Kling vs Veo: Google moved to Gemini Omni, so we tested Kling and Omni on the same shots

Most Kling vs Veo pages compare Kling with Veo 3 or Veo 3.1. Since October 22, 2026, Google's Gemini API no longer serves Veo 3.1: its deprecation page lists the three Veo 3.1 preview models as shut down and names Gemini Omni 1.2 Flash as the replacement. So this post tests the Google model you can call today. On November 14, 2026, Hook Studio added Gemini Omni 1.2 Flash, and we ran it through the same three shots we had already run on Kling 3 and Kling 4, and kept every take, including the ones that failed.

Veo is Google DeepMind's video model. Veo 3.1 came out on October 15, 2025, in the Gemini API, the Gemini app, and Flow. It added up to three reference images, a first and a last frame, scene extension, and native sound, in clips of 4, 6, or 8 seconds.

On May 19, 2026, at Google I/O, Google announced Gemini Omni, a family of video models that take text, images, sound, and video as input. Omni 1.2 Flash became generally available on November 5, 2026: clips of 3 to 15 seconds, vertical 9:16, image to video, reference images, a first and a last frame, editing by conversation, and what Google calls clearer lip sync, at $0.07 a second at 720p on Google's price list. Google says it is still improving on fast sports action, quick script changes, and complex physics. Every Omni clip carries Google's SynthID watermark.

That table comes from Google's deprecation page for the Gemini API, read on November 14, 2026. It names Gemini Omni 1.2 Flash as the replacement for each Veo 3.1 model. Google's own posts about Omni never say it replaces Veo, and Veo 3.1 is still listed on Google Cloud's Vertex platform. So the accurate version is: Google moved its video API from Veo to Omni.

Why Hook Studio added Omni, not Veo

Every clip in Hook Studio starts from a still you approved, which is how the person, the room, and the light stay the same from shot to shot. Google's own API takes an image for Veo 3.1, but the video provider Hook Studio runs on offers Veo 3.1 from text only, so it could not start from your still. Gemini Omni 1.2 Flash starts from the still, makes its own sound, and renders vertical. It went into Hook Studio's model menu on November 14, 2026, as "Gemini Omni · Google's newest".

Who makes them

• Kling is made by Kuaishou, the Chinese company behind the Kuaishou and Kwai short-video apps. Kling 4 launched on October 15, 2026.
• Veo and Gemini Omni are made by Google DeepMind. Google puts them in YouTube Shorts, Flow, the Gemini app, and Google Vids.

For the other makers, from ByteDance to Alibaba, see who makes Kling, Seedance, and Veo.

• Same start. Each shot began from one still we approved before any clip, so every model started from the same first frame.
• Same words. One prompt per shot, word for word the same for every model, in Hook Studio's organic style: filmed on a phone by someone who was there.
• Same length. 5 seconds. Kling 3 renders only at 1080p, so it ran at 1080p. Kling 4 and Gemini Omni ran at 720p.
• Two checks. Hook Studio's realism check reads the frames and flags what looks fake. The lip-sync check compares the mouth with the voice, sentence by sentence. Whether a clip did what the prompt asked is our own read of its frames.
• One take each. No retakes and no picking the best of three.

The people are AI people from Hook Studio's premade cast: Camila, Marco, and Nia. Every clip starts on the still its shot was made from. The clips play muted: tap the speaker for sound.

Camila sits at a café table in Mexico City and says one line in the clip's own voice: no voice-over, no lip-sync pass after. Google lists native sound for Veo 3.1 and clearer lip sync for Omni 1.2 Flash, so this is the shot to start with. The prompt, word for word:

Seated at the café table, she is already talking to her friend across the table, calm and friendly, with a small natural nod; her hand stays near the cup. She says: "Same photo, same line, different AI model. Which one looks real?"

