# Veo

> Source: https://aiwiki.ai/wiki/veo
> Updated: 2026-10-30
> Fact-checked: 2026-10-30
> Categories: Deep Learning, Generative AI, Google DeepMind, Video Generation
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Veo." aiwiki.ai, 30 Oct 2026. https://aiwiki.ai/wiki/veo
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Veo** is a family of [text-to-video](https://aiwiki.ai/wiki/text_to_video) generative [AI models](https://aiwiki.ai/wiki/model) developed by [Google DeepMind](https://aiwiki.ai/wiki/google_deepmind). The original Veo was announced at Google I/O on May 14, 2024, and TechCrunch described it as [Google](https://aiwiki.ai/wiki/google)'s answer to [OpenAI](https://aiwiki.ai/wiki/openai)'s [Sora](https://aiwiki.ai/wiki/sora).[^2][^10] Successive versions followed: [Veo 2](https://aiwiki.ai/wiki/veo_2) (December 16, 2024), [Veo 3](https://aiwiki.ai/wiki/veo_3) (May 20, 2025), Veo 3 Fast (June 2025), [Veo 3.1](https://aiwiki.ai/wiki/veo_3_1) (October 15, 2025), a January 13, 2026 update that added 4K output, and Veo 3.1 Lite (March 31, 2026).[^3][^4][^5][^35][^36][^41] Veo 2 raised the maximum resolution to 4K, and Veo 3 added native audio: dialogue, sound effects and ambient sound generated together with the picture. Google called Veo 3 "our first video model to incorporate high-fidelity video outputs and native audio."[^3][^17] On May 19, 2026, Google announced [Gemini Omni](https://aiwiki.ai/wiki/gemini_omni), a multimodal video generation and editing model that Google says "will replace Veo in the Gemini app."[^8][^37] In the [Gemini API](https://aiwiki.ai/wiki/gemini), all remaining Veo endpoints were shut down by October 22, 2026, with Gemini Omni Flash as the listed replacement. On Google Cloud, Veo 3.1 and Veo 3.1 Fast remained available.[^9][^39]

<figure class="w-56">
<img src="/images/wd-2026-10-28-veo-logo-4f2a963aba3e-tile.png" alt="Veo logo" width="480" height="299" />
<figcaption>Veo logo. Source: <a href="https://commons.wikimedia.org/wiki/File:Google_Veo.png">Wikimedia Commons</a>, <a href="https://commons.wikimedia.org/wiki/File:Google_Veo.png#Licensing">Public domain</a>.</figcaption>
</figure>

Veo has been offered to consumers through Google's [Gemini app](https://aiwiki.ai/wiki/gemini_app) and the Flow filmmaking tool, and to developers and enterprises through the Gemini API, [Google AI Studio](https://aiwiki.ai/wiki/google_ai_studio) and [Vertex AI](https://aiwiki.ai/wiki/google_vertex_ai).[^6][^22][^24] All Veo-generated videos carry an invisible [SynthID](https://aiwiki.ai/wiki/synthid) watermark embedded in every frame.[^7] Adoption was rapid. On July 10, 2025, about seven weeks after Veo 3's launch, Google reported "over 40 million Veo 3 videos generated across the Gemini app and Flow over the last seven weeks." On October 15, 2025, it said more than 275 million videos had been generated in Flow.[^33][^23]

Veo has competed with [Sora](https://aiwiki.ai/wiki/sora) and [Sora 2](https://aiwiki.ai/wiki/sora_2) (OpenAI), [Runway Gen-3](https://aiwiki.ai/wiki/runway_gen_3) and [Runway Gen-4](https://aiwiki.ai/wiki/runway_gen_4) ([Runway](https://aiwiki.ai/wiki/runway_ml)), [Pika Labs](https://aiwiki.ai/wiki/pika_labs), and [Kling](https://aiwiki.ai/wiki/kling) ([Kuaishou](https://aiwiki.ai/wiki/kuaishou)). When Veo 3 launched in May 2025, most commercial video generators produced silent clips. TechCrunch called audio a potential "big differentiator" for Veo 3, while noting that "AI-powered sound-generating tools aren't novel." DeepMind CEO [Demis Hassabis](https://aiwiki.ai/wiki/demis_hassabis) told a press briefing, "For the first time, we're emerging from the silent era of video generation."[^32]

## What is Veo used for?

Veo generates short video clips from text prompts, still images or reference images, with synchronized audio from Veo 3 onward. Applications include content creation, filmmaking pre-visualization, [advertising](https://aiwiki.ai/wiki/advertising), education, prototyping and entertainment. A full breakdown of applications appears in the [Use cases](#use-cases) section below.

<figure class="embedvideo">
<iframe src="https://www.youtube-nocookie.com/embed/KIRh5ZnWZ5o" title="Veo demo | Compilation"></iframe>
<figcaption>"Veo demo | Compilation", official video from Google DeepMind, published October 8, 2024. <a href="https://www.youtube.com/watch?v=KIRh5ZnWZ5o">Watch on YouTube</a>.</figcaption>
</figure>

## Background

Google says Veo "builds upon years of our generative video model work, including Generative Query Network (GQN), DVD-GAN, Imagen-Video, Phenaki, WALT, VideoPoet and Lumiere."[^10] [Google Brain](https://aiwiki.ai/wiki/google_brain) and DeepMind worked on video synthesis separately before they merged into Google DeepMind in April 2023.

DeepMind's early contributions included the Generative Query Network (GQN), which learned 3D scene representations from 2D observations, and DVD-GAN, a [generative adversarial network](https://aiwiki.ai/wiki/generative_adversarial_network) for video. On the [Google Research](https://aiwiki.ai/wiki/google_research) side, Imagen Video (October 2022) applied cascaded [diffusion model](https://aiwiki.ai/wiki/diffusion_model) techniques to generate 1280x768 video at 24 frames per second from text prompts.[^11] Phenaki (October 2022) took an autoregressive [transformer](https://aiwiki.ai/wiki/transformers) approach: it could produce variable-length video from a sequence of text prompts, which allowed basic storytelling by chaining prompts.[^56]

Later projects narrowed the gap between research demonstrations and practical use. WALT (December 2023) explored latent video diffusion with a window-[attention](https://aiwiki.ai/wiki/attention) transformer.[^57] VideoPoet (December 2023) used a [large language model](https://aiwiki.ai/wiki/large_language_model) backbone to handle text-to-video, image-to-video, video stylization and video inpainting in a single architecture.[^58] [Lumiere](https://aiwiki.ai/wiki/lumiere) (January 2024) introduced a Space-Time [U-Net](https://aiwiki.ai/wiki/unet) (STUNet) architecture that generated the entire temporal duration of a video in a single pass, improving global temporal consistency.[^12]

Google said Veo combined "architecture, scaling laws and other novel techniques to improve quality and output resolution."[^10] DeepMind's 2024 Veo page added two specifics. Google had "added more details to the captions of each video in its training data." The model also used "high-quality, compressed representations of video (also known as latents)," which made it more efficient.[^48]

