Gemini Omni vs Veo 3.1: Should You Migrate in May 2026?

Google's Gemini Omni page says, in five words, "Gemini Omni will replace Veo in the Gemini app."^[3] If you only read the marketing, you'd assume Veo 3.1 is sunset. It isn't. Veo 3.1 still ships on Vertex AI, on the Gemini API, and on every aggregator that lists it. Google's own April 28, 2026 update to the Veo 3.1 Gemini API docs is more recent than Omni's launch^[4].

So when you're choosing Gemini Omni vs Veo 3.1 in May 2026, you're not choosing between a deprecated model and its replacement. You're choosing between a 7-month-old per-second model with five channels and a 2-day-old per-generation model with one. The right answer depends on which Veo 3.1 channel you'd otherwise be calling, and whether the workflow needs audio, 4K, first-and-last-frame control, or multi-turn editing.

TL;DR

• Scope of the "replacement." Google replaced Veo with Omni in the Gemini app, Flow, and YouTube Shorts only^[3]. Veo 3.1 remains the documented Gemini API video model^[4].
• API access timing. Veo 3.1 Gemini API: live since October 2025. Gemini Omni public API: "coming weeks" from May 19, 2026^[5]. Through reAPI, both are callable today on the same endpoint.
• Cheapest no-audio path. Veo 3.1 Lite at $0.05 per 8-second 720p/1080p clip^[6] beats Gemini Omni's $0.216 for 8s 1080p^[7] by roughly 4x.
• Cheapest with-audio path. Gemini Omni Flash at $0.216 for 8s 1080p^[7] beats Veo 3.1 Fast Official at $1.20 (8s × $0.15/s)^[6] by 5.5x.
• 4K. Veo 3.1 Fast alt at $0.30 per 8s clip wins on cost. Gemini Omni at $0.432 per 8s wins on bundled audio.
• Duration flexibility. Gemini Omni supports 4/6/8/10s^[8]. Veo 3.1 alt locks duration at 8s; Veo 3.1 official supports 4/6/8s.
• First/last-frame interpolation. Veo 3.1 Official only. Gemini Omni doesn't expose first/last anchoring.
• Multi-turn conversational edit. Gemini Omni only^[1].
• The split. Stay on Veo 3.1 Lite for cheap silent clips. Stay on Veo 3.1 Official for first/last-frame chains. Move to Gemini Omni for any with-audio workload at 1080p or 4K, or any flow where iteration matters more than one-shot quality.

What Google actually replaced

The "Gemini Omni will replace Veo" claim runs across exactly three consumer surfaces: the Gemini app, Google Flow, and YouTube Shorts^[3]. Inside those, Omni Flash is now the default video model. Veo 3.1 disappears from the consumer UI in those products.

Outside those surfaces, Veo 3.1 is the same product it was on May 18. The Gemini API video docs at ai.google.dev still list Veo 3.1 as the supported text-to-video and image-to-video model^[4]. Vertex AI still exposes Veo 3.1 with person_generation, resize_mode, and other production controls. Aggregators routing through Google's Veo 3.1 weights are unaffected. Even Google's own April 28, 2026 announcement of new Veo 3.1 capabilities in the Gemini API (added safety filters, expanded prompt sizes) hit the public blog less than a month before Omni shipped^[4], hardly the cadence of a model being retired.

Nicole Brichtova, Google DeepMind's director of product management, told TechCrunch that Omni is "more than a Veo update" and described it as "the next step towards the progression of combining the intelligence of Gemini with the rendering capabilities of our media models."^[5] The framing is intentional. Omni is a different product category (reasoning + editing), not a drop-in upgrade for Veo's per-second video synthesis.

For the developer choosing Gemini Omni vs Veo 3.1, the takeaway is: don't migrate because the consumer UI did. Migrate when the capability or price math points at Omni for your specific workload.

The capability matrix

Three asymmetries do most of the work in the decision.

Veo 3.1's first/last-frame interpolation. Omni doesn't have an equivalent. If your pipeline anchors clips on a known start frame and a known end frame (storyboard chaining, hero-clip retakes with locked composition), Veo 3.1 Official is the only model of these two that gives you that control.

Omni's multi-turn conversational edit. Veo doesn't have an equivalent. "Make the violin invisible. Now change the camera angle." Each instruction reshapes the existing clip without losing the scene^[1]. Veo can't iterate on its own output, you have to regenerate with a new prompt every time.

Veo 3.1's audio toggle on Official. Veo Official lets you generate silent video at a discount. Omni bundles audio into every generation at one price. If you don't want audio, you can save 33% on Veo Fast Official by setting generate_audio: false; Omni doesn't expose that switch.

