.1, not a 15-second talking model and not a silent Turbo. The prompt grammar changes with the clock.

Duration and audio are the contract

Veo 3.1 is Google's enhanced VEO 3: text-to-video, no still required, native audio on, 4K ceiling, durations 4s, 6s, 8s, 40 credits. That closed set is the whole movie per call. Sora 2 text-to-video accepted 4s, 8s, 12s, 16s, 20s at 720p for 10 credits. The 12s, 16s and 20s jobs do not fit. Extend is a continuation from a last frame, not a way to smuggle 20 seconds into one timestamp list.

Native audio is on for both families. The port is not "add sound." The port is write the sound, because Veo will invent a mix if you do not. A Sora prompt that never named the room still often came back with a usable bed. On Veo, a picture-only sentence is how a quiet product hero returns with a trailer score. Quote the line. Name the air. Say "no music" if you mean it.

4K is a ceiling on this row, with 720p and 1080p also listed. Sora 2 T2V's published resolution is 720p. Regenerating on Veo is how you get the 4K master. Upscaling an old Sora file is a different job and a different row.

Veo 3.1 reference-to-video is not the T2V slug. It is 8s only, 64 credits, native audio, up to three reference images, still required. If the Sora job was "this face, this pack, this room," you have three files of budget. A twelve-reference Seedance stack does not belong here. A one-shot cinematic with three locked subjects does.

Sora 2 Pro Storyboard (10s / 15s / 25s, 30 credits) is the wrong object for Veo. Timestamp blocks inside 8 seconds are cuts in one clip. They are not a board. Do not port a twelve-scene storyboard into [00:00–00:08].

Timestamp blocks replace Sora prose

Sora tolerated a paragraph: "she walks in, then we cut to her hands, then a crane over the valley, she says the line." Veo 3.1 will squash that into one wandering take or spend six seconds on the first clause. Google's documented multi-shot method is timed prompt blocks that add up to the duration you actually requested.

The production shorthand:

0-2s: Medium, behind her, she pushes the vine aside.
      SFX: leaves, distant birds.
2-5s: Reverse, face, ruins behind her. She says, "This is it."
5-8s: Wide crane, she is small in the complex. No new dialogue.

Rules that survive a Sora paste:

• Ranges must sum to 4, 6, or 8 seconds. 0-2 / 2-5 / 5-12 on an 8-second call eats the ending.
• One camera instruction per block. A block that says "medium then snap zoom then handheld" is still a paragraph.
• Dialogue lives in the block where we see the mouth.
• Do not recast the person in block three. Continuity is the point of one generate.
• Do not timestamp a single take. One action, one camera, 8s or less: write one shot.

Timestamp blocks in Veo 3.1 prompts is the grammar page. This migration page is when to reach for it: Sora prompts with "then we cut" inside eight seconds or less. Longer Sora prompts are a split, an extend, or a different model (Seedance to 15s), not more timestamps.

Even blocks (0-2 / 2-4 / 4-6 / 6-8) for four beats. Uneven (0-2 / 2-5 / 5-8) when the land needs time. Six blocks in 8 seconds will smear.

Negative prompt stays for exclusions (watermarks, extra limbs, on-screen text you did not want). It is not "please do not cut." If you did not want a cut, do not timestamp.

4K is a ceiling, not a habit

Forty credits is the T2V generate. Use 4K when the plate needs it: a hero, a paid placement, a frame that will be cropped. A daily UGC test does not need 4K because Sora used to be the default. 720p and 1080p are on the same row. Pick them on purpose.

Aspect is 16:9, 9:16, or auto. Sora 9:16 jobs stay 9:16. Do not "fix" a vertical prompt into a cinematic wide because Veo can do 4K. The feed is still a phone.

Always-on audio plus 4K is a large file. Captioning is still required for mute autoplay. Burned-in captions sit in the caption pass, not inside a timestamp block. A block can name the spoken line so the mouth matches. Type on the picture is editor work.

If the Sora job was silent-on-purpose, Veo is a poor port. Native audio is on. You will get a soundtrack. Either write "room tone only, no music, no VO" and plan to replace, or pick a published silent row (Kling 3 Turbo text-to-video is silent) and stop forcing Veo to mute.

Veo 4 is not announced as of mid-2026 reporting (WaveSpeed, Evolink). Do not wait for it. Veo 3.1 (January 2026) is the current cinematic short.

When Veo is the wrong landing

Leave this row if:

• The Sora piece is longer than 8 seconds and the extra time is load-bearing, not air. Seedance 2.0 goes to 15s.
• You need phoneme lipsync in a non-English line on a referenced face. That is Seedance's vendor claim, not Veo's page.
• You need more than three reference stills.
• You need 15 seconds of physics-heavy product motion at native 4K and you are willing to use a Kling row.
• The object was Storyboard mode. Rebuild with locked stills and I2V/R2V, not timestamps.

Stay on Veo if:

• The piece is a 4 / 6 / 8s scene with dialogue or designed SFX.
• You want the 4K ceiling on a short cinematic.
• The cuts are the gag and they fit in one 8-second identity event.

Archive the Sora masters before 24 September so you are not regenerating from memory. Then rewrite the prompt as blocks, set 4 or 6 or 8, and run Veo 3.1. Open the studio for that generate. Keep Seedance and Kling for the jobs this clock cannot hold.

FAQ

Can I timestamp a 12-second Sora prompt onto Veo 3.1?

Not as one call. Veo 3.1 accepts 4s, 6s, or 8s. Twelve seconds is a split, an extend from a last frame, or a different model.

Do I have to write [00:00-00:02] the way Google's examples do?

No. 0-2s is the same instruction. The ranges must be closed, sequential, and sum to the duration you send.

Is Veo 3.1 R2V the migration default?

Only if you have up to three stills and an 8-second cinematic. It is 64 credits and still-required. Paragraph-only Sora jobs land on the 40-credit T2V slug.