## Veo 1 (May 2024)

Google announced the original Veo model on May 14, 2024, at Google I/O 2024, the same day it introduced [Imagen](https://aiwiki.ai/wiki/imagen) 3.[^10] Demis Hassabis discussed it with reporters at a virtual roundtable, saying, "We're exploring features like storyboarding and generating longer scenes to see what Veo can do." [Douglas Eck](https://aiwiki.ai/wiki/douglas_eck), who leads DeepMind's research efforts in generative media, gave a press briefing.[^2] TechCrunch wrote that "Google's gunning for OpenAI's Sora with Veo."[^2]

### Capabilities at launch

Google said Veo generated "high-quality 1080p resolution videos in a wide range of cinematic and visual styles that can go beyond a minute."[^10] TechCrunch described several input modes:[^2]

- **Text-to-video**: generating clips from natural-language descriptions of scenes, actions and visual styles
- **Image-to-video**: generating video from a still image
- **Masked editing**: changing specific areas of a video
- **Longer narrative videos**: given a sequence of prompts that together tell a story, Veo could generate videos longer than a minute

Google said Veo understood cinematic terms such as "timelapse" and "aerial shots of a landscape."[^10] TechCrunch wrote that Veo had "somewhat of a grasp on physics," such as fluid dynamics and gravity, but noted that objects in its videos could disappear and reappear, and that it often got physics wrong.[^2]

### Initial availability

Veo was initially available to "select creators in private preview" in VideoFX, a new video tool in Google Labs, through a waitlist.[^10][^2] Google said it would later bring "some of Veo's capabilities to YouTube Shorts and other products."[^10] All videos Veo generated on VideoFX were watermarked with SynthID.[^7]

## Veo 2 (December 2024)

Google announced Veo 2 on December 16, 2024, as an upgrade in quality, realism and creative control.[^3] The announcement was written by Aäron van den Oord (Research Scientist, Google DeepMind) and Elias Roman (Senior Director, Product Management, Google Labs). It appeared on the Google Blog alongside an updated [Imagen 3](https://aiwiki.ai/wiki/imagen_3) and the introduction of the [Whisk](https://aiwiki.ai/wiki/whisk) experiment.[^3]

### Key improvements

| Feature | Veo 1 | Veo 2 |
|---|---|---|
| Maximum resolution | 1080p | 4K |
| Maximum duration | "Beyond a minute" | "Extended to minutes in length" |
| Physics understanding | Basic | Improved real-world physics |
| Human motion | Limited | Better nuance in movement and expression |
| Cinematography control | Moderate | Lens types, depth of field, genre cues |

Google said Veo 2 hallucinated unwanted details such as extra fingers or unexpected objects less often.[^3] It also understood "the unique language of cinematography," interpreting requests such as "18mm lens" and "shallow depth of field."[^3] Fortune reported that 4K clips were limited to eight seconds by default but could be extended to two minutes or more, according to a Google spokesperson.[^13]

### How did Veo 2 perform in benchmarks?

Google ran head-to-head tests on 1,003 prompts from Meta's MovieGenBench dataset. Human raters judged 720p clips from Veo 2 against output from Meta [Movie Gen](https://aiwiki.ai/wiki/movie_gen), Kling v1.5, MiniMax and OpenAI's Sora Turbo.[^14] Google reported that Veo 2 came out ahead on both overall preference and prompt adherence. Against Sora Turbo, 59% of raters preferred Veo 2 and 27% Sora Turbo. Fortune reported that the preference for Veo 2 slipped slightly below 50% only against Kling v1.5, and its headline said Veo 2 "trounces" Sora.[^13][^14]

Google ran these tests with its own methodology, and independent benchmarks may rank the models differently.

### Availability

On December 16, 2024, Google brought Veo 2 to VideoFX in Google Labs and expanded the number of users who could access it, though a waitlist remained. Google said it was "slowly rolling it out via VideoFX, YouTube and Vertex AI."[^3][^15] Fortune noted that VideoFX was capped at 720p.[^13] On February 13, 2025, YouTube integrated Veo 2 into Dream Screen in Shorts and added standalone Veo 2 video clips, first in the United States, Canada, Australia and New Zealand.[^45] On April 9, 2025, Google released `veo-2.0-generate-001` as a generally available model in the Gemini API.[^34] On April 15, 2025, Veo 2 came to Gemini Advanced subscribers in the Gemini app, generating eight-second 720p clips, and to the Whisk Animate feature.[^44]

## Veo 3 (May 2025)

Veo 3 was announced on May 20, 2025, at Google I/O.[^4][^16] Its headline feature was native audio generation. Google later described it as "our first video model to incorporate high-fidelity video outputs and native audio."[^17]

### Native audio generation

Veo 3 generates audio together with the video rather than relying on a separate audio model or a post-production step.[^4][^17] Google's Gemini API post says it natively generates dialogue, effects and music and "synchronizes it with video in a single pass."[^17] The audio falls into three broad categories:

| Audio type | Description | Examples |
|---|---|---|
| Dialogue | Character speech with lip synchronization | Conversations, narration, monologues |
| Sound effects | Sounds matching on-screen actions | Footsteps, door creaking, water splashing |
| Ambient noise and music | Background sound that sets the scene | City traffic, wind, ocean waves, music |

Hassabis told reporters, "For the first time, we're emerging from the silent era of video generation."[^32] According to Google's technical report, the diffusion process is applied jointly to audio and video latents, so the model generates both together rather than adding audio afterwards.[^25] Prompts can include dialogue in quotation marks and cues for sound effects and ambient noise.[^24]

### Additional improvements

Google also cited gains in realism, physics and prompt adherence. Its Veo page says Veo 3 "delivers high quality, excelling in physics, realism and prompt adherence."[^1]

The output format changed over time. At its Gemini API launch on July 17, 2025, Veo 3 produced eight-second 720p clips in 16:9. In September 2025 Google added 1080p output and the 9:16 vertical format, and in October 2025 it added clip lengths of 4 and 6 seconds as well as 8.[^34][^49] Through 2025, Veo 3's options in the Gemini API topped out at 1080p; 4K output arrived with the January 2026 Veo 3.1 update.[^24][^35]

### Viral impact and rapid adoption

Veo 3 drew wide public attention, and demo videos spread quickly on social media. Users posted fake news-anchor clips within days of the launch.[^19] TIME reported that "in the first week of Veo 3's release, online users posted fake news segments in multiple languages," including an anchor announcing the death of J.K. Rowling.[^20] On July 10, 2025, Google reported "over 40 million Veo 3 videos generated across the Gemini app and Flow over the last seven weeks." The same day, it launched photo-to-video in the Gemini app, which turns photos into eight-second clips with sound, for Google AI Pro and Ultra subscribers in select countries.[^33]

