Google Unveils Veo 3.1 to Rival OpenAI’s Sora 2—But Does It Deliver?

Google launched Veo 3.1 today, an upgraded variation of its AI video generator that includes audio throughout all functions and presents brand-new modifying abilities developed to offer developers more control over their clips.

The statement comes as OpenAI’s completing Sora 2 app climbs up app shop charts and triggers disputes about AI-generated material flooding social networks.

The timing recommends Google wishes to place Veo 3.1 as the expert option to Sora 2’s viral social feed technique. OpenAI released Sora 2 on September 30 with a TikTok-style user interface that focuses on sharing and remixing.

The app struck 1 million downloads within 5 days and arrived area in Apple’s App Shop. Meta took a comparable technique, with its own sort of virtual social networks powered by AI videos.

Users can now produce videos with integrated ambient sound, discussion, and Foley results utilizing “Components to Video,” a tool that integrates numerous recommendation images into a single scene.

The “Frames to Video” function produces shifts in between a beginning and ending image, while “Extend” develops clips lasting as much as a minute by continuing the movement from the last second of an existing video.

New modifying tools let users include or eliminate aspects from produced scenes with automated shadow and lighting modifications. The design produces videos in 1080p resolution at horizontal or vertical element ratios.

The design is offered through Circulation for customer usage, the Gemini API for designers, and Vertex AI for business clients. Videos lasting as much as a minute can be developed utilizing the “Extend” function, which continues movement from the last second of an existing clip.

The AI video generation market has actually ended up being crowded in 2025, with Runway’s Gen-4 design targeting filmmakers, Luma Labs using quick generation for social networks, Adobe incorporating Firefly Video into Creative Cloud, and updates from xAI, Kling, Meta, and Google targeting realism, noise generation, and timely adherence.

However how great is it? We checked the design, and these are our impressions.

Checking the design

If you wish to attempt it, you ‘d much better have some deep pockets. Veo 3.1 is presently the most pricey video generation design, on par with Sora 2 and just behind Sora 2 Pro, which costs more than two times as much per generation.

Free users get 100 regular monthly credits to check the system, which suffices to create around 5 videos monthly. Through the Gemini API, Veo 3.1 expenses around $ 0.40 per 2nd of produced video with audio, while a much faster alternative called Veo 3.1 Quick expenses $0.15 per second.

For those ready to utilize it at that rate, here are its strengths and weak points.

Text to Video

Veo 3.1 is a certain enhancement over its predecessor. The design manages coherence well and shows a much better understanding of contextual environments.

It works throughout various designs, from photorealism to elegant material.

We asked the design to mix a scene that began as an illustration and transitioned into live-action video footage. It managed the job much better than any other design we checked.

With no recommendation frame, Veo 3.1 produced much better lead to text-to-video mode than it did utilizing the very same timely with a preliminary image, which was unexpected.

The tradeoff is motion speed. Veo 3.1 focuses on coherence over fluidity, making it challenging to create hectic action.

Components move more gradually however preserve consistency throughout the clip. Kling still leads in fast motion, although it needs more efforts to accomplish functional outcomes.

Image to Video

Veo developed its track record on image-to-video generation, and the outcomes still provide– with cautions. This seems a weaker location in the upgrade. When utilizing various element ratios as beginning frames, the design had a hard time to preserve the coherence levels it as soon as had.

If the timely strays too far from what would rationally follow the input image, Veo 3.1 discovers a method to cheat. It produces incoherent scenes or clips that dive in between areas, setups, or totally various aspects.

This lose time and credits, given that these clips can’t be modified into longer series since they do not match the format.

When it works, the outcomes look wonderful. Arriving is part ability, part luck– primarily luck.

Components to Video

This function works like inpainting for video, letting users place or erase aspects from a scene. Do not anticipate it to preserve best coherence or utilize your specific recommendation images, however.

For instance, the video listed below was produced utilizing these 3 recommendations and the timely: a guy and a female come across each other while running in a futuristic city, where a Bitcoin indication hologram is turning. The male informs the lady, “QUICK, BITCOIN CRASHED! WE MUST BUY MORE!!

As you can see, neither the city nor the characters are really there. Nevertheless, characters are using the clothing of recommendation, the city looks like the one in the in the image, and things represent the concept of the aspects, not the aspects themselves.

Veo 3.1 deals with uploaded aspects as motivation instead of rigorous design templates. It produces scenes that follow the timely and consist of items that resemble what you supplied, however do not lose time attempting to place yourself into a motion picture– it will not work.

A workaround: utilize Nanobanana or Seedream to publish aspects and create a meaningful beginning frame initially. Then feed that image to Veo 3.1, which will produce a video where characters and items reveal very little contortion throughout the scene.

Text to Video with Discussion

This is Google’s selling point. Veo 3.1 manages lip sync much better than any other design presently offered. In text-to-video mode, it produces meaningful ambient noise that matches scene aspects.

The discussion, modulation, voices, and feelings are precise and beat completing designs.

Other generators can produce ambient sound, however just Sora, Veo, and Grok can create real words.

Of those 3, Veo 3.1 needs the least efforts to get great lead to text-to-video mode.

Image to Video with Discussion

This is where things break down. Image-to-video with discussion struggles with the very same problems as basic image-to-video generation. Veo 3.1 focuses on coherence so greatly that it overlooks timely adherence and recommendation images.

For instance, this scene was produced utilizing the recommendation displayed in the aspects to video area.

As you can see, our test produced an entirely various topic than the recommendation image. The video quality was outstanding– modulation and gestures were spot-on– however it wasn’t the individual we submitted, making the outcome worthless.

Sora’s remix function is the very best option for this usage case. The design might be censored, however its image-to-video abilities, reasonable lip sync, and concentrate on tone, accent, feeling, and realism make it the clear winner.

Grok’s video generator can be found in 2nd. It appreciated the recommendation image much better than Veo 3.1 and produced exceptional outcomes. Here is one generation utilizing the very same recommendation image and timely.

If you do not wish to handle Sora’s social app or absence access to it, Grok may be your finest alternative. It’s likewise uncensored however moderated, so if you require that specific technique, Musk has you covered.