Veo / Flow

Of four common requests, one gets a "no" - the puppy homonym - while the app, free trial, and Veo 3.1 work.

Who it is for. For people making short videos with speech and sound for ads, social media, and presentations who do not want to assemble them from three services: Flow creates the frames, voices the lines, extends the clip, and stitches the scene together itself. And for people editing with words: the agent in the window recolors sixteen videos from day to night with one phrase, while every edit is saved as a separate version.

Who else. For people who already have a Google AI subscription for Gemini: Flow credits are included at no extra charge, 1,000 per month on Pro. And for people who need a model from the top five arenas: in September 2026, Gemini Omni Flash ranks in the top 5 of all four Artificial Analysis tables.

Who it is not for. Not for anyone who needs text in the frame or a long film with one click: the model does not render text, clips last up to 10 seconds, and longer videos are assembled by extending them in parts.

Good fit: section 3 takes you from entry to your first video. Choosing between Flow and Kling? See the decision guide in section 7.

What vibecoding isStarting from scratch: how to set tasks for agents with no coding experience

Three steps: describe the scene, get a video with sound, edit it with words, and credits are charged for every step at the selected model's price.

Two models in one window. According to Help as of Jul 9, 2026, Flow has four video models: Veo 3.1 Lite, Veo 3.1 Fast, Veo 3.1 Quality, and Gemini Omni Flash 1.1. Omni is enabled by default; Google calls it "our newest video model" and compares it to Nano Banana, only for video: 4, 6, 8, and 10-second clips, 360p drafts at half price, and word-based editing of uploaded video for up to three consecutive turns. Veo 3.1 retains what Omni does not yet have: only it can extend an eight-second clip, and the vendor itself points to "Veo 3.1 lite" when you need a first and last frame.

How to order a video. Four inputs: text, first and last frame, references ("ingredients" in the vendor's terms: up to three images of a character, object, or style), and your own video for editing. Sound comes with the image: scene noise, music, and character lines written directly in the description. Finished clips go into the scene builder, where they are trimmed with handles, reordered, and downloaded as one track.

Agent and tools. Since 05.19.2026, everyone has an agent: it suggests a plot, selects a model, creates five lighting options with one request, and organizes results into collections, while its requests cost no credits. You can also build your own tools from a description: "make a filter with a gray slider," and Flow writes the code and interface itself. Everyone can use other people's tools; only subscribers can create their own.

Images and avatar. Nano Banana Pro, Nano Banana 2, and Lite create video frames; images have been free since 02.25.2026, and a lasso selects an area and changes it with words. An avatar is made from your own face and voice on a phone via QR code and invoked in a prompt with "@me"; avatars and uploaded-video editing are unavailable in the European Economic Area, Switzerland, and the UK.

What the vendor itself shows. The header video comes from Google's 05.19.2026 Omni post: the agent recolors sixteen videos from day to night with one phrase, and the line "Using the Omni Flash Model" is visible on screen. The frames above were taken from the vendor's landing page on Jul 9, 2026; their selectors show Omni Flash, while Veo 3.1 does not appear once. We did not run Flow for this task ourselves; the figures come from Help, the storefront, terms, and other people's tests.

You can start in the browser or on Android without installing anything.

Step 1. Open labs.google/fx/tools/flow or the short address flow.google, sign in with a Google account, and confirm your age: access starts at 18. Without a subscription, you get 50 credits per day; they do not accumulate, and without a subscription videos are not generated during peak hours from 2:00 to 5:00 UTC.

Step 2. Click Video, describe the scene, pick a model and a length: Omni Flash is the default, and Veo 3.1 is selected in the same panel. The price in credits is shown in the prompt field settings, and a single request can return two clips and charge double.  

Surfaces. A computer with a Chromium-based browser, as the vendor itself recommends, and the Google Flow Android app, listed in the storefront as Google Labs: the package was inherited from Whisk along with its 5M installs, updated Jan 9, 2026, beta status according to the vendor, requiring Android 12 and 2 GB of memory. There is no iPhone app as of Jul 9, 2026, and "Google Flow Music" in the App Store is a separate music product. The agent and tools work on the web only.

The same models outside. Omni Flash and Veo 3.1 also live in the Geminiapp, in Google AI Studio and in the API for dollars per second, and Omni is in YouTube Shorts as well. Veo 3.1 is also sold inside other storefronts, for example in Higgsfield from the Plus tier.

Steps checked Sep 7, 2026 against the Flow Help, the storefront config, Google Play and clips by practitioners

Google has two different positions in the arenas: the new Omni Flash is in the top five, while Veo 3.1 has dropped out of the top ten, so the word "best" means nothing here without a model name.

Arenas. In Artificial Analysis on Sep 7, 2026, Gemini Omni Flash ranks second in both "text to video" tables, fourth in "image to video with audio," and first without audio. Veo 3.1 ranks 16th, 27th, 10th, and 32nd in the same tables, below Wan 3.0, MiniMax H3, Seedance 2.0 and Kling 3.0. In Arena, formerly LMArena, Omni 1.1 Flash is first in "text to video" and second in "image to video," while Veo is absent from all three top 10 lists.

Artificial Analysis Video Arena and Arena, blind user votes, captured on Sep 7, 2026; the Veo 3.1 row was captured with all models displayed, while the default view shows 14th, 24th, 10th, and 26th places; Omni 1.1's first place in Arena is based on 1,777 votes, with rank uncertainty from 1 to 5

Vendor measurements. Google compares models using its own evaluators: Veo 3.1 charts are "updated in October 2025," Omni has no date, and the vendor gives both the conclusion of "best performance" on the MovieGenBench dataset. The scale is self-reported too: more than 275 million videos in five months (Oct 15, 2025, together with Veo 2), more than 1.5 billion images and videos since launch (Feb 25, 2026), and "millions of creators" without a number (May 19, 2026). Google does not publish Flow's user count; the Android app uses the package of the former Whisk, which has more than 5 million installs.