### Veo 3 Fast

Veo 3 Fast is a variant optimized for speed and cost. It came to the Gemini app on June 9, 2025. 9to5Google reported that it was twice as fast to generate as Veo 3 while keeping the same 720p resolution, and that Google AI Pro subscribers ($19.99 per month) got three Veo 3 Fast generations per day.[^41] Veo 3 Fast came to the Gemini API in preview on July 31, 2025, at $0.40 per second of video with audio, against $0.75 for Veo 3.[^50] On September 8, 2025, Google cut Gemini API prices: Veo 3 fell from $0.75 to $0.40 per second and Veo 3 Fast from $0.40 to $0.15. The Gemini API changelog records both models becoming generally available there on September 9.[^49][^34] On Vertex AI, Veo 3 and Veo 3 Fast reached general availability on July 29, 2025.[^39]

### Availability

Veo 3 launched in the Gemini app for U.S. subscribers to the $249.99-per-month Google AI Ultra plan, and in private preview on Vertex AI.[^32][^43][^4] Flow, Google's AI filmmaking tool introduced at I/O 2025 and "custom-designed for Veo," launched the same day for Google AI Pro and Ultra subscribers in the United States. Pro included Flow features with 100 generations per month, and Ultra had early access to Veo 3 with audio.[^22] Developers got Veo 3 in paid preview through the Gemini API on July 17, 2025, at $0.75 per second of video with audio.[^17][^34] On September 16, 2025, YouTube launched a custom version of Veo 3 Fast in Shorts, free for creators, generating 480p clips with sound, first in the United States, the United Kingdom, Canada, Australia and New Zealand.[^46]

## Veo 3.1 (October 2025)

Veo 3.1 and Veo 3.1 Fast were released on October 15, 2025. They were available in paid preview in the Gemini API and also came to Vertex AI, the Gemini app and Flow.[^5][^23] Google said Veo 3.1 "brings richer audio, more narrative control, and enhanced realism that captures true-to-life textures."[^23] In the Gemini API, "Veo 3.1 is the same price as Veo 3."[^5]

### What is new in Veo 3.1?

- **Reference images**: users can provide up to three reference images of a character, object or scene to guide generation (the "Ingredients to Video" feature in Flow). This helps keep characters consistent across shots or apply a specific style.[^5][^23] The Gemini API docs listed reference images as "Veo 3.1 only," so the feature was not available with Veo 3.1 Fast there.[^24][^59]
- **Scene extension**: new clips continue a previous video, "even lasting for a minute or more," with each new clip generated "based on the final second of the previous clip."[^5] In the Gemini API, Veo 3.1 could extend a Veo-generated video by 7 seconds up to 20 times, producing up to 148 seconds of video at 720p.[^24]
- **First and last frame**: users provide a starting and an ending image, and the model generates the transition between them, "complete with accompanying audio."[^5]
- **Audio across Flow features**: Flow's Ingredients to Video, Frames to Video and Extend features gained audio "for the first time."[^23]
- **Object insertion in Flow**: an "Insert" tool adds new elements to a scene, handling shadows and lighting. Google said object removal would come "Soon" and that adding and removing objects was not available in the Gemini API.[^23]
- **Image-to-video**: Google cited improved prompt adherence and audiovisual quality when turning images into videos.[^23]

In the Gemini API, Veo 3.1 generated videos of 4, 6 or 8 seconds at 720p, or 8 seconds at 1080p, in 16:9 or 9:16.[^24] Veo 3.1 and Veo 3.1 Fast became generally available on Vertex AI on November 17, 2025.[^38][^39]

### January 2026 update

On January 13, 2026, Google updated Veo 3.1's Ingredients to Video mode with livelier, more expressive output, better identity consistency for characters across scenes, and native vertical (9:16) output.[^35] The update also improved 1080p output, which had been available in the Gemini API since Veo 3.1's launch for eight-second clips, and added 4K.[^24][^35] Google's sources describe the 4K output in two ways. The announcement called it "state-of-the-art upscaling to 1080p and 4K resolution," available only in Flow, the Gemini API and Vertex AI.[^35] The Gemini API changelog recorded "4k output resolutions for Veo" that day, and the Gemini API docs said Veo 3.1 "can also directly generate 720p, 1080p or 4k videos."[^34][^59] The update also brought Ingredients to Video to YouTube Shorts and the YouTube Create app for the first time.[^35]

### Veo 3.1 Lite

On March 31, 2026, Google released Veo 3.1 Lite in the Gemini API and Google AI Studio, calling it "Google's most cost-effective video model." Google said it ran "at less than 50% of the cost of Veo 3.1 Fast, but with the same speed."[^36] It supports text-to-video and image-to-video at 16:9 and 9:16, at 720p and 1080p, in clips of 4, 6 or 8 seconds.[^36] In the same post, Google said it would reduce the price of Veo 3.1 Fast on April 7, 2026.[^36] Veo 3.1 Lite entered public preview on Vertex AI on April 2, 2026.[^38]

## Successor: Gemini Omni

At Google I/O on May 19, 2026, Google DeepMind introduced Gemini Omni and its first model, Gemini Omni Flash, as a model that can "create anything from any input," starting with video. It accepts combinations of images, audio, video and text as input, and users can edit videos through conversation.[^8] Gemini Omni Flash began rolling out that day to Google AI Plus, Pro and Ultra subscribers in the Gemini app and Flow, and that week at no cost in YouTube Shorts and the YouTube Create app. All Omni videos carry SynthID.[^8] Google's Gemini app video page now says that "Gemini Omni will replace Veo in the Gemini app."[^37]

For developers, Gemini Omni Flash entered public preview in the Gemini API on June 30, 2026, generating 3- to 10-second 720p videos. On August 27, 2026, it became generally available as `gemini-omni-1.1-flash`, with options up to 4K, where 1080p and 4K outputs are made by upscaling.[^34] The Gemini API then retired its Veo models:[^9]

| Gemini API model | Released | Shut down |
|---|---|---|
| `veo-3.0-generate-preview`, `veo-3.0-fast-generate-preview` | July 2025 | November 12, 2025 |
| `veo-2.0-generate-001` | April 9, 2025 | June 30, 2026 |
| `veo-3.0-generate-001`, `veo-3.0-fast-generate-001` | September 9, 2025 | June 30, 2026 |
| `veo-3.1-generate-preview`, `veo-3.1-fast-generate-preview` | October 15, 2025 | October 22, 2026 |
| `veo-3.1-lite-generate-preview` | March 31, 2026 | October 22, 2026 |

The deprecations page lists `gemini-omni-1.1-flash` as the replacement for the preview models, and `gemini-omni-1.1-flash` or the generally available models on the Gemini Enterprise Agent Platform for the stable Veo 2 and Veo 3 models.[^9]

## Technical approach

Google DeepMind has not published a full research paper on Veo's architecture. For Veo 3 it published a short technical report and a model card.[^25]

### Latent diffusion transformer

According to the Veo 3 technical report, "Veo 3 uses latent diffusion, in which the diffusion process is applied jointly to the temporal audio latents, and the spatio-temporal video latents."[^25] Video and audio are encoded by separate [autoencoders](https://aiwiki.ai/wiki/autoencoder) into compressed latent representations "in which learning can take place more efficiently than with the raw pixels or waveform."[^25] Working in a compressed [latent space](https://aiwiki.ai/wiki/latent_space) makes the expensive diffusion process far cheaper than working on raw frames, a technique DeepMind already described for the original Veo.[^48]

### Diffusion process

Veo follows the standard diffusion paradigm. During training, the report says, "a transformer-based denoising network is optimized to remove noise from noisy latent vectors." At sampling time, the network is applied iteratively to Gaussian noise to produce a video, guided by the text prompt or image conditioning.[^25] A decoder then turns the latents back into frames. Google has not published parameter counts or layer-level details.

