Veo 3.1 reviews: what it actually does

Veo 3.1 reviews describe the model in nearly identical terms: it is Google's video generator that produces picture and sound in a single pass — speech, ambience and effects included — and with a detailed prompt it delivers commercial-grade framing. The same reviews repeat three complaints: clips last only seconds, on-screen text and infographics fall apart, and direct access usually means a VPN plus a foreign subscription. In IIshki studio, Veo 3.1 runs without a VPN, with durations of 4, 6 or 8 seconds, 16:9 and 9:16 framing and resolutions up to 4K — and Kling 3.0, Seedance 2.0, Hailuo H3 and Wan 3.0 sit next to it for the jobs where Veo hits its limits. Here is which parts of those reviews hold up, what to do about each, and how to build your first clip.

What Veo 3.1 reviews agree on

Collecting hands-on tests and discussions from the past year gives a consistent picture.

The takeaway from the reviews: Veo 3.1 is a model for one strong scene with sound, not a one-click minute-long video. The moment the task drifts from that, another model fits better.

What Veo 3.1 actually does

Here is what the generation form offers, without the claims from promotional round-ups:

• Video from text and from frames. First and last frame can be uploaded, or skipped entirely — then the scene is built from the prompt alone.
• Native audio. Speech, effects and ambience are generated together with the footage; there is no separate voicing step.
• Durations of 4, 6 or 8 seconds. Three fixed options, nothing in between.
• 16:9 and 9:16. Landscape for YouTube and presentations, vertical for Reels, Shorts and stories. No square, no 4:3.
• 720p, 1080p and 4K.
• Fast and Standard modes. Fast is cheaper and quicker; Standard is for the final take.

What the model does not do, and what reviews sometimes skip: you cannot feed it a set of reference photos of a character or product the way Seedance 2.0 accepts up to nine images plus video and audio references, or Hailuo H3 does. Control here runs through frames and text, not through a pile of source material.

Five recurring complaints, and what to do about each

"The clip is too short"

The most common complaint, and the most accurate one: eight seconds is the ceiling of a single run. The fix is editing logic rather than prompt magic — generate a scene, take its final frame, drop it in as the first frame of the next run and continue the action. The cut is seamless because the model starts from exactly the image the previous scene ended on.

How to do it here: in Veo 3.1, upload the still from your previous clip into the first-frame slot and describe what happens next. If you need one uninterrupted take, Kling 3.0 runs up to 14 seconds with built-in multi-shot storyboarding, and Wan 3.0 up to 15.

"You need a VPN, a foreign card and a subscription"

In hands-on write-ups this takes up half the text: registration, a foreign IP, payment, a free tier of a few clips per month. None of it says anything about the model — these are the terms of direct access to Google's own products.

How to do it here: the studio opens in a normal browser, sign-in is Telegram, email, Yandex ID or VK ID, and generations are paid with tokens. No VPN, no foreign card, no separate Veo subscription; the cost of each run is shown on the model card.

"Lettering, logos and infographics fall apart"

Confirmed: video models hold small text, digits and brand marks badly — letters shimmer and mutate across the shot. That is a weakness of the generation, not a bug in this particular version.

How to do it here: build the lettered frame as a still first — GPT Image 2.5 Sunburst and Nano Banana Pro render readable text — then feed that image into Veo 3.1 as the first frame and ask only for camera and light movement, not for a redraw. Overlay titles are still safer to add in an editor.

"A short prompt gives a dull result"

Also true, and the easiest complaint to fix. Prompt breakdowns converge on one shape: who is in frame → what they do → where and in what light → how the camera moves → what it sounds like. Describe background sound explicitly: stay silent about it and the model invents something, occasionally a studio laugh track nobody wanted.

There is one specific trap: a line in quotation marks is often read as a request for subtitles. Write dialogue after a colon, without quotes, and keep it short — a long monologue gets rushed to fit the clip.

How to do it here: two working templates for Veo 3.1 —

Cinematic shot: a barista sets a white cup on a wooden counter, steam rising in slanted morning light, camera pushes in slowly. Sound: the hum of an espresso machine, muffled voices in the room, no music.

A presenter in a bright studio looks into the lens and says: today I will show how to shoot an ad without a film crew. Static camera, soft frontal light. Sound: a quiet room, no music, no subtitles.

"It refuses to generate famous people"

Scenes with recognisable public figures are rejected by provider-side moderation — that is a rule, not a failed generation. Reviews reporting "an error on celebrities" are describing exactly this.

How to do it here: describe the type instead of the name — age, look, clothing, manner — or upload your own frame with the person you need. If a character has to stay recognisable across several clips, build them in Seedance 2.0 from reference photos: it takes up to nine images and holds a face far better than a text description.

Which model fits which job

Choosing between the two main contenders has its own breakdown, Veo 3.1 versus Kling 3.0, and the wider field is covered in the best AI video models round-up.

Building your first Veo 3.1 clip in five steps

Step 1. Pick the mode

Open Veo 3.1 and start on Fast: it is quicker and cheaper, and while you are still hunting for the idea, picture quality rarely decides anything. Switch to Standard once the scene, the prompt and the duration are settled.

Step 2. Set framing and resolution

16:9 for YouTube, a site or a deck; 9:16 for Reels, Shorts and stories. Resolution: 720p is fine for a draft, 1080p is the working choice for publishing, 4K is worth it when the shot goes into an edit with cropping or onto a big screen.

Step 3. Write the prompt to the pattern

Who → what they do → where and in what light → camera movement → sound. The more concrete each part, the less the model improvises. Put dialogue after a colon, keep it to one sentence, and name the background sound outright.

Step 4. Add frames if you need control

The first frame sets where the scene starts, the last frame sets where it ends. Both slots are optional — without them the model builds the shot from scratch. Continuing a previous scene? Put its final still in the first slot.

Step 5. Choose the duration and run it

4, 6 or 8 seconds. Four is enough to test an idea: a scene that does not work at four seconds will not fix itself at eight, it will only cost more. Chaining several runs together is easier on the canvas.

Common errors and fixes

Where to go next

Veo 3.1 earns its place when you need a short, dense, audible shot — an ad, a teaser, an intro, a talking presenter. As soon as the task turns towards length, repeatable characters or precise motion control, the neighbouring models win, and in the studio they are open in the same window. Starting from a photograph rather than text? See how to animate a photo with AI. Want to push prompts for controlled animation? Read the Kling 3.0 guide.