How to Use Google Veo 3.1 for Free (2026): Gemini, Flow, AI Studio, Vids

Rate limits on Google's video model are a routing problem, not a hard wall. Google puts the same model behind several surfaces, and each one runs its own quota. When one cuts you off, you switch to the next. That was the whole trick with Veo 2 in the video, and it's still the trick with Veo 3.1 today; only the surfaces and the numbers have changed.

What changed since I recorded this#

Checked against Google DeepMind's Veo model page and the Flow plans page on September 5, 2026:

• Veo 2 is no longer what Google serves. The current model is Veo 3.1: 8-second clips, native audio (dialogue and sound generated with the video), 1080p and 4K output, image-to-video with character and style consistency, and scene extension for longer sequences.
• Where it lives: the Gemini app, Google Flow, Google Vids, Google AI Studio, and the Gemini API. VideoFX, the Labs experiment with the waitlist I walked through in a second video, became Flow, so the approval process is gone; sign in and generate.
• Flow is the clearest free quota: 50 daily credits on the free plan, with access to Veo 3.1, Gemini Omni Flash, and Nano Banana Pro. Google AI Plus ($4.99 a month) adds 200 monthly credits on top, AI Pro ($19.99) adds 1,000, AI Ultra adds 10,000 at $99.99 or 25,000 at $199.99. Google says per-clip costs vary by model and resolution and does not publish a fixed figure on that page.
• The Gemini app offers Veo 3.1 too, with daily limits Google does not publish for the free tier. Expect a handful of clips a day and more on the paid plans.
• AI Studio exposes Veo through the Gemini API, which is the developer route; check the API pricing before you rely on it as a free surface.

Everything below the fold is my original testing: what the model produced, where it hallucinated, and the prompt patterns that got usable clips. The prompt advice holds. The Veo 2 results are a floor; 3.1 is better on every axis I tested, and it has sound.

The 2026 Rotation#

1. Flow first. 50 credits a day, no waitlist, and the newest model. Generate in Flow, download, move on.
2. Gemini app second. When Flow's daily credits run out, describe the clip in a Gemini chat. Separate quota.
3. Google Vids third if you have a Google account with Vids access: it is aimed at work presentations, but it renders Veo clips into a deck without leaving the doc.
4. AI Studio and the API for anything programmatic, or when you want the negative-prompt and reference-image controls the consumer apps hide.

When all four are exhausted and you still have a batch to run, I use Artlist's unlimited AI video generator, which resells Veo 3.1 (and Sora 2, Kling, and Seedance) without a per-clip credit meter. It is a subscription, not free, and I tested whether its output matches Google's directly in Sora 2 vs Veo 3.1 by use case.

Start With Google AI Studio (the original walkthrough)#

Go to Google AI Studio, sign in with your Google account, and select Video Generation. When I recorded this, Veo 2 was the active model and generation was free with no waitlist; today the panel shows the current Veo model and routes through the Gemini API.

Before you run anything, adjust three settings:

• Number of results: Set this to 2. You get two videos per generation, which doubles your output for the same quota spend. If you hit a quota wall, drop it to 1; that's sometimes enough to squeeze out one more generation.
• Aspect ratio: 16:9 for screens, 9:16 if your content is primarily for mobile.
• Duration: shorter clips generate faster. Start there.

You'll also need to connect Google Drive when prompted. One click, confirm, done. Your generations get saved there automatically; nothing stays in AI Studio's interface once you close the tab.

One workflow trick: open a second AI Studio tab and start a new generation while the first one is running. They process in parallel, so you're not sitting idle waiting for one clip to finish before starting the next.

What Veo Produced in My Tests#

I ran four test generations on Veo 2 to see where the model holds up and where it breaks. I've kept them because the failure modes, not the successes, are what you need to prompt around.

Dog running on the beach at sunset. Both outputs were cinematic. The shadow and reflection looked accurate, and the dog's back paws kicking up sand were rendered with real physical weight. The only tell: the front paws weren't sharp. If no one told me this wasn't real footage, I wouldn't have known.

iPhone product showcase. I prompted a close-up, cinematic video of the front of the latest iPhone highlighting the camera. The result looked like something out of an Apple ad, except the phone showed two camera lenses instead of three. That's the model hallucinating hardware details. The framing, lighting, and motion were all convincing. The specific product spec was wrong. Keep that in mind for any product work where accuracy matters; Veo 3.1's reference-image consistency helps, but check every frame.

San Francisco skyline hyperlapse with a reference image. You can upload a reference image and prompt against it. I uploaded a San Francisco skyline and asked for a Ghibli-style hyperlapse at sunset. The clouds moved. The ocean didn't; it stayed static. A more specific prompt ("animate the ocean waves") would have fixed that. The reference image was matched accurately; the motion was just incomplete.

Clown riding a unicycle on the moon. Completely absurd, rendered perfectly. Cinematic lens flare, dust kicked up behind the unicycle, the whole thing. The model handles surreal prompts just as well as realistic ones.

Writing Prompts That Get Usable Results#

The iPhone video worked because the prompt was specific: subject, shot type, purpose, and style all in one sentence. "Create a video of the front of the latest iPhone to showcase the camera. Close-up, cinematic." That's the pattern.

For reference image videos, you need to tell the model what to do with the image, not just describe it. "Create a hyperlapse of [subject] during sunset in Ghibli style" is a direction. Uploading the image without a motion instruction gets you something that barely moves.

Since 3.1 generates audio, say what you want to hear. "Waves and gulls, no music" or "a single line of dialogue: 'we made it'" steers the soundtrack the same way the shot description steers the picture.

Negative prompts are available in AI Studio and worth using. If you're generating product footage and want to avoid lens distortion, blurry backgrounds, or text overlays, put those in the negative prompt field before you run.

If you want a structured starting point for prompts across product showcases, ads, and branded content, I put together a free Marketing Prompt Collection that covers these workflows.

The Old VideoFX Waitlist, For the Record#

My second Veo video walked through getting approved for VideoFX: profession "YouTuber," purpose "educational and research," opting into research invitations, and waiting anywhere from minutes to weeks for the email. None of that is needed now. Flow is the successor and it opens on sign-in. I'm keeping the note here because the old URL still gets searched and I'd rather you land on the answer than on a dead form.

Flow is the cleanest entry point in 2026: no approval, 50 credits a day, the newest model. Start there, learn what your prompts actually produce, and use the Gemini app, Vids, and AI Studio as overflow when the credits run out. The model is good enough that the limiting factor is prompt quality, not the tool. For how Veo 3.1 stacks up against the competition on real briefs, Kling 3.0 vs Sora 2 vs Veo 3.1 is the head-to-head.

Watch the full video on YouTube: https://youtu.be/D7KcRkURhLg

This post contains affiliate links. I only recommend tools I actually use.