### Conditioning and prompt understanding

Rich text conditioning matters a great deal to Veo's output quality. For the first Veo, Google said it added more detailed captions to the training videos.[^48] For Veo 3, Google says it captioned training data "at different levels of detail, leveraging multiple Gemini models," and filtered out unsafe captions and personally identifiable information.[^25]

The model understands specialized cinematic terminology. Google's Veo 2 announcement invited users to "ask it for a genre, specify a lens, suggest cinematic effects," giving examples such as a low-angle tracking shot, a close-up, an "18mm lens" and "shallow depth of field."[^3]

Starting with Veo 3, the conditioning system was extended to audio. The model generates synchronized dialogue, sound effects and ambient audio from the same text prompt, producing a single audiovisual output.[^25]

## Availability

Veo moved from a waitlisted Labs experiment in 2024 to a set of consumer, developer and enterprise products. In 2026 Google began moving its consumer products and the Gemini API to Gemini Omni.

### VideoFX

VideoFX was the first consumer-facing tool for Veo, launched with the original model in May 2024 as part of Google Labs.[^10] It offered a simple web interface for text-to-video generation. With Veo 2, VideoFX was capped at 720p, although the model supported higher resolutions through other channels.[^13] Google later described Flow as "the evolution of VideoFX."[^22]

### Flow

**Flow** is Google's AI filmmaking tool, introduced at Google I/O on May 20, 2025.[^22] Google called it "the only AI filmmaking tool custom-designed for Google's most advanced models," naming Veo, Imagen and Gemini. At launch its features included Camera Controls, Scenebuilder (for editing and extending shots "with continuous motion and consistent characters"), Asset Management, and Flow TV, a showcase of Veo clips that shows the prompts used.[^22] By October 15, 2025, Google said more than 275 million videos had been generated in Flow.[^23]

Flow is metered by credits. On August 18, 2025, Google doubled monthly Flow credits for Google AI Ultra subscribers from 12,500 to 25,000, saying users had created 100 million videos in Flow since May.[^40] As of October 2026, Flow's help page lists 50 free credits per day for everyone, plus monthly allocations of 200 credits for Google AI Plus, 1,000 for Pro, 10,000 for "Ultra $100" and 25,000 for "Ultra $200." Veo 3.1 Lite, Fast and Quality generations cost 10, 20 and 100 credits for non-Ultra subscribers, alongside Gemini Omni Flash.[^26]

### Google AI Studio and the Gemini API

Developers accessed Veo through Google AI Studio and the Gemini API, paying per second of generated video only for successfully generated videos. Veo was available only on the paid tier.[^21] The Gemini API carried Veo 2 (from April 2025), Veo 3 and Veo 3 Fast (from July 2025), Veo 3.1 and Veo 3.1 Fast (from October 2025), and Veo 3.1 Lite (from March 2026).[^34] All Veo models had been shut down there by October 22, 2026, and as of October 2026 the Gemini API pricing page lists Gemini Omni Flash and no Veo models.[^9][^21]

### Vertex AI

For enterprise customers, Veo is available on Vertex AI, [Google Cloud](https://aiwiki.ai/wiki/google_cloud)'s managed [machine learning](https://aiwiki.ai/wiki/machine_learning) platform.[^6] On April 22, 2026, Vertex AI became part of the Gemini Enterprise Agent Platform.[^38] The platform's lifecycle table lists these release and retirement dates for Veo:[^39]

| Model | Released | Retirement |
|---|---|---|
| `veo-2.0-generate-001` | May 27, 2025 | June 30, 2026 |
| `veo-3.0-generate-001`, `veo-3.0-fast-generate-001` | July 29, 2025 | June 30, 2026 |
| `veo-3.1-generate-001`, `veo-3.1-fast-generate-001` | November 17, 2025 | November 17, 2026 or later |

Google announced the deprecation of the Veo 2 and Veo 3 endpoints on March 24, 2026, recommending a move to Veo 3.1.[^38] Gemini Omni 1.1 Flash entered public preview on the Agent Platform on August 27, 2026.[^38]

### Gemini app

Veo 2 came to Gemini Advanced subscribers on April 15, 2025, and Veo 3 to Google AI Ultra subscribers on May 20, 2025.[^44][^43] Pro subscribers received three Veo 3 Fast videos per day from June 9, 2025.[^41] In January 2026, the Gemini Apps limits page listed up to three videos per day with Veo 3.1 Fast for Google AI Pro and up to five per day with Veo 3.1 for Google AI Ultra.[^42] Since May 19, 2026, the Gemini app has generated video with Gemini Omni, which Google says replaces Veo there.[^37][^43]

## SynthID watermarking

All videos generated by Veo are watermarked with SynthID, a Google DeepMind [AI watermarking](https://aiwiki.ai/wiki/watermarking) technology that embeds an imperceptible watermark "directly into the pixels of every video frame."[^7] The watermark is invisible to viewers but detectable by Google's tools. Google says it is designed to stand up to "cropping, adding filters, changing frame rates, or lossy compression."[^27] Discussing its text watermark, Google says SynthID "isn't built to directly stop motivated adversaries like cyberattackers or hackers from causing harm," so it is not a defense against determined [adversarial attacks](https://aiwiki.ai/wiki/adversarial_attack).[^7]

On May 20, 2025, Google said "over 10 billion pieces of content have already been watermarked with SynthID," covering content from Gemini, Imagen, Lyria and Veo. The same day, it announced a SynthID Detector portal for checking content.[^47] Google's SynthID page says the detector can check whether an image, video or audio file was made with AI "from Google or our partners."[^27]

Beyond watermarking, Google said in 2024 that Veo videos are "passed through safety filters and memorization checking processes that help mitigate privacy, copyright and bias risks."[^48]

## How does Veo compare with competitors?

The [AI video generation](https://aiwiki.ai/wiki/ai_video_generation) field has grown much more competitive since 2024. Native audio, Veo 3's distinguishing feature at launch, soon spread to rival models:

| Model | Developer | Date | Audio | Source note |
|---|---|---|---|---|
| Veo 3 | [Google DeepMind](https://aiwiki.ai/wiki/google_deepmind) | May 20, 2025 | Native audio at launch | Google's "first video model" with native audio[^17] |
| Sora 2 | [OpenAI](https://aiwiki.ai/wiki/openai) | September 30, 2025 | Native audio at launch | Launched with an invite-only Sora social app[^51] |
| Kling Video 2.6 | [Kuaishou](https://aiwiki.ai/wiki/kuaishou) | December 3, 2025 | "Simultaneous audio-visual generation" | Chinese and English speech, clips up to 10 seconds[^53] |
| Runway Gen-4.5 | [Runway](https://aiwiki.ai/wiki/runway_ml) | December 2025 | Native audio added December 11, 2025 | Update released with Runway's first world model, GWM-1[^54] |

