Veo 4 Video Generator
Google has not shipped Veo 4. Status checked 4 September 2026. Until it does, every Veo scene in Zebracat runs on Veo 3.1, the current top model in the family, and the finished video still comes out fully edited with voiceover, captions and music. Veo 4 will run here the day it launches.
.svg.webp){kind=logo}
.svg)
.svg)
.svg)
.webp)
.svg)
.webp)
.webp)
.webp)
.svg)
The model at a glance
Videos generated with this model
Every clip below was generated in Zebracat with the exact prompt shown. No cherry-picking, no post-production.
What is this model?
Status, 4 September 2026: Google has not announced, named or documented a model called Veo 4. Google DeepMind's Veo page, the Gemini API docs and Vertex AI all list Veo 3.1 as the current model. There is no Veo 4 model ID, no model card, no pricing page and no changelog entry. We check these sources and update this line whenever it changes.
That matters because almost every other page ranking for "Veo 4 video generator" tells you the opposite. They quote 4K, 30-second clips, multi-angle generation and voice-cloned avatars as if they were on a spec sheet. None of that is sourced to Google. It is guesswork, and some of it is guesswork copied from other guesswork. The sites making those claims offer a "Try Veo 4" button that generates with something else.
What Google has actually done since Veo 3 is ship point releases. Veo 3.1 arrived on 15 October 2025 with reference images, scene extension and better audio. Veo 3.1 Fast and, on 3 April 2026, Veo 3.1 Lite added cheaper tiers, and a separate Veo upscaler now lifts any clip to 1080p or 4K. At I/O in May 2026 there was no Veo 4 at all. Google used the keynote for Gemini Omni Flash and for renaming Flow to Google Flow. Some analysts read that as a sign the next generation of Google video may ship inside Gemini rather than under the Veo name. We do not know, and neither does anyone selling you Veo 4 today.
So here is the honest version of this page. Below is what Veo 3.1 does now, what it still gets wrong, what a Veo 4 would need to fix, and how Zebracat handles the switch so you never have to think about it. When Veo 4 ships, this page becomes its product page, same URL.
What our Veo 3.1 runs show, and what Veo 4 has to fix
- Sound is the family's edge, and it is real. Spoken lines come back with usable lip sync in most attempts, with effects and ambience generated in the same pass, which no other model on our roster does as cleanly for a realistic English scene. Any Veo 4 that loses this loses the reason to pick Veo.
- Eight seconds is the wall. One generation is 4, 6 or 8 seconds. Extension chains shots, but only at 720p, and continuity across the join is hit and miss. Longer single takes are the most requested fix, and the most repeated rumor. Until it ships, 30-second takes go to Seedance 2.5.
- Props do not stay put. A microphone, a bottle, a phone: in busy scenes objects drift or vanish between the 4-second and 8-second mark. Object permanence is a model-generation problem, not a prompt problem.
- On-screen text is unusable. Veo 3 rendered broken letters. Veo 3.1 mostly refuses. Zebracat adds captions and titles in the edit instead, which is why the finished video has legible text and the raw clip does not.
- 4K is not free. Veo 3.1 renders 4K only at 8 seconds, and Google's separate upscaler is what most "4K Veo" output on the web actually is. Treat every "native 4K Veo 4" claim as a prediction.
- Peak-hour latency doubles. One to three minutes per clip is normal. At US peak hours the API has stretched to six. A new flagship model launches slower, not faster, so expect the first weeks of Veo 4 to queue.
What this model is best at, and where it falls short
Best at
What Veo 3.1 does well today. This is the floor a Veo 4 has to clear, and what your Veo scenes get in Zebracat right now.
- Dialogue with lip sync. Write the line in quotes and the character says it, mouth and timing matching.
- Sound design without a sound designer. Effects and ambience generate with the picture: rain, traffic, a door, crowd noise.
- Product and material shots. Glass, liquid, fabric and skin hold up at 1080p, which is why ecommerce and DTC creative runs on it.
