7 AI Video Models, Same Prompt, Real Prices

The bill

Exact settings and price per second

Clip lengths differ — both Veo 3.1 tiers always return 8 seconds and Grok's minimum is 6 — so the fair price is per second of video. Settings are read back from the stored generation inputs; prices are what our wallet was actually charged. Render time is submit-to-finished, averaged over the three prompts.

Round 1 · Motion & physics

A kickflip over a neon puddle

A kickflip is a hard test: the board has to rotate a full turn under airborne feet and land back under them. Most models fake it.

• Seedance 2.0. A clean kickflip — the board visibly rotates under the skater's feet mid-air, then lands.
• Wan 3.0. Also lands it: the board flips over the puddle with a believable splash, on a convincing Tokyo street.
• Kling 3.0. Lands it too: the board flips under the skater's feet over the puddle, comes down in a pink-lit splash, and rolls away.
• Veo 3.1 Fast. The surprise of the test: the board flips a full turn through a big slow-motion splash and the skater lands back on it — for $0.30.
• Veo 3.1 Quality. The best atmosphere and a huge slow-motion splash, cut like a trailer — but the trick itself is hard to read.
• Runway. Low, cinematic angle with a big splash, ending on a face close-up. No readable flip.
• Grok Imagine. The most stylised — a magenta splash and an airborne frame — with loose physics.

Seedance, Wan, Kling and Veo 3.1 Fast actually perform the trick. Veo 3.1 Quality wins on atmosphere, not on the move.

Round 2 · Character & coherence

A golden retriever surfing a giant wave

Can the model keep one character — a dog, its sunglasses, its board — consistent while the camera moves and water flies?

• Veo 3.1 Quality. A GoPro fisheye shot from the board, inside the barrel, dog front-on with its tongue out. The most shareable clip of the test.
• Veo 3.1 Fast. Almost the same shot as Quality — POV from the board, inside the barrel, sunglasses on, water drops on the lens — at a quarter of the price.
• Grok Imagine. The most energetic — crouched into the turn, spray everywhere, lens flare — and the dog stays on the board.
• Runway. A clear dog on a board in the barrel with consistent anatomy and sunglasses that stay put.
• Wan 3.0. Rock-solid consistency in every frame, but a nearly static camera — the least energy.
• Seedance 2.0. The most cinematic arc: wide shot, POV in the barrel, then a wipeout as the wave swallows the camera.
• Kling 3.0. A low GoPro shot from the nose of the board inside the barrel; the sunglasses slide from forehead to eyes and the dog never leaves the board.

Either Veo tier for the viral shot (Fast got it for $0.30), Grok for energy on a budget, Runway for the cleanest result per cent.

Round 3 · Fire, fluids & objects

A chef tossing a flaming wok

Fire and steam are hard to render, and food and cookware tend to morph mid-clip. We watched for both.

• Veo 3.1 Quality. The most believable kitchen: noodles tossed mid-air above the flame, a second cook behind, and real kitchen sound.
• Veo 3.1 Fast. A real line kitchen with cooks behind, a tall burst of fire with sparks, then noodles tossed out of the wok — the wok and the food stay the same throughout.
• Seedance 2.0. Ignored “close-up” and opened wide — then a wall of fire engulfs the whole frame. The most dramatic fire.
• Wan 3.0. Noodles launched out of a flaming wok, staff in the background, food that stays the same food.
• Runway. An action-movie edit: wok toss, vegetables in the air, ending on an intense close-up of the chef.
• Kling 3.0. A fireball that swallows the frame around the chef, then a steaming toss — and the wok and the noodles stay the same throughout.
• Grok Imagine. A food-commercial look with huge flames, but the vegetables turn into chunks of meat mid-clip.

Both Veo tiers, Wan and Kling keep objects stable; Grok morphed what was in the pan.

The verdict

Which model to pick for what

• Veo 3.1 Quality: Best overall. The most believable scenes, real ambient sound, and an editor's sense of framing.
• Veo 3.1 Fast: Best value. Veo's look and sound at a quarter of the Quality price — and it landed the kickflip Quality didn't.
• Seedance 2.0: Best physics. Landed the kickflip cleanly and produced the most dramatic fire, with sound.
• Kling 3.0: The most reliable all-rounder. It landed the kickflip, kept the dog on its board and the wok a wok, with sound.
• Wan 3.0: The most consistent — subjects never morphed — but the slowest render and the calmest camera.
• Grok Imagine: The most energetic, social-ready look for pennies — with looser physics and one ingredient morph.
• Runway: The cheapest by far. Cinematic, coherent clips in every round for six cents — but no sound.

How we tested — and what was not equal

Held equal for every model:

• The identical prompt text, shown under each round.
• 16:9 framing and 720p, where the model offers a choice.
• Sound on wherever the model can generate it.
• The first take — no rerolls, no cherry-picking.
• The price is what our own wallet was charged on the day of the test.

Not equal, and how we handled it:

• Clip length. Both Veo 3.1 tiers always return 8 seconds and Grok's minimum is 6; everyone else ran 5. Prices are compared per second, and the grids show the first 5 seconds of every clip.
• Defaults. Kling 3.0 defaults to 1080p with sound off; we switched it to 720p with sound so it ran at the same settings as everyone else.
• Runway has no sound option, so its clips are silent by design, not by setting.
• One retry. One Seedance 2.0 generation failed before rendering for a capacity reason on our side; it was refunded automatically and re-run once with the same prompt.

Prices change as models update; the live price for any model and setting is on its page.

FAQ