The five Veo 3.1 channels Omni doesn't replace

Veo 3.1 is five channels on reAPI, not one model, and each one has a different surface^[6]. Each maps to a different reason to stay or migrate.

veo3.1-lite is the budget channel. Per generation, 8-second duration locked, no audio, prompt-only (no image input). $0.05 per 8s 720p/1080p clip^[6]. Gemini Omni at $0.216 for 8s 1080p is roughly 4x more expensive on the same output spec. If your workload is "draft fast, throw most away," Lite is unbeaten.

veo3.1-fast is the workhorse alt channel. Per generation, accepts up to 3 image references, 8s fixed duration. $0.10 per 8s 720p/1080p, $0.30 per 8s 4K. Omni's three-image fusion mode covers a similar surface but at 2x the price for 1080p output.

veo3.1-quality is the high-fidelity alt channel. $0.75 per 8s 720p/1080p, $2.40 per 8s 4K. Used when face rendering, text legibility, or character consistency are make-or-break. Omni's raw quality is unconfirmed at this price tier; reviewers writing the day after launch said Omni's aggregate quality "trails" the leaderboard's frontier^[5].

veo3.1-fast-official unlocks the Vertex-grade controls: first/last-frame anchoring, generate_audio toggle, person_generation enum, 4 / 6 / 8s durations. Per second. $0.15/s for 720p/1080p with audio. None of these controls have an Omni equivalent.

veo3.1-quality-official is the production-grade tier. $0.40/s for 720p/1080p with audio. Same control surface as Fast Official, higher visual fidelity. Used for hero shots and paid creative.

When you're deciding Gemini Omni vs Veo 3.1, the right question isn't "which model wins" but "which of the five Veo 3.1 channels does this workload currently use, and does Omni's surface cover those controls?"

Migrating the request body

On reAPI, both models run on POST /api/v1/videos/generations. The migration is a request-body edit, not an endpoint change. Here's the field-by-field translation from the most common Veo 3.1 setup to Omni.

Veo 3.1 Fast (alt channel) request:

{
  "model": "veo3.1-fast",
  "prompt": "A neon city street in the rain, slow camera pan",
  "image_urls": ["https://your-cdn.com/ref.jpg"],
  "generation_type": "frame",
  "aspect_ratio": "16:9",
  "resolution": "1080p"
}

Same request, migrated to Gemini Omni:

{
  "model": "gemini-omni",
  "prompt": "A neon city street in the rain, slow camera pan",
  "image_urls": ["https://your-cdn.com/ref.jpg"],
  "duration": 8,
  "aspect_ratio": "16:9",
  "resolution": "1080p"
}

Three changes to know about. First, generation_type doesn't exist on Omni; the mode is implicit from the image_urls count (0 text-to-video, 1 image-to-video, 3 fusion)^[8]. Second, Omni accepts duration as a first-class field, where Veo alt locks it to 8s. Third, image_urls length 2 is rejected by Omni with a 400 error code 20003^[8]; Veo Fast accepts 1, 2, or 3.

If your source request was veo3.1-fast-official, the migration loses more. Drop first_frame_image, last_frame_image, generate_audio, person_generation, and resize_mode — Omni exposes none of these. If any of those were load-bearing, don't migrate.

Price math, the channels that matter

For an 8-second 1080p clip on reAPI in May 2026:

Two observations. First, if you need audio and you're not on Veo Official, Gemini Omni is the cheapest path on reAPI. Omni's $0.216 with-audio rate is 5.5x cheaper than Veo Fast Official's $1.20 and 14.8x cheaper than Veo Quality Official's $3.20. Second, if you don't need audio, Veo Lite at $0.05 stays unbeaten. Omni doesn't have a silent-mode discount because the price doesn't depend on audio.

For 4K with audio at 8s: Gemini Omni at $0.432 vs Veo Quality Official at $4.80 (8s × $0.60/s). Omni is 11x cheaper for 4K with audio. This is where the per-generation model lands hardest in Omni's favor.

When to stay on Veo 3.1

• The workflow uses first-and-last-frame anchoring to chain shots into a longer sequence
• The workflow needs person_generation set to disallow for safety compliance
• The output spec is silent video and Veo Lite's $0.05 floor matters at scale
• The pipeline is shipping today and Omni's two-day-old release introduces uncomfortable risk
• Prompt length regularly exceeds 2,000 chars (Veo allows 4,000)
• The team has a Vertex AI contract with negotiated terms

When to migrate to Gemini Omni

• You're paying for veo3.1-fast-official with audio at $1.20+ per 8s clip and the iterative-edit workflow doesn't need first-frame anchoring
• You need 4K with audio (Omni's $0.432 per 8s beats Veo Quality Official's $4.80 by 11x)
• The workflow is editing-heavy: regenerating from scratch for each tweak burns budget and breaks character consistency
• Duration flexibility (4 / 6 / 8 / 10s) matters and you're tired of the 8-second alt lock
• You want to consolidate down to one model behind one endpoint and the team can absorb the loss of Veo's official controls

FAQ

Is Veo 3.1 being deprecated?

No. Google removed Veo 3.1 from the Gemini app, Flow, and YouTube Shorts consumer surfaces and replaced it with Gemini Omni Flash^[3]. The Gemini API video docs continue to list Veo 3.1 as supported, with a documentation refresh as recent as April 28, 2026^[4]. Vertex AI's Veo 3.1 surface is unchanged.