In Google's December 2024 evaluation on 1,003 MovieGenBench prompts, human raters preferred Veo 2 over Sora Turbo, Meta Movie Gen, Kling v1.5 and MiniMax overall, with the narrowest margin against Kling v1.5.[^13][^14] Rankings on public leaderboards have shifted quickly. TechCrunch reported that Runway's Gen-4.5, released in December 2025, "surpassed both Google and OpenAI on the Video Arena leaderboard."[^54]

### How does Veo differ from Sora?

Veo and Sora were the two best-known text-to-video systems from frontier AI labs. TIME wrote that, unlike the outputs of earlier generators such as the original Sora, "Veo 3 videos can include dialogue, soundtracks and sound effects."[^20] Fortune reported in December 2024 that Sora Turbo was limited to 1080p clips of up to 20 seconds, while Veo 2 reached 4K.[^13] OpenAI's Sora 2, released on September 30, 2025, added audio.[^51] Veo marks its videos with SynthID, an in-pixel watermark, while Sora used a visible watermark and C2PA metadata, according to Fortune.[^7][^13] In March 2026 OpenAI announced that it would discontinue Sora, closing the web and app versions on April 26, 2026 and the Sora API on September 24, 2026.[^52]

## Limitations and controversies

### Known limitations

Veo has several documented limitations:

- **Duration constraints**: a single generation is limited to about 8 seconds. Longer videos require scene extension, which in the Gemini API ran at 720p and topped out at 148 seconds.[^24]
- **Complex scenes**: The Decoder reported that, when announcing Veo 2, Google acknowledged that the model still struggled with complex scenes and motion sequences.[^14]
- **Text rendering**: like most video models, Veo struggles to render legible text, signs or written content within frames. Google's Veo 3 technical report says that, "while much better than previous Veo models, Veo 3 is still poor at generating text."[^25]
- **Cost**: at $0.40 per second for 720p or 1080p output from Veo 3 or Veo 3.1 (the Gemini API rate from September 2025, also the Vertex AI rate), a minute of video costs $24, or $36 at Veo 3.1's 4K rate of $0.60 per second.[^6][^21]

### Subtitle artifacts

In July 2025, *MIT Technology Review* reported that Veo 3 was still adding garbled, nonsensical captions to videos more than a month after launch, "even when users explicitly ask for no captions at all."[^28] Advertising creative director Mona Weiss told the magazine that "if you're creating a scene with dialogue, up to 40% of its output has gibberish subtitles that make it unusable." Google Labs' Josh Woodward had posted on June 9, 2025 that Google had developed fixes. An expert quoted by the magazine said the training data was likely to include [YouTube](https://aiwiki.ai/wiki/youtube) videos, vlogs and TikTok edits with embedded subtitles.[^28]

### Training-data controversy

In June 2025, *CNBC* reported that Google was turning to its catalog of 20 billion YouTube videos to train AI models including Gemini and Veo 3, citing a person not authorized to speak publicly. Google confirmed it uses "a subset" of YouTube videos for training and said it honors specific agreements with creators and media companies.[^29] Multiple leading creators and intellectual-property professionals told CNBC they had not been informed that their content could be used this way. CNBC noted a September blog post in which YouTube said content could be used to improve "the product experience … including through machine learning and AI applications," and reported that users have no way to opt out. Training on just 1% of the catalog would amount to 2.3 billion minutes of content, which experts told CNBC is more than 40 times the training data used by competing AI models. Google includes an indemnification clause for its generative AI products, including Veo, for users facing copyright challenges.[^29]

### Misinformation and deepfake concerns

Veo 3's realism, combined with native audio, quickly raised concerns about misinformation. In a June 3, 2025 investigation, *TIME* reported that it had used Veo 3 to make realistic clips including an election worker shredding ballots. After TIME contacted Google, the company said it would begin adding a visible watermark to Veo 3 videos.[^20] The AI Incident Database records a case in which Philippine Senator Ronald "Bato" dela Rosa and Davao City Mayor Sebastian Duterte reportedly shared a Veo 3-generated street-interview video of fake students opposing Vice President Sara Duterte's impeachment.[^30]

On July 1, 2025, Media Matters for America reported that racist and antisemitic videos apparently made with Veo were garnering millions of views on TikTok.[^55] Ars Technica's Ryan Whitwam wrote that "in a perfect world, Veo 3 would refuse to create these videos, but vagueness in the prompt and the AI's inability to understand the subtleties of racist tropes (i.e., the use of monkeys instead of humans in some videos) make it easy to skirt the rules."[^31]

### Quality concerns at the low end

In a Gizmodo article published two days after the launch, James Pero described users steering Veo 3 toward what he called "AI slop." His examples were fake man-on-the-street interviews, iPhone unboxing videos, stand-up comedy clips and simulated game streams.[^18]

## Use cases

Veo has found applications across a range of creative and professional domains:

- **Content creation**: short-form video for YouTube Shorts, Instagram Reels and [TikTok](https://aiwiki.ai/wiki/tiktok), including through YouTube's own Veo integrations in Shorts[^45][^46]
- **Filmmaking and pre-visualization**: directors and producers use Flow to prototype scenes, test camera angles, explore lighting and visualize narrative sequences before production. Google cited Promise Studios, which used Veo 3.1 for generative storyboarding and previsualization.[^5]
- **Advertising and marketing**: drafting video ads, product demos and concept videos
- **Education**: explanatory videos, historical recreations and visualizations of scientific concepts
- **Prototyping and product design**: mock-up videos to visualize user experiences, app flows or physical product concepts
- **Accessibility**: lower barriers to video creation for individuals and small teams without traditional production resources
- **Gaming and entertainment**: Google cited the game studio Volley, which used Veo 3 for in-game cut-scenes, and Latitude, which experimented with Veo 3.1 in its narrative game engine.[^17][^5]

## How much does Veo cost?

Google has changed Veo's prices several times. The prices below were published by Google on the dates shown.

