Veo 3.1 video length limit. Clip and extension planner.
In the Gemini API and in Vertex AI, one Veo 3.1 clip is 4, 6 or 8 seconds long. In the Gemini API, you can extend a 720p video from Veo 3.1 or Veo 3.1 Fast by 7 seconds, up to 20 times. The maximum is then 148 seconds. Vertex AI states different extension limits. Each number on this page names its surface and its source.
- 4, 6 or 8 sOne clip. Gemini API and Vertex AI.
- 7 sOne extension. Gemini API and Vertex AI.
- 148 sMaximum with extension. Gemini API only.
- 37 sMaximum with extension. Vertex AI only.
Plan a target length. Clips or extensions.
Type a target length in whole seconds. The planner shows two paths. Path one uses separate 8-second clips. Path two uses one 8-second clip and 7-second extensions, with the limits of the Gemini API.
Separate 8-second clips
- Clips
- Generated seconds
- Requests
- Frames at 24 fps
Each clip is a separate file. The clip length is for the Gemini API and Vertex AI. The request count uses the Gemini API rule of 1 video for each request.
Extension in the Gemini API
- Extensions of 7 seconds
- Final length in seconds
- Requests
- Frames at 24 fps
One 8-second clip first. 720p only. Veo 3.1 and Veo 3.1 Fast only. The maximum of 148 seconds is the Gemini API limit.
Separate clips. The clip count is the target divided by 8, rounded up. The planner counts one request for each clip. This agrees with the Gemini API, which gives 1 video for each request. Vertex AI permits up to 4 output videos for one prompt, so the request count can be lower there when clips use the same prompt. The official pages give no method to join separate clips.
Extension. A target of 8 seconds or less needs no extension. For a longer target, the extension count is the target minus 8, divided by 7, rounded up. Each extension is one more request, and the result is one video. The Gemini API permits 20 extensions, so the longest result is 8 + 7 × 20 = 148 seconds.
A request is one generation call. The planner does not count status polling calls. A frame count is our arithmetic at 24 frames per second, not a number from Google. For example, 8 seconds is 192 frames and 148 seconds is 3,552 frames.
The planner always uses 8-second clips. A 720p clip can also be 4 or 6 seconds, so a mix can give less surplus. For example, 8 + 8 + 8 + 6 = 30 seconds.
| Target | Separate 8-second clips | Extension in the Gemini API |
|---|---|---|
| 30 seconds | 4 clips, 32 seconds, 4 requests | 4 extensions, 36 seconds, 5 requests |
| 60 seconds | 8 clips, 64 seconds, 8 requests | 8 extensions, 64 seconds, 9 requests |
| 148 seconds | 19 clips, 152 seconds, 19 requests | 20 extensions, 148 seconds, 21 requests |
| 149 seconds | 19 clips, 152 seconds, 19 requests | Not possible. The maximum is 148 seconds. |
Primary source for the extension limits: Generate videos with Veo 3.1 in Gemini API, ai.google.dev, last updated 8 Oct 2026, fetched 11 Oct 2026. The input range of 1 to 3600 seconds and the method with 8-second clips are choices of this planner. They are not Google limits.
Gemini API limits. Veo 3.1, Fast and Lite.
These limits are from the Gemini API page for Veo 3.1. That page gives one column for Veo 3.1 and Veo 3.1 Fast and one column for Veo 3.1 Lite.
| Limit in the Gemini API | Veo 3.1 and Veo 3.1 Fast | Veo 3.1 Lite |
|---|---|---|
| Clip length for one request | 4, 6 or 8 seconds | 4, 6 or 8 seconds |
| Settings that need 8 seconds | 1080p, 4k, reference images, extension | 1080p, reference images |
| Resolution | 720p (default), 1080p, 4k | 720p (default), 1080p |
| Aspect ratio | 16:9 (default), 9:16 | 16:9 (default), 9:16 |
| Frame rate | 24 fps | 24 fps |
| Videos for each request | 1 | 1 |
| Extension | 7 seconds each, up to 20 times, 720p only | Not available |
| Longest input for an extension | 141 seconds | Not available |
| Longest video with extension | 148 seconds | Not available |
| Storage on the server | 2 days | 2 days |
| Status | Preview | Preview |
An extension input must be a video that Veo generated. It must be 720p, in 9:16 or 16:9, and 141 seconds or less. With an 8-second first clip, the 20th extension starts from 141 seconds and ends at 148 seconds. The output is one video that combines the input video and the new part.
For Veo 3.1 Lite, the duration cell of the parameter table names reference images. The same table gives "n/a" for the referenceImages parameter of Veo 3.1 Lite.
The Gemini API stores a generated video for 2 days. A reference for extension resets the 2-day timer. You can extend only a video that was generated or referenced in the last two days, so download each video that you want to keep.
Primary source: Generate videos with Veo 3.1 in Gemini API, ai.google.dev, last updated 8 Oct 2026, fetched 11 Oct 2026.
Vertex AI limits. Not the same numbers.
