Model · Last updated April 10, 2026

Veo 3.1 for AI agents

Veo 3.1 is a premium video generation model exposed through AnyCap. It handles both text-to-video and image-to-video workflows: agents can generate a cinematic clip from a text brief, or animate an existing image into motion without leaving the same CLI. The result stays inside one capability runtime alongside image generation, video analysis, and other multimodal steps.

Generated example

Illustrative keyframe for a premium text-to-video brief

Video output is time-based, so this page uses a companion still to anchor the brief visually. The image reflects the kind of cinematic scene planning teams often do before sending a premium text-to-video request.

Cinematic aerial still of a futuristic city at dawn with a drone moving between tall towers in warm sunrise light. — Companion keyframe

This still was generated through AnyCap as a visual proxy for the kind of premium scene brief that pairs well with Veo 3.1.

Illustrative still prompt

cinematic aerial keyframe of a futuristic city at dawn, a drone gliding between towers, soft haze, warm sunrise rim light, premium sci-fi film still, no text, no watermark

When agents should use Veo 3.1

Generate short product demos from a written concept (text-to-video)
Animate a product screenshot, design frame, or reference photo into a cinematic clip (image-to-video)
Create motion prototypes during agent-led content workflows
Turn a text brief into an explainer or teaser draft
Keep video generation inside the same agent runtime used for image and analysis tasks

Call Veo 3.1 through AnyCap

Text-to-video

anycap video generate --model veo-3.1 --prompt "a cinematic flyover of a futuristic city at dawn" -o city.mp4

Image-to-video

anycap video generate --model veo-3.1 --mode image-to-video --prompt "slow push-in with soft parallax and ambient light shifts" --param images='["./keyframe.jpg"]' -o animated.mp4

List available video models

anycap video models

Veo 3.1 specs at a glance

Output: Video clips, up to 8 seconds, up to 1080p
Modes: Text-to-video, image-to-video
Audio: Native synced ambient sound and speech
Strengths: Premium cinematic quality, strong prompt adherence
Reference image: Strong character and composition preservation
Provider: Google DeepMind
AnyCap CLI: anycap video generate --model veo-3.1

Workflow placement

In an agent workflow, Veo 3.1 is usually the generation step that follows planning and precedes review. A coding or automation agent may draft the concept, call Veo 3.1 for the video output, then route the result into review, asset packaging, or documentation.

Upstream
Context engineering, prompt preparation, story framing, and asset selection.
Downstream
Review, editing notes, video analysis, and distribution inside the rest of the agent stack.

Veo 3.1 vs Kling 3.0 vs Seedance 2.5

All three are available through AnyCap. Compare the verified input modes below, then inspect the selected model's live schema before generating.

Dimension	Veo 3.1	Kling 3.0	Seedance 2.5
Best fit	Premium cinematic first pass, prompt fidelity	Realistic motion, image-to-video continuity	Text, image, first/last-frame, and multimodal-reference inputs
Max duration	Up to 8 seconds	Up to 15 seconds	Inspect the live mode schema
Native audio	Yes — synced ambient + speech	Yes — dialogue, ambient, SFX	Audio can be used as a multimodal reference input
Image-to-video	Strong, preserves character + composition	Strong, preserves source frame style	Supported
Provider	Google DeepMind	Kuaishou	ByteDance
AnyCap CLI	--model veo-3.1	--model kling-3.0	--model seedance-2.5

Veo 3.1 vs nearby choices

Dimension	Veo 3.1	Alternative
Best fit	Premium cinematic output from a text brief or a reference image	Choose Kling 3.0 when first/last-frame transitions or multi-shot control matter, or Seedance 2.5 when multimodal references are required
Text-to-video	Strong first-pass quality when the clip needs to land close to final from a prompt alone	Use Kling 3.0 for first/last-frame or multi-shot workflows, or Seedance 2.5 for text, image, first/last-frame, or multimodal-reference input modes
Image-to-video	Animate a reference frame into premium cinematic motion while preserving the source composition	Choose Kling 3.0 when multi-shot control matters, or Seedance 2.5 when the workflow also needs first/last-frame or multimodal references
Typical agent task	Turn a written concept or product screenshot into a polished teaser, demo, or concept clip	Route the output into review, packaging, or follow-up analysis after the initial generation step

FAQ

What is Veo 3.1 best for?

Veo 3.1 is best for premium video generation — both text-to-video and image-to-video — when an agent needs a stronger cinematic first pass from a written brief or a reference image.

How do agents use Veo 3.1 for image-to-video?

Agents can animate a reference image by running anycap video generate --model veo-3.1 --mode image-to-video with the source image passed via --param images. The CLI handles the upload and returns the video output.

How do agents call Veo 3.1 through AnyCap?

Agents call it with the AnyCap CLI using anycap video generate --model veo-3.1 and a prompt for text-to-video, or add --mode image-to-video with a reference image. Same auth as every other capability.

Should I use Veo 3.1 or Kling 3.0?

Use Veo 3.1 when the first-pass result needs the most premium look from a text brief or a reference image. Use Kling 3.0 when the workflow needs longer clips (up to 15s) or more flexible image-to-video iteration.

How long can a Veo 3.1 clip be?

Veo 3.1 generates clips up to 8 seconds at 1080p with native synced ambient audio and speech in a single pass.

Veo 3.1 vs Sora — which should an agent use?

Veo 3.1 is the production-ready API option with native audio sync and strong image-to-video. Sora is broadly available but currently lighter on multimodal API integration. For agent workflows that need a stable API surface, Veo 3.1 through AnyCap is the more dependable choice today.

Video Generation Kling 3.0 Seedance 2.5

CLI version and updating · 0.6.5

Current release checked October 21, 2026: 0.6.5. Use the update command for your installation method, then verify the version and login in the same environment that runs your agent.

Direct binary installation

anycap update

npm installation

npm install -g @anycap/cli@latest

anycap --version
anycap status

Version 0.6.5 retries transient downloads, recovers an expired video link by querying the same task, and saves multiple image outputs to numbered files. After a video download failure, resume the existing task instead of generating again.

anycap video tasks wait <task-id> -o result.mp4

Dated examples retain the CLI version, output and charge from the original run. Updating the CLI does not revalidate an old result. Check the live model schema and current pricing before a new billed task.