### Gemini API (developer pricing)

All prices are per second of generated video with audio (for Veo 2, video only).

| Date | Model | 720p / 1080p | 4K |
|---|---|---|---|
| April 2025 to May 2026 (launch post and archived pages) | [Veo 2](https://aiwiki.ai/wiki/veo_2) | $0.35 | Not offered |
| July 17, 2025 | [Veo 3](https://aiwiki.ai/wiki/veo_3) (preview) | $0.75 | Not offered |
| July 31, 2025 | Veo 3 Fast (preview) | $0.40 | Not offered |
| September 8, 2025 | Veo 3 / Veo 3 Fast | $0.40 / $0.15 | Not offered |
| October 15, 2025 | [Veo 3.1](https://aiwiki.ai/wiki/veo_3_1) / Veo 3.1 Fast | $0.40 / $0.15 | Not listed |
| February 2026 | Veo 3.1 / Veo 3.1 Fast | $0.40 / $0.15 | $0.60 / $0.35 |
| March 31, 2026 | Veo 3.1 Lite | $0.05 (720p), $0.08 (1080p) | Not supported |
| May 2026 (after the April 7 cut) | Veo 3.1 Fast | $0.10 (720p), $0.12 (1080p) | $0.30 |

Sources: Google's launch posts and archived Gemini API pricing pages.[^60][^17][^50][^49][^5][^21][^36] The May 2026 pricing page also listed Veo 3 Fast at $0.10, $0.12 and $0.30 (4K), while Veo 3 stayed at $0.40 with no 4K price.[^21] Charges applied only to successfully generated videos, and there was no free tier.[^21] Veo models were removed from the Gemini API in 2026. As of October 2026, its successor, Gemini Omni Flash, is billed per token, at an effective price of about $0.10 per second of 720p video.[^9][^21]

### Consumer subscriptions

| Date | Plan | Monthly price | Veo access |
|---|---|---|---|
| May 20, 2025 | Google AI Ultra | $249.99 | Veo 3 in the Gemini app and Flow (U.S.)[^32][^22] |
| May 20, 2025 | Google AI Pro | $19.99 | Key Flow features, 100 generations per month[^22][^41] |
| June 9, 2025 | Google AI Pro | $19.99 | Three Veo 3 Fast videos per day in the Gemini app[^41] |
| August 18, 2025 | Google AI Ultra | $249.99 | Monthly Flow credits doubled from 12,500 to 25,000[^40] |
| January 2026 | Google AI Pro / Ultra | | Up to 3 videos per day (Veo 3.1 Fast) / up to 5 per day (Veo 3.1) in the Gemini app[^42] |
| May 19, 2026 | Google AI Ultra (new tier) | From $100 | Video generation with Gemini Omni[^43] |

### Vertex AI (enterprise)

On Vertex AI, Veo 2 was priced at $0.50 per second. In July 2025, Veo 3 cost $0.75 per second with audio and $0.50 without. By October 2025 the price list showed $0.40 and $0.20 for Veo 3, and $0.15 and $0.10 for Veo 3 Fast.[^6] As of October 2026, the Agent Platform (formerly Vertex AI) price list shows Veo 3.1 at $0.40 per second with audio at 720p and 1080p and $0.60 at 4K. Veo 3.1 Fast is $0.10, $0.12 and $0.30, and Veo 3.1 Lite $0.05 and $0.08, with lower prices for video without audio.[^6]

## When was each Veo version released? (Timeline)

| Date | Event |
|---|---|
| May 14, 2024 | Veo announced at Google I/O 2024; available to select creators in VideoFX through a waitlist |
| December 16, 2024 | [Veo 2](https://aiwiki.ai/wiki/veo_2) announced with up to 4K resolution and improved physics; rolled out in VideoFX |
| February 13, 2025 | Veo 2 comes to Dream Screen in YouTube Shorts |
| April 9, 2025 | Veo 2 generally available in the Gemini API |
| April 15, 2025 | Veo 2 comes to Gemini Advanced subscribers in the Gemini app |
| May 20, 2025 | [Veo 3](https://aiwiki.ai/wiki/veo_3) announced at Google I/O 2025 with native audio; Flow introduced |
| June 9, 2025 | Veo 3 Fast arrives in the Gemini app; Pro gets three Veo 3 Fast videos per day |
| June 19, 2025 | CNBC reports that Google uses YouTube videos to train Veo 3 and Gemini |
| July 10, 2025 | Google reports over 40 million Veo 3 videos; photo-to-video launches in the Gemini app |
| July 15, 2025 | *MIT Technology Review* documents Veo 3's subtitle problem |
| July 17, 2025 | Veo 3 in paid preview in the Gemini API at $0.75 per second |
| July 29, 2025 | Veo 3 and Veo 3 Fast generally available on Vertex AI |
| September 8-9, 2025 | Veo 3 and Veo 3 Fast generally available in the Gemini API; prices cut to $0.40 and $0.15 per second; 1080p and 9:16 added |
| September 16, 2025 | Custom Veo 3 Fast launches free in YouTube Shorts |
| October 15, 2025 | [Veo 3.1](https://aiwiki.ai/wiki/veo_3_1) and Veo 3.1 Fast released with scene extension and first-and-last-frame control, plus reference images (Veo 3.1 only in the Gemini API); Flow passes 275 million videos |
| November 17, 2025 | Veo 3.1 generally available on Vertex AI |
| January 13, 2026 | Veo 3.1 update adds native vertical Ingredients to Video, improves 1080p and adds 4K output |
| March 31, 2026 | Veo 3.1 Lite released in the Gemini API |
| April 7, 2026 | Veo 3.1 Fast price cut in the Gemini API |
| April 22, 2026 | Vertex AI becomes part of Gemini Enterprise Agent Platform |
| May 19, 2026 | [Gemini Omni](https://aiwiki.ai/wiki/gemini_omni) announced at Google I/O 2026; set to replace Veo in the Gemini app |
| June 30, 2026 | Veo 2 and Veo 3 stable models shut down in the Gemini API and retired on Vertex AI |
| October 22, 2026 | All Veo 3.1 preview models, including Lite, shut down in the Gemini API |

## Legacy and current status

As of October 2026, Google DeepMind's Veo page still presents Veo 3.1 as "our leading video generation model."[^1] The video model Google announced at I/O 2026 was not a new Veo version but Gemini Omni, which Google says replaces Veo in the Gemini app and which the Gemini API lists as the replacement for its retired Veo models.[^8][^9][^37] Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite remain available in Flow and on Google Cloud's Agent Platform, where the Veo 3.1 models carry a retirement date of "November 17, 2026 or later."[^26][^39]

Within two years, Veo went from a waitlisted Labs experiment in May 2024 to a production video platform with native audio, reference-image conditioning and availability across Google's consumer, developer and cloud products, with SynthID watermarking on every video. Its rise also brought sharp [generative AI](https://aiwiki.ai/wiki/generative_ai) disputes over training-data provenance, misinformation and deepfakes, and the economic relationship between AI labs and the creators whose videos train their models.