- Reference-driven consistency. Up to three reference images lock a character, a product or a location across shots.
- Clean camera language. Dolly, push-in, handheld and rack focus are followed rather than interpreted.
- Vertical and horizontal in one model. 9:16 and 16:9 render natively, no cropping.
Not good at
Where Veo 3.1 falls short. These are the gaps people are searching "Veo 4" hoping to close, and where Zebracat routes around Veo today.
- Anything longer than 8 seconds in one generation. Longer stories are cut from several shots.
- Legible on-screen text, logos and UI. Added in the edit, never in the generation.
- Object permanence in busy scenes. Props drift or disappear across the take.
- Choreography with many people. Crowds and group action lose coherence.
- Real people and copyrighted characters. Blocked by Google's filters, and the filters tighten over time.
- Cheap iteration. Veo is a premium model priced per second, and a re-roll costs what the first attempt cost.
- Non-English dialogue. English is strong; other languages are inconsistent, which is why Zebracat generates voiceover separately in 170+ languages.
Which version of this model you get in Zebracat
Veo 4
Not released. Not announced. No model ID in the Gemini API or Vertex AI as of 4 September 2026. Third-party pages listing Veo 4 specs are describing something they have not used. Google's release rhythm was I/O in May 2024 (Veo 1), December 2024 (Veo 2), I/O May 2025 (Veo 3), then an off-cycle Veo 3.1 in October 2025 and Veo 3.1 Lite in April 2026. I/O 2026 came and went without a Veo 4. Read from that what you like. We will update this section the day Google names it.
Veo 3.1, 3.1 Fast, 3.1 Lite (what runs in Zebracat today)
The current family. 720p by default, 1080p, and 4K at 8 seconds. 4, 6 or 8 second clips, native audio always on, image to video, up to three reference images, scene extension at 720p, 16:9 and 9:16 at 24fps. Fast trades a little detail for speed and price. Lite drops reference images and extension and tops out at 1080p, built for volume. Zebracat's routing picks between them per scene. Full detail on the Veo 3 page.
Veo 3 (May 2025)
The first Veo with sound. 720p and 1080p, 8-second clips only, no reference images. Superseded by 3.1 for every scene we route.
Veo 2 (December 2024)
Silent, 720p, 5 to 8 seconds. Historically interesting, not something we run.
What could replace "Veo 4"
At I/O 2026 Google announced Gemini Omni Flash, a Gemini-family model, and no new Veo. If Google folds video generation into Gemini, the next flagship may never carry the Veo 4 name. Zebracat integrates whichever model Google ships, under whichever name, and this page will say which.
This model inside a full video pipeline, not a download button
You do not need to wait for Veo 4 to get the benefit of Veo 4. That sounds like marketing, so here is the mechanism.
Today
You write a script, paste an idea or drop in a blog URL. Zebracat breaks it into scenes and picks a model for each one. A line of dialogue with lip sync goes to Veo 3.1. A 30-second unbroken product story goes to Seedance 2.5. A close-up performance goes to Kling 3.0. Stills come from Nano Banana. Each prompt is rewritten in that model's own grammar before it runs. You can override any scene by hand.
The day Veo 4 ships
We add the model to the router with its real specs, costs and failure modes, measured on our own test set, not copied from a launch post. From that point the scenes Veo 4 wins go to Veo 4. Your scripts, your brand kit, your voice clone and your templates do not change. If Veo 4 turns out to be better at long takes but worse at faces, the router sends it long takes and keeps faces on Kling. That is the point of not being loyal to one model.
After the shot, every time
Clips land in one tool, not a downloads folder. Voiceover in 170+ languages or your cloned voice, captions synced word by word, music, on-screen titles the model could not render. Swap a scene, rerun it on a different model, change the hook, nothing else moves. The finished video runs up to 5 minutes at 9:16, 16:9 or 1:1, or schedules straight to TikTok, YouTube and Instagram.