Demand. The place among video tools is shown under the card name, the whole tool ranking sits on the shared page.

Free access includes 50 credits a day, while a Google AI subscription from $4.99 adds monthly credits and removes caps.

Credits. One currency covers all models, and the rate is listed in Help: Omni Flash 720p costs 7, 10, 12, and 15 credits for 4, 6, 8, and 10 seconds, 360p draft is half the price, editing an uploaded video costs 40, Veo 3.1 Lite 10, Fast 20, and Quality 100 per video. Upscaling, meaning increasing a finished video's resolution, to 1080p is free for subscribers; 4K costs 50 credits and is available only on Ultra. One request can produce two videos, in which case twice as many credits are charged. Between 3 and 7 September 2026, Omni dropped from 25 to 12 credits for 8 seconds, and a 360p draft mode appeared.

Plans. In the US storefront on Sep 7, 2026: Plus is $4.99 for 200 monthly credits, Pro $19.99 for 1,000, Ultra $99.99 for 10,000, and Ultra $199.99 for 25,000, all in addition to the daily 50. There is no annual payment, the grid has not changed since Jul 28, 2026, and before May 19, 2026 Ultra cost $249.99. On Pro, monthly credits cover 10 Veo 3.1 Quality videos or 83 Omni Flash 720p videos, meaning a Quality second costs $0.25 and an Omni second $0.03. On Ultra, Veo Lite and Fast are half the price, while Quality still costs 100.

Free. The daily 50 credits do not roll over: they cover four Omni 720p videos or five Veo Lite videos, while Veo 3.1 Quality at 100 can never be accumulated. Upscaling is unavailable on the free plan, videos cannot be made without a subscription during peak hours from 2:00 to 5:00 UTC, and remaining free credits expire when a subscription is purchased.

API. A second of Veo 3.1 with audio costs $0.40 at 720p and 1080p and $0.60 at 4K; Fast costs $0.10 and Lite $0.05 only at 720p, and without audio they cost half as much. Omni Flash is priced in tokens: 5,792 tokens per second of 720p at $17.50 per million, or about $0.10 per second. A subscription is cheaper than the API almost everywhere: a Quality second on Pro costs $0.25 versus $0.40.

Three questions: who owns these videos, what Google does with your data, and what scandals have already happened.

Whose videos. Flow has no separate terms; Google's general terms apply: "your content remains yours," and Google's license is non-exclusive and needed to store and display what is made. Neither the terms nor Help has a direct statement on commercial use of generated video. The terms prohibit presenting a video as filmed by a person and training your own models on it.

Watermarks. An invisible SynthID mark is present in every Flow video and image; Omni also claims C2PA provenance metadata, through which software can recognize that a video was made by AI. The visible mark is the "Visible watermarking" toggle in the profile menu, and it is mandatory for residents of India, South Korea, and Vietnam. The vendor introduced it on May 30, 2025 for everyone except Ultra subscribers in Flow after TIME asked about deepfakes of riots and ballot damage; according to the same article, the visible mark can be removed with a regular editor.

Your data. With “Help improve Google Flow” enabled, Google uses prompts, uploads, and generated videos to improve products and train models; selected interactions are reviewed by trained reviewers and stored separately from the account for up to 18 months. The setting can be turned off in the menu, and the vendor does not state the default. Projects are deleted all at once; tool and agent data is deleted permanently without confirmation; exports use Google Takeout.

Restrictions. Videos of prominent people and editing videos featuring identifiable minors are prohibited in all regions; avatars and editing uploaded videos are unavailable in the European Economic Area, Switzerland, and the UK; access is 18+. In summer 2025, Media Matters found racist videos labeled Veo on TikTok, one reaching 14.2 million views; attribution is indirect, based on the label and clip length.

Disney. On Dec 11, 2025, the studio sent Google a claim alleging infringement “on a massive scale”; according to Engadget’s account, the letter names Veo and Nano Banana as trained on protected works. Within a week, Google removed dozens of Disney-character videos from YouTube. As of Sep 7, 2026, there is no lawsuit against Google over the video model: rightsholder cases are proceeding against Midjourney and MiniMax, while cases against Google concern books and Gemini.

The distinction is not who makes prettier videos, but speech and text-based edits versus affordable dynamics.

Kling. Chinese video model with its own site: $10 per month for 660 credits and daily free credits. Kling excels at motion and price, while Flow delivers speech in the frame, text-based edits to finished videos, and a scene builder. In the September 2026 arenas, Kling 3.0 ranks above Veo 3.1 but below Omni Flash. Choose Kling if you need affordable, energetic videos; choose Flow if you need character dialogue.

Higgsfield. A showcase of third-party models where Veo 3.1 sits next to Kling and Seedance, starting with the $59 Plus plan: a route to Veo. Omni isn't there, but Veo's sound and speech are. Veo 3.1 is also available in Luma AI, where usage depends on the model, length and resolution of the clip.

The closed Sora is listed in the catalog as an obituary, and its card itself calls Veo 3.1 the closest equivalent with sound.

In two years, Google’s lab model gained its own studio, audio, a summer deepfake scandal, a Disney claim, and a new neighbor, Gemini Omni, which became the face of Flow in May 2026.