## See also

- [Veo 2](https://aiwiki.ai/wiki/veo_2)
- [Veo 3](https://aiwiki.ai/wiki/veo_3)
- [Veo 3.1](https://aiwiki.ai/wiki/veo_3_1)
- [Gemini Omni](https://aiwiki.ai/wiki/gemini_omni)
- [Sora](https://aiwiki.ai/wiki/sora)
- [Sora 2](https://aiwiki.ai/wiki/sora_2)
- [Runway Gen-3](https://aiwiki.ai/wiki/runway_gen_3)
- [Runway Gen-4](https://aiwiki.ai/wiki/runway_gen_4)
- [Kling](https://aiwiki.ai/wiki/kling)
- [Gemini app](https://aiwiki.ai/wiki/gemini_app)
- [Vertex AI](https://aiwiki.ai/wiki/google_vertex_ai)
- [SynthID](https://aiwiki.ai/wiki/synthid)
- [Lumiere](https://aiwiki.ai/wiki/lumiere)
- [Diffusion model](https://aiwiki.ai/wiki/diffusion_model)
- [Text to video](https://aiwiki.ai/wiki/text_to_video)
- [Google DeepMind](https://aiwiki.ai/wiki/google_deepmind)

## References

[^1]: Google DeepMind. "Veo." Models page (accessed October 30, 2026). https://deepmind.google/models/veo/
[^2]: Kyle Wiggers. "Google Veo, a serious swing at AI-generated video, debuts at Google I/O 2024." TechCrunch, May 14, 2024. https://techcrunch.com/2024/05/14/google-veo-a-serious-swing-at-ai-generated-video-debuts-at-google-io-2024/
[^3]: Aäron van den Oord and Elias Roman. "State-of-the-art video and image generation with Veo 2 and Imagen 3." Google Blog, December 16, 2024. https://blog.google/technology/google-labs/video-image-generation-update-december-2024/
[^4]: Google Cloud Blog. "Announcing Veo 3, Imagen 4, and Lyria 2 on Vertex AI." May 21, 2025. https://cloud.google.com/blog/products/ai-machine-learning/announcing-veo-3-imagen-4-and-lyria-2-on-vertex-ai
[^5]: Google Developers Blog. "Introducing Veo 3.1 and new creative capabilities in the Gemini API." October 15, 2025. https://developers.googleblog.com/introducing-veo-3-1-and-new-creative-capabilities-in-the-gemini-api/
[^6]: Google Cloud. "Vertex AI pricing" (now "Agent Platform pricing," accessed October 30, 2026). https://cloud.google.com/vertex-ai/generative-ai/pricing ; archived snapshots: July 5, 2025 https://web.archive.org/web/20250705053458/https://cloud.google.com/vertex-ai/generative-ai/pricing ; October 2, 2025 https://web.archive.org/web/20251002151121/https://cloud.google.com/vertex-ai/generative-ai/pricing
[^7]: Google DeepMind. "Watermarking AI-generated text and video with SynthID." May 2024. https://deepmind.google/blog/watermarking-ai-generated-text-and-video-with-synthid/
[^8]: Koray Kavukcuoglu. "Introducing Gemini Omni." Google Blog, May 19, 2026. https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni/
[^9]: Google AI for Developers. "Gemini deprecations" (accessed October 30, 2026). https://ai.google.dev/gemini-api/docs/deprecations
[^10]: Google. "Google I/O 2024: Introducing Veo and Imagen 3 generative AI tools." Google Blog, May 14, 2024. https://blog.google/technology/ai/google-generative-ai-veo-imagen-3/
[^11]: Google Research. "Imagen Video: High-definition video generation with diffusion models." October 2022. https://imagen.research.google/video/
[^12]: Google Research. "Lumiere: A Space-Time Diffusion Model for Video Generation." January 2024. https://lumiere-video.github.io/
[^13]: Fortune. "Google DeepMind's new Veo 2 AI video generator trounces OpenAI's Sora with 4K resolution." December 16, 2024. https://fortune.com/2024/12/16/google-deepmind-veo-2-ai-video-generator-4k-openai-sora/
[^14]: The Decoder. "Google's Veo 2 outperforms OpenAI's Sora Turbo in head-to-head AI video generation tests." December 2024. https://the-decoder.com/googles-veo-2-outperforms-openais-sora-turbo-in-head-to-head-ai-video-generation-tests/
[^15]: 9to5Google. "Google announces Veo 2 video generation model, expanding VideoFX access." December 16, 2024. https://9to5google.com/2024/12/16/google-veo-2/
[^16]: Business Standard. "Google I/O 2025: All AI products announced last night, from Beam to Veo 3." May 21, 2025. https://www.business-standard.com/technology/tech-news/google-io-2025-ai-launches-beam-veo3-imagen4-gemini-updates-125052100158_1.html
[^17]: Google Developers Blog. "Build with Veo 3, now available in the Gemini API." July 17, 2025. https://developers.googleblog.com/en/veo-3-now-available-gemini-api/
[^18]: James Pero. "Google's Veo 3 Is Already Deepfaking All of YouTube's Most Smooth-Brained Content." Gizmodo, May 22, 2025. https://gizmodo.com/googles-veo-3-is-already-deepfaking-all-of-youtubes-most-smooth-brained-content-2000605140
[^19]: BizzBuzz. "Google Launches Veo 3 AI Model, Sparking Viral Videos and Misinformation Fears." May 2025. https://www.bizzbuzz.news/technology/google-launches-veo-3-ai-model-sparking-viral-videos-and-misinformation-fears-1362803
[^20]: Andrew R. Chow and Billy Perrigo. "Google's New AI Tool Generates Convincing Deepfakes of Riots, Conflict, and Election Fraud." TIME, June 3, 2025. https://time.com/7290050/veo-3-google-misinformation-deepfake/
[^21]: Google AI for Developers. "Gemini Developer API pricing" (accessed October 30, 2026). https://ai.google.dev/gemini-api/docs/pricing ; archived snapshots: November 3, 2025 https://web.archive.org/web/20251103074340/https://ai.google.dev/gemini-api/docs/pricing ; February 1, 2026 https://web.archive.org/web/20260201102619/https://ai.google.dev/gemini-api/docs/pricing ; April 1, 2026 https://web.archive.org/web/20260401191258/https://ai.google.dev/gemini-api/docs/pricing ; May 1, 2026 https://web.archive.org/web/20260501160638/https://ai.google.dev/gemini-api/docs/pricing
[^22]: Google Blog. "Meet Flow: AI-powered filmmaking with Veo 3." May 20, 2025. https://blog.google/innovation-and-ai/products/google-flow-veo-ai-filmmaking-tool/
[^23]: Google Blog. "Introducing Veo 3.1 and advanced capabilities in Flow." October 15, 2025. https://blog.google/technology/ai/veo-updates-flow/
[^24]: Google AI for Developers. "Generate videos with Veo 3.1 in Gemini API." Archived November 20, 2025. https://web.archive.org/web/20251120/https://ai.google.dev/gemini-api/docs/video
[^25]: Google DeepMind. "Veo 3 Technical Report." 2025. https://storage.googleapis.com/deepmind-media/veo/Veo-3-Tech-Report.pdf
[^26]: Google Flow Help. "Manage your Google Flow credits" (accessed October 30, 2026). https://support.google.com/flow/answer/16526234
[^27]: Google DeepMind. "SynthID" (accessed October 30, 2026). https://deepmind.google/models/synthid/
[^28]: MIT Technology Review. "Google's generative video model Veo 3 has a subtitles problem." July 15, 2025. https://www.technologyreview.com/2025/07/15/1120156/googles-generative-video-model-veo-3-has-a-subtitles-problem/