How to prompt this model
There is no Veo 4 prompt guide because there is no Veo 4. What we can say is that Google has kept Veo's prompt conventions stable from Veo 3 to 3.1, so the syntax below is the best available bet for whatever ships next. Zebracat writes these prompts for you either way.
- Put dialogue in quotes and name the speaker. The barista says, "Your oat latte is ready." Unquoted speech gets narrated or dropped.
- Label sound explicitly. SFX: espresso machine hiss. Ambient noise: light rain on the awning. Veo reads those labels. Vague "add sound" prompts produce generic sound.
- Timestamp beats when order matters. [00:00-00:03] she looks up. [00:03-00:08] she smiles and walks out. Without timing, beats overlap.
- One camera instruction per shot. Slow push-in, or handheld follow, or static. Two moves in one prompt fight each other.
- Keep the frame simple past four seconds. Fewer props, fewer people. Object permanence degrades with clutter, and that will still be true on launch day of any new model.
- Do not ask for text. No signs, no labels, no packaging copy. Add it in the edit.
- Use reference images for anything that must match across shots. Character, product, location. Three images, consistent lighting.
- Say the aspect ratio in the prompt and in the setting. 9:16 for social, 16:9 for YouTube. Veo composes differently for each, and a crop later loses the framing.
- Re-roll the prompt, not the seed. A failed Veo generation usually failed on ambiguity. Rewrite the sentence that could be read two ways.
This model vs other AI video models
Spec-level comparison of the models you can run in Zebracat. Every model here is selectable in the same editor.
1,000+ 5-Star Reviews from Creators, Marketers & Makers
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
Questions about this model, answered
Is Veo 4 released?
No. As of 4 September 2026 Google has not announced or released Veo 4. Google DeepMind, the Gemini API and Vertex AI all list Veo 3.1 as the current model. We update this answer when that changes.
When is the Veo 4 release date?
Google has not given one. Past releases were May 2024, December 2024, May 2025 and October 2025, but I/O 2026 in May brought Gemini Omni Flash and no Veo 4, so the old cadence is not a reliable guide. Any site quoting a date is guessing.
Why do other sites say I can use Veo 4 now?
Because the query gets traffic. Those pages generate with Veo 3.1 or another model behind a "Veo 4" button, and their spec tables are unsourced. Check whether a page dates its claims and links to a Google source. Almost none do.
Will Veo 4 have 4K, 30-second clips or avatars?
Unknown. Those are the most repeated rumors and none is sourced to Google. What exists today: Veo 3.1 renders 4K at 8 seconds, extends clips at 720p, and Google's separate upscaler lifts any video to 4K. Avatars and voice cloning are not Veo features in any released version.
Will Veo 4 be in Zebracat?
Yes, on the day Google makes it available through its API. Zebracat already runs Veo 3.1, Kling 3.0, Seedance 2.5 and Nano Banana through one router, and adding a model does not change anything on your side. This page becomes the Veo 4 product page at the same URL.
What should I use instead of Veo 4 right now?
Veo 3.1 for dialogue, sound and product realism. Seedance 2.5 for takes longer than 8 seconds. Kling 3.0 for close-up performance. Zebracat picks per scene automatically, so the practical answer is: write the script and let the router decide.
Might Google skip the Veo 4 name entirely?
Possibly. I/O 2026 was about Gemini Omni Flash, a Gemini-family model, and Flow became Google Flow. If video generation moves into Gemini, the next flagship may not be called Veo. We will state plainly on this page what shipped and under which name.
Will Veo 4 be free?
No released Veo model is free at production quality, and there is no reason to expect the next one to be. In Zebracat, premium video models need a paid plan. The free plan covers script writing, AI voices and lower-tier image models without a card.
Can I use Veo videos commercially?
Yes. Videos made on paid Zebracat plans carry a full commercial license for ads, social and client work, whichever model rendered each scene.
How will I know when this page changes?
The status line at the top carries the date it was last checked, and the Versions section logs each change. If the top of this page still says "not released", it is not released.
Other AI video models in Zebracat
One editor, every model. Zebracat routes each scene to the model that fits, or you pick one yourself.