New Heights in AI Video Technology: A Preview of Veo 3.1

Technical Background: Developments in Sora 2 and Veo 3

Generative AI technology is entering the professional video creation field at an unprecedented pace. In September 2025, OpenAI released the groundbreaking Sora 2 model, marking a significant leap in text-to-video technology. OpenAI's research team stated that Sora 2 represents a qualitative improvement over its predecessor in terms of physical simulation accuracy and visual realism, capable of providing perfectly synchronized dialogue and sound effects while maintaining scene consistency when handling complex multi-shot instructions. This technical upgrade positions Sora 2 as a new benchmark in the AI video generation domain.

As a major competitor in this field, Google DeepMind's Veo 3 has also demonstrated excellent performance. According to official DeepMind information, Veo 3 comes with advanced audio generation capabilities, capable of creating high-quality sound effects, ambient sounds, and dialogue content, while achieving industry-leading levels in video quality, visual realism, and prompt execution precision. In September 2025, Google made important updates to both the standard Veo 3 and the economical Veo 3 Fast, adding 1080p video and 9:16 vertical format support, along with significant price adjustments—reducing costs to $0.40 per second (original $0.75) and $0.15 per second (original $0.40) respectively. These rapid iterations have sparked significant industry interest and speculation about the incremental Veo 3.1 version.

Veo 3.1 Technical Preview

While Google has not yet officially announced Veo 3.1, multiple sources indicate that a "Veo 3.1 Early Access Program" has already begun. Analysis suggests that Veo 3.1 will serve as a feature-enhanced version of the Veo 3/Fast series, primarily focusing on extended video duration and improved creative control precision. Leaked interface screenshots reveal that the new version is expected to offer multiple output resolution options including 480p, 720p, and 1080p, with video length potentially breaking through to 30 seconds. It may also introduce advanced features like natural product placement and precise object positioning (such as the industry-followed "banana placement" precision test).

Google Veo Model Technical Evolution

The Google Veo series showcases the rapid iterative development of video generation AI models:

• Veo 1 (early 2024) – the initial system produced silent clips with moderate resolution. It established the basic architecture for text‑to‑video generation.
• Veo 2 (mid‑2024) – introduced creative controls. DeepMind’s model page notes that Veo 2 gained “reference powered video” and “match your style” features, allowing users to provide reference images to guide video generation and style matching. These controls improved consistency and creative intent.
• Veo 3 (May 2025) – re‑designed for greater realism. The DeepMind page describes Veo 3 as adding native audio generation (sound effects, ambient noise and dialogue) and improving prompt adherence. The update also claims 4K output and better physics simulation, though the Gemini API initially restricted outputs to 720 p.
• Veo 3 Fast (July 2025) – an optimized version offering lower cost and slightly reduced quality. Both Veo 3 and Veo 3 Fast were updated in September 2025 to support 1080 p video and vertical 9:16 aspect ratio.
• Veo 3.1 (Unknown) is expected to build on these improvements by providing longer video durations, higher resolution in API, and finer control over scenes and objects.

Veo 3.1's Core Upgrades and Technical Advantages

Based on leaked information analysis and industry expert predictions, Veo 3.1 may bring the following breakthroughs:

Expected advance	Evidence/rumor	Why it matters
Longer clips (up to 30 s)	Rumored waitlist screens show 30 s duration settings, extending the current 8–10 s cap in Veo 3.	Longer duration allows narrative sequences and multi‑shot stories without stitching clips.
Higher resolution & improved aspect ratios	Leaked settings mention 480 p, 720 p and 1080 p outputs by default; this would align with the September 2025 update that added 1080 p to Veo 3/Fast.	Native 1080 p would make Veo competitive with Sora 2, which already produces high‑fidelity videos.
Enhanced physics & object interactions	Veo 3 already excels at physics and prompt adherence; leaks suggest 3.1 will further refine motion realism, enabling subtle interactions like product placement or balancing objects (the “banana placement” challenge).	Stronger physics simulation would close the gap with Sora 2’s accurate modeling of buoyancy and inertia.
Product placement & character consistency tools	Higgsfield demos reportedly show UI options for inserting branded objects and maintaining consistent characters across shots.	Allows marketers and storytellers to embed products or maintain consistent actors throughout a video.
Better multi‑prompt and editing controls	Speculation includes multi‑prompt support and editing tools akin to Google’s AI Studio MediaSim app (demoed for Veo 3).	Improves creative control, enabling multi‑scene composition, transitions and style changes within one generation.

How does Veo 3.1 perform? Potential improvements in tests

• Realism and audio – Veo 3 already generates audio and realistically simulates physics. If 3.1 adds longer clips, we can expect greater temporal coherence and more varied soundscapes.
• Prompt adherence – Google claims Veo 3 follows prompts “like never before”; 3.1 could improve multi‑shot prompt following, aligning with Sora 2’s ability to persist world state across complex instructions.
• Resolution – September 2025 updates allow 1080 p output via the API, so 3.1 may enable this resolution for all aspect ratios and possibly 4K for premium tiers. The Mindstream newsletter noted that 1080 p is currently only available for 16:9 clips, while vertical videos remain lower resolution; a 3.1 update could unify these.
• Durations – If 3.1 supports 30‑s videos, we should see improved motion continuity and narrative flow. However, longer sequences could also amplify artifacts if the model fails to maintain consistency.

When will Veo 3.1 and the Veo 3.1 API be online?

Google has not released an official timeline for Veo 3.1. Observers note that the Gemini 3 and Veo 3.1 rumors coincide with Google’s October 9 2025 event, suggesting a potential announcement. In the absence of official confirmation, it is safest to assume that Veo 3.1 will roll out initially through Google’s preview channels — Gemini API, AI Studio and Flow — as Veo 3 did, followed later by general availability.

Where can we try Veo 3 (and likely Veo 3.1) first?

The DeepMind site lists several portals for trying Veo. Users can access Veo via Flow, the Gemini app, Google AI Studio, the Gemini API, Google Vids and DeeVid AI. These channels currently offer Veo 3 and Veo 3 Fast; a future Veo 3.1 preview would likely appear in the same interfaces. Early testers should watch for announcements on the Google AI Developers X account.

FAQ

Q: Is Veo 3.1 officially released?

A: As of October 8 2025 there is no official release. It is highly possible that Google will release it on Google’s October 9 2025 event.

Q: Will Veo 3.1 support longer videos?

A: Multiple leaks indicate a 30‑second option, but this has not been confirmed. Current Veo 3 clips are limited to around 10 seconds via the API.

Q: Does Veo 3.1 generate audio?

A: Veo 3 already generates sound effects, ambient noise and dialogue. Rumored updates will likely retain or improve this capability.

Q: How does Veo compare with Sora 2 and Seedance?

A: Veo focuses on realistic physics and prompt adherence with integrated audio; Sora 2 emphasizes high‑fidelity physical simulation and controllability; Seedance excels at smooth motion and multi‑shot storytelling but lacks audio.

Q: Where can I follow news on Veo 3.1?

A: Monitor the Google AI Developers blog and social media channels, the DeepMind Veo product page, and creative platforms like DeeVid AI. These sources often post updates and will host any official preview of Veo 3.1.