Veo¶

Development line: project:veo · thread veo-development
Last event: 2024-05-14 · 1 dated since 2024-05-14 · Researched: 2026-09-05 · confidence: high

What it is¶

Veo is Google DeepMind’s video-generation model family for filmmakers, creators, and API users.

• Video clips: generates text-to-video and image-to-video clips.
• Frame controls: supports reference images, first/last-frame generation, and extension.
• Audio tracks: generates synchronized dialogue, sound effects, and ambience.

Development line¶

• 2024-05-14 — Veo introduced at Google I/O 2024. On 2024-05-14, Google introduced Veo at Google I/O 2024. This marks the start of the identified Veo development line and is material to its public history.

What changed¶

• 2024-05-14 — Google introduced Veo as a 1080p video model, available only to select creators through VideoFX private preview.
• 2025-05-20 — Veo 3 added native audio. Google introduced Flow and made Veo 3 available in Gemini, Flow, and Vertex AI under announced access conditions.
• 2025-09-26 — Veo 3 was the subject of a zero-shot research report covering visual perception, manipulation, physical modeling, and simple visual reasoning. This was research evidence, not a new product endpoint.
• 2025-10-15 — Veo 3.1 added audio to Ingredients to Video, Frames to Video, and Extend in Flow, with stronger image-to-video prompt adherence.
• 2026-01-13 — Veo 3.1 Ingredients to Video added native 9:16 output and 1080p/4K options in supported professional workflows.

How to use this¶

From 2024-05-14, practitioners should treat Veo as a distinct video-generation project and track its subsequent releases, access changes, and model capabilities separately from general Google I/O announcements.

1. Create a Gemini API client, submit a generation request with a Veo model name, poll the long-running operation, then download the generated video. — https://ai.google.dev/gemini-api/docs/veo?hl=en
2. Write the prompt as subject, action, style, camera, composition, focus, and ambience. Include quoted dialogue or explicit sound cues when audio matters. — https://ai.google.dev/gemini-api/docs/veo?hl=en
3. Supply an initial image, up to three reference images, or first and last frames for controlled image-to-video work. Use extension only for a Veo-generated clip. — https://ai.google.dev/gemini-api/docs/veo?hl=en
4. Download the output promptly: Google documents a two-day server retention period. — https://ai.google.dev/gemini-api/docs/veo?hl=en

Best practices¶

• Specify camera position or movement, framing, lens/focus, and ambience instead of relying on style labels alone. — https://ai.google.dev/gemini-api/docs/veo?hl=en
• Use reference images for character, product, or scene consistency; the documented limit is three assets. — https://ai.google.dev/gemini-api/docs/veo?hl=en
• Plan high-resolution, reference-image, and extension shots as eight-second units; extension is limited to 720p. — https://ai.google.dev/gemini-api/docs/veo?hl=en
• Keep speech and sound requirements explicit, then review outputs. Google documents remaining limitations with natural, consistent spoken audio. — https://deepmind.google/models/veo/

Superseded by this¶

• 2024-05-14 — Veo is no longer accurately described as VideoFX private-preview-only; current documentation exposes it through the Gemini API and multiple Google products.
• 2024-05-14 — The original “beyond a minute” positioning does not describe the current documented API unit, which is a 4-, 6-, or 8-second generated clip.

Still unknown¶

• The 2024-05-14 item carried no URL, so its association is based on Google’s same-date Veo announcement rather than the original linked material.
• The 2025-09-26 item links to a Veo 3 research project, not a product-release page. Its official publication page is dated 2025-09-24; the exact relationship between the September 26 item and a particular paper revision is not stated.
• The zero-shot research result should not be treated as a supported production workflow or as evidence that Veo exposes segmentation, editing, or reasoning APIs.

Sources¶

Agent brief¶

• Subject: project:veo, thread veo-development, 1 dated events 2024-05-14 → 2024-05-14.
• Practical note: From 2024-05-14, practitioners should treat Veo as a distinct video-generation project and track its subsequent releases, access changes, and model capabilities separately from general Google I/O announcements.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.