Google's documents now use the name Gemini Enterprise Agent Platform for Vertex AI. The clip lengths agree with the Gemini API. The extension limits and the videos for each prompt do not.
| Limit in Vertex AI | Veo 3.1 | Veo 3.1 Fast and Veo 3.1 Lite |
|---|---|---|
| Clip length | 4, 6 or 8 seconds. Reference image to video supports only 8 seconds. | 4, 6 or 8 seconds |
| Output resolution | 720p, 1080p, 4K | 720p, 1080p |
| Aspect ratio | 9:16, 16:9 | 9:16, 16:9 |
| Frame rate | 24 FPS | 24 FPS |
| Output videos for each prompt | Maximum 4 | Maximum 4 |
| Extension length | 7 seconds for each extension | |
| Input video for an extension | 1 to 30 seconds, MP4, 24 frames per second | |
| Longest video with extension | 37 seconds | |
| Extension resolution | 720p, 1080p or 4k in the section "Veo limitations". The REST steps on the same page give 4k for "Veo 3.1 Preview models only". | |
| Storage period | Not stated on the Vertex AI model page | |
The Vertex AI pages state no maximum number of extensions. The limit of 20 extensions and the maximum of 148 seconds are not on the Vertex AI pages that we read. The Vertex AI page for extension lists all three models, with Veo 3.1 Lite as a preview. The Gemini API page says that extension is not available for Veo 3.1 Lite. The two Vertex AI pages do not agree in full on 4k. The model page lists only 720p and 1080p as output resolutions for Veo 3.1 Fast and Veo 3.1 Lite.
Primary sources: Veo 3.1 model page and Extend videos, docs.cloud.google.com, each last updated 10 Oct 2026, fetched 11 Oct 2026.
Combinations the Gemini API rules out. Check before you send.
Each rule below is a quote from the Gemini API page. We did not send test requests, so this page does not show the error text of the API.
Rule 1
1080p or 4k with a 4-second or 6-second clip
The parameter table says: "1080p" (only supports 8s duration), "4k" (only supports 8s duration). Use 720p for a clip of 4 or 6 seconds. The Vertex AI model page does not state this rule.
Source: Gemini API Veo 3.1 page, parameter
resolution, last updated 8 Oct 2026, fetched 11 Oct 2026.Rule 2
Extension above 720p
The page says: "Video extension is also limited to 720p videos." A 1080p or 4k clip from the Gemini API cannot go into the extension path. The Vertex AI page for extension states 720p, 1080p or 4k, and its REST steps give 4k for "Veo 3.1 Preview models only".
Source: Gemini API Veo 3.1 page, section "Control the resolution", last updated 8 Oct 2026, fetched 11 Oct 2026.
Rule 3
Extension with Veo 3.1 Lite
The page says: "This feature is available for Veo 3.1 & Veo 3.1 Fast models only, not Veo 3.1 Lite." The Vertex AI page lists Veo 3.1 Lite as a preview model for extension.
Source: Gemini API Veo 3.1 page, section "Extending Veo videos", last updated 8 Oct 2026, fetched 11 Oct 2026.
Rule 4
An extension input that is too long, too old or not from Veo
The page says: "Veo-generated videos only up to 141 seconds long." It also says: "You can only extend videos that were generated or referenced in the last two days."
Source: Gemini API Veo 3.1 page, section "Extending Veo videos", last updated 8 Oct 2026, fetched 11 Oct 2026.
Rule 5
4k with Veo 3.1 Lite
The page says: "4k not available for Veo 3.1 Lite". The Vertex AI model page also lists only 720p and 1080p for Veo 3.1 Lite.
Source: Gemini API Veo 3.1 page, section "Control the resolution", last updated 8 Oct 2026, fetched 11 Oct 2026.
Model codes and dates. Different on each surface.
Check the dates before you start a project that needs these models. The limits on this page apply to the model codes below.
Gemini API
Three preview codes
The codes are
veo-3.1-generate-preview,veo-3.1-fast-generate-previewandveo-3.1-lite-generate-preview. The Gemini API page shows the status Preview for all three. The Gemini API deprecations page lists October 22, 2026 as the shutdown date of these three codes, withgemini-omni-1.1-flashas the recommended replacement. The same page says that the dates in its table are "the earliest possible dates on which a model might be retired". It says that Google will communicate the exact date with advance notice.Sources: Gemini API Veo 3.1 page, last updated 8 Oct 2026, and Gemini deprecations, last updated 9 Oct 2026. Both fetched 11 Oct 2026.
Vertex AI
Three model IDs
The model IDs are
veo-3.1-generate-001,veo-3.1-fast-generate-001andveo-3.1-lite-generate-001. The first two have the launch stage GA and the retirement date "November 17, 2026 or later". Veo 3.1 Lite has the launch stage Preview, and the page lists no retirement date for it.Source: Veo 3.1 model page, docs.cloud.google.com, last updated 10 Oct 2026, fetched 11 Oct 2026.
Four open points.
1. A maximum number of extensions in Vertex AI. The pages give only the input limit of 30 seconds and the maximum of 37 seconds.
2. Whether a clip of 4 or 6 seconds can be the input of an extension in the Gemini API. The parameter table says that the duration must be 8 when you use extension.
3. A method to join separate clips, or a limit for the number of separate clips.
4. A storage period for Veo videos on the Vertex AI model page.
Google can change these pages. Open the source links and compare the "last updated" dates with the dates on this page before you use a number.