[^29]: CNBC. "Creators say they didn't know Google uses YouTube to train AI." June 19, 2025. https://www.cnbc.com/2025/06/19/google-youtube-ai-training-veo-3.html
[^30]: AI Incident Database. "Incident 1128: Philippine Officials Reportedly Share Veo 3-Generated Video to Support Vice President Sara Duterte During Impeachment." https://incidentdatabase.ai/cite/1128/
[^31]: Ryan Whitwam. "TikTok is being flooded with racist AI videos generated by Google's Veo 3." Ars Technica, July 2, 2025. https://arstechnica.com/ai/2025/07/racist-ai-videos-created-with-google-veo-3-are-proliferating-on-tiktok/
[^32]: Kyle Wiggers. "Veo 3 can generate videos, and soundtracks to go along with them" (headline uses a dash). TechCrunch, May 20, 2025. https://techcrunch.com/2025/05/20/googles-veo-3-can-generate-videos-and-soundtracks-to-go-along-with-them/
[^33]: Google Blog. "Turn your photos into videos in Gemini." July 10, 2025. https://blog.google/products-and-platforms/products/gemini/photo-to-video/
[^34]: Google AI for Developers. "Release notes" (Gemini API changelog, accessed October 30, 2026). https://ai.google.dev/gemini-api/docs/changelog
[^35]: Google Blog. "Veo 3.1 Ingredients to Video: More consistency, creativity and control." January 13, 2026. https://blog.google/innovation-and-ai/technology/ai/veo-3-1-ingredients-to-video/
[^36]: Google Blog. "Build with Veo 3.1 Lite, our most cost-effective video generation model." March 31, 2026. https://blog.google/innovation-and-ai/technology/ai/veo-3-1-lite/
[^37]: Google. Gemini app video generation page (accessed October 30, 2026). https://gemini.google/overview/video-generation/ ; archived February 23, 2026 (Veo 3.1 version): https://web.archive.org/web/20260223103843/https://gemini.google/overview/video-generation/
[^38]: Google Cloud. "Vertex AI release notes" and "Gemini Enterprise Agent Platform release notes" (accessed October 30, 2026). https://docs.cloud.google.com/vertex-ai/generative-ai/docs/release-notes ; https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes
[^39]: Google Cloud. "Model versions and lifecycle." Gemini Enterprise Agent Platform documentation (accessed October 30, 2026). https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/model-versions
[^40]: Google Blog. "Get more out of Flow in Google AI Ultra." August 18, 2025. https://blog.google/innovation-and-ai/models-and-research/google-labs/flow-whisk-ai-credits-update/
[^41]: 9to5Google. "Gemini app Pro subscribers now get three 'Veo 3 Fast' videos per day." June 9, 2025. https://9to5google.com/2025/06/09/gemini-veo-3-fast/
[^42]: Google. "Gemini Apps limits & upgrades for Google AI subscribers." Gemini Apps Help, archived January 28, 2026. https://web.archive.org/web/20260128183520/https://support.google.com/gemini/answer/16275805
[^43]: Google. "Gemini Apps' release updates & improvements" (entries of May 20, 2025 and May 19, 2026). https://gemini.google/release-notes/
[^44]: Google Blog. "Generate videos in Gemini and Whisk with Veo 2." April 15, 2025. https://blog.google/products/gemini/video-generation/
[^45]: YouTube Official Blog. "Imagine it, create it: Veo 2 is coming to YouTube Shorts." February 13, 2025. https://blog.youtube/news-and-events/veo-2-shorts/
[^46]: YouTube Official Blog. "Unpacking the magic of our new creative tools." September 16, 2025. https://blog.youtube/news-and-events/generative-ai-creation-tools-made-on-youtube-2025/
[^47]: Google Blog. "SynthID Detector: a new portal to help identify AI-generated content" (headline uses a dash). May 20, 2025. https://blog.google/technology/ai/google-synthid-ai-content-detector/
[^48]: Google DeepMind. "Veo." Technologies page, archived June 2024. https://web.archive.org/web/20240601000000/https://deepmind.google/technologies/veo/
[^49]: Google Developers Blog. "Veo 3 and Veo 3 Fast: new pricing, new configurations and better resolution." September 8, 2025. https://developers.googleblog.com/en/veo-3-and-veo-3-fast-new-pricing-new-configurations-and-better-resolution/
[^50]: Google Developers Blog. "Veo 3 Fast and new image-to-video capabilities." July 31, 2025. https://developers.googleblog.com/en/veo-3-fast-image-to-video-capabilities-now-available-gemini-api/
[^51]: TechCrunch. "OpenAI is launching the Sora app, its own TikTok competitor, alongside the Sora 2 model." September 30, 2025. https://techcrunch.com/2025/09/30/openai-is-launching-the-sora-app-its-own-tiktok-competitor-alongside-the-sora-2-model/
[^52]: Matthias Bastian. "OpenAI sets two-stage Sora shutdown with app closing April 2026 and API following in September." The Decoder, March 28, 2026. https://the-decoder.com/openai-sets-two-stage-sora-shutdown-with-app-closing-april-2026-and-api-following-in-september/
[^53]: Kuaishou Technology. "Kling AI Launches Video 2.6 Model with 'Simultaneous Audio-Visual Generation' Capability, Redefining AI Video Creation Workflow." PR Newswire via Nasdaq, December 5, 2025. https://www.nasdaq.com/press-release/kling-ai-launches-video-26-model-simultaneous-audio-visual-generation-capability
[^54]: Ivan Mehta and Rebecca Bellan. "Runway releases its first world model, adds native audio to latest video model." TechCrunch, December 11, 2025. https://techcrunch.com/2025/12/11/runway-releases-its-first-world-model-adds-native-audio-to-latest-video-model/
[^55]: Abbie Richards. "Racist AI-generated videos are the newest slop garnering millions of views on TikTok." Media Matters for America, July 1, 2025. https://www.mediamatters.org/tiktok/racist-ai-generated-videos-are-newest-slop-garnering-millions-views-tiktok
[^56]: Ruben Villegas et al. "Phenaki: Variable Length Video Generation From Open Domain Textual Description." arXiv:2210.02399, October 2022. https://arxiv.org/abs/2210.02399
[^57]: Agrim Gupta et al. "Photorealistic Video Generation with Diffusion Models" (WALT). arXiv:2312.06662, December 2023. https://arxiv.org/abs/2312.06662
[^58]: Dan Kondratyuk et al. "VideoPoet: A Large Language Model for Zero-Shot Video Generation." arXiv:2312.14125, December 2023. https://arxiv.org/abs/2312.14125
[^59]: Google AI for Developers. "Generate videos with Veo 3.1 in Gemini API." Archived January 31, 2026. https://web.archive.org/web/20260131223404/https://ai.google.dev/gemini-api/docs/video ; October 16, 2025 snapshot: https://web.archive.org/web/20251016115333/https://ai.google.dev/gemini-api/docs/video
[^60]: Google Developers Blog. "Gemini 2.5 Flash and Pro, Live API, and Veo 2 in the Gemini API." April 9, 2025. https://developers.googleblog.com/en/gemini-2-5-flash-pro-live-api-veo-2-gemini-api/