We made the same three clips in Spanish, from the same still, with the line "Misma foto, misma frase, otro modelo de IA. ¿Cuál se ve real?" Gemini Omni passed again, with the mouth in sync. The check noted a rounded "oh" with raised brows that stays within normal speech, and a short glance to the side. Kling 3 passed too, but its voice ran 0.5 seconds ahead of the mouth in the last second. Kling 4 failed with Camila's eyes shut for about three seconds while she talked. The Spanish takes play in the Spanish version of this post.

So the cheapest clip in the test was also the only one that passed and stayed in sync in both languages. That is one take per language. Kling 3 is the one to try next: it passed both too. Kling 4 kept the mouth on the voice but turned a calm line into a performance, in both languages.

Marco pours coffee from a glass carafe into a green mug, then sets the carafe down. No sound. Liquid, glass, and hands show how a model handles physics.

He is already pouring: the coffee streams from the carafe into the green mug and the mug fills about halfway; then he sets the carafe down on the counter beside the mug. Calm, unhurried, no talking.

Nia gets three instructions in one prompt: take one sip, set the mug on the shelf, then turn and wave with both hands. This shot tests whether a model follows direction, and whether the room stays the room while it does.

Three actions, in this order, starting at once: first she takes one sip from the white mug; then she steps toward the narrow wooden wall shelf at waist height and sets the mug down on it; then she turns back to the camera and waves with both hands, smiling. No talking.

All three did the three actions in the right order. What separated them was the room. In the still, the shelf is small, with one strap. Kling 3 grew it into a long two-strap shelf and lost the mug, and the realism check failed it. Kling 4 passed, but a yellow patch appeared in the painting behind her, which the check missed. Gemini Omni kept the shelf small with one strap: in our read of the frames, the only model in the whole test, Seedance included, to do that. Its notes: the shelf hangs at chest height, not waist height, and the wave is a little big. New decor also appears at the right edge as the camera moves: a eucalyptus sprig and books on the side table.

Prices are in credits. A top-up is $1 for 100 credits, so a 5 second Gemini Omni clip at 720p is about 60 cents, and a 5 second Kling 3 clip with speech is about $1.43. You see the price on a card before any clip runs. Veo 3.1 is not in Hook Studio.

In Hook Studio every clip starts from your approved still; the extra reference files in the table are what each model takes in its maker's own tools.

What the leaderboards say

On the Artificial Analysis video arena, read on November 14, 2026, Gemini Omni Flash 1.2 ranked 5th for text to video and 14th for image to video. Kling 3.0 ranked 22nd and 23rd. Veo 3.1 ranked 25th and 27th. We did not find Kling 4 on those pages. The arena ranks blind votes on short clips. It does not measure whether a clip followed your prompt, kept your character, or what it cost, which is why we run our own shots.

• A person talking to the camera, in English or Spanish: start with Gemini Omni, 60 credits at 720p or 89 at 1080p. It was the only model here that passed and stayed in sync in both languages. Kling 3 is the one to try next, at 143. Plan a retake on any talking clip.
• A product in someone's hands, with liquid or small details: Kling 4 passed the pour, at 134 credits. Gemini Omni poured well and then over-acted, so watch the last second of every Omni take.
• Several actions in a set order: all three did the order. Gemini Omni kept the room closest to the still; Kling 4 passed but changed the painting.
• One long take: Kling 4 reaches 30 seconds in Kling's own app. Gemini Omni stops at 15 and Veo 3.1 at 8 per clip. In Hook Studio, every clip is 4 to 10 seconds.
• You need Veo 3.1 itself: it is still on Google Cloud's Vertex platform. It is not in Hook Studio.

A reel does not have to stay on one model. In Hook Studio each clip runs on the model you pick for it, so a reel can open on a Gemini Omni talking hook and cut to a Kling 4 product shot. For Seedance on the same three shots, see Kling vs Seedance. For why the most realistic model is not always the right one, see realism vs control. To run a test like this on your own shots, follow how to pick an AI video model, and for a model for each kind of shot, see which AI video model to use.