When does the Gemini Omni public API open?

Google said developer and enterprise API access would land "in the coming weeks" after the May 19, 2026 announcement^[5]. As of this post, no public Gemini API model ID for Omni has been documented. On reAPI, Gemini Omni is callable today at gemini-omni on the standard videos endpoint, ahead of Google's direct API rollout.

Can I use both models behind one endpoint?

Yes. Both run on POST /api/v1/videos/generations on reAPI. Switching is a "model" field change. Field mappings are not identical (see the migration section above) — Omni drops first_frame_image, last_frame_image, generate_audio, person_generation, and resize_mode.

Does Gemini Omni support first-frame and last-frame anchoring?

No. Omni exposes only image fusion at cardinality 0, 1, or 3^[8]. There is no concept of a "first" or "last" frame. If your pipeline depends on locking the opening or closing composition, Veo 3.1 Official is the only model that gives you that.

Which model is cheaper for cheap iteration?

Depends on whether you need audio. Veo 3.1 Lite at $0.05 per 8s silent clip is the cheapest path for any kind of throw-away draft^[6]. If you need audio in every draft, Gemini Omni at $0.216 per 8s 1080p with audio is the cheapest with-audio rate on reAPI^[7].

Will the Gemini Omni API undercut reAPI when it ships?

Possibly. Independent forecasters project Omni Flash at $0.20–$0.60 per second of video output on Google direct, plus token-based input pricing. reAPI's $0.216 per 8s 1080p clip is roughly $0.027/s flat, which sits below the low end of the Google direct forecast. The actual gap won't be known until Google publishes official pricing.

What about Gemini Omni vs Veo 3.1 quality?

Omni Flash launched two days before this post and isn't on the Artificial Analysis Video Arena leaderboard yet^[10]. Veo 3.1 has been on the leaderboard for months and sits behind Seedance 2.0 but ahead of Sora 2 in aggregate. Early hands-on coverage of Omni called the consumer demos impressive but flagged that aggregate quality "trails" the leaderboard frontier^[5]. Treat Omni as the most novel editing surface, not the verified quality leader, until the Arena gets enough votes.

Routing both for the next quarter

You don't decide Gemini Omni vs Veo 3.1 once for the whole pipeline in May 2026. You route per request. Send draft, silent, and first-frame-anchored workloads to the right Veo 3.1 channel. Send the with-audio, 4K, and multi-turn-editing workloads to Gemini Omni Flash. Both behind one OpenAI-compatible endpoint makes that routing a 30-line config change.

If forced to standardize on one model right now, I'd hold on Veo 3.1 for any pipeline that ships to paying customers, on the strength of seven months of production track record. I'd let Gemini Omni take the new workloads where its multi-turn editing and bundled audio matter. The "Gemini Omni replaces Veo" framing is true for Google's consumer UI and misleading for everywhere else. Read the Gemini Omni vs Veo 3.1 question as "what does each channel still earn its keep on" and the migration plan writes itself.

References

1. Google. Introducing Gemini Omni. Koray Kavukcuoglu, May 19, 2026. Retrieved May 2026 from blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni
2. Google DeepMind. Gemini Omni — Model page. Retrieved May 2026 from deepmind.google/models/gemini-omni
3. Google. Gemini Omni — Video Generation overview. Retrieved May 2026 from gemini.google/overview/video-generation
4. Google. Enhanced Veo 3.1 capabilities are now available in the Gemini API. April 28, 2026. blog.google/innovation-and-ai/technology/developers-tools/veo-3-1-gemini-api
5. Rebecca Bellan. Google's Gemini Omni turns images, audio, and text into video — and that's just the start. TechCrunch, May 19, 2026. techcrunch.com/2026/05/19/googles-gemini-omni-turns-images-audio-and-text-into-video-and-thats-just-the-start
6. reAPI. Veo 3.1 — Model page (live pricing across 5 channels). Retrieved May 2026 from reapi.ai/models/veo3-1
7. reAPI. Gemini Omni — Model page (live pricing). Retrieved May 2026 from reapi.ai/models/gemini-omni
8. reAPI. Gemini Omni — API reference. Retrieved May 2026 from reapi.ai/docs/gemini-omni
9. reAPI. Veo 3.1 — API reference. Retrieved May 2026 from reapi.ai/docs/veo3-1
10. Artificial Analysis. Video Arena Leaderboard. Retrieved May 2026 from artificialanalysis.ai/video/arena

Further reading

• Google. Generate videos with Veo 3.1 in the Gemini API. ai.google.dev/gemini-api/docs/video
• reAPI. Gemini Omni vs Seedance 2.0: The 2026 Video Model Split. reapi.ai/blog/gemini-omni-vs-seedance-2-0-2026
• reAPI. Veo 3.1 vs Seedance 2.0: Picking a Video Model in 2026. reapi.ai/blog/veo-3-1-vs-seedance-2-0-2026
• reAPI. Cheapest Veo 3.1 API in 2026. reapi.ai/blog/cheapest-veo-3-1-api-2026