4.36 / 5
Quality 4 · Cost 5 · Speed 4 — highest
AI video tools, tested by actually running them.
Six briefs — a drama scare scene, a rainy-night life scene, a sci-fi portrait, a creature clip, a food-hall promo — fed word for word into three tools. Every run logged: credits, retries, wall-clock time, export specs, and watermarks verified frame by frame. What separates them is not whether they work. It is fidelity, format, cleanliness and cost.
Brief shown: T1 — “Midnight Signal,” Ep. 1, convenience-store scare scene
Kling 4.0 · Standard plan
Veo 3.1 Quality via Flow · Google AI Pro
Runway Gen-4.5 · Standard plan
True vertical frame. The only tool that fully rendered the scripted beat — a second figure in the CCTV grid, then an empty store.
True vertical frame. The beat landed only in part — a distant figure, not “a person behind her” — but the face is the best of the three.
Text-to-video came back landscape — the 9:16 control was not selectable. The scare beat could not be verified from stills.
Quality 4 · Cost 5 · Speed 4 — highest
Quality 4 · Cost 3 · Speed 3
Quality 3 · Cost 2 · Speed 4
At the introductory price; $12.00 at renewal. Zero retries in the battery.
Calculated at the $19.99/mo list price of Google AI Pro.
And the minute is landscape — worth less to a vertical workflow.
Bottom-right on standard exports, frame-verified by us. A no-watermark download path on Standard is still unverified.
Exports clean across all six briefs.
Exports clean across the briefs it ran.
1080 × 1920, as briefed.
720 × 1280 to 1080 × 1920, as briefed.
In Gen-4.5 text-to-video. Vertical exists on Runway only via image-to-video (Gen-4 Turbo) — a different workflow, not a setting.
Wall-clock across briefs; zero retries.
Almost all of it queue — one brief waited ~30 minutes.
Fastest single run of the battery: ~3.5 minutes.
Pick Kling for vertical drama and serialized shorts on a budget — if the watermark is acceptable on your channel.
Pick Veo for commercial or client-facing clips, faces that must stay clean — and budget for the queue.
Pick Runway for speed on simple scenes — and enter through image-to-video for anything vertical.
Scored over the 70% of our protocol we have logged evidence for; languages, rights and editing control are unscored until their runs are done.
Kling tells the story best and charges the least for it. It was the only tool that fully rendered the scripted scare beat, it took the strongest single food frame of the test, and at $9.54 per finished minute it undercuts both rivals. Its catch is printed on every export: a “KlingAI 4.0” watermark, bottom-right — which is why, if Kling’s Standard tier turns out to have no clean-download path, the commercial ranking flips.
Veo is the clean professional. The strongest faces, the only tool with zero garbled on-screen text across all six briefs, and no watermark — paid for in time, with queues that stretched to roughly half an hour, and one real consistency failure when our creature grew extra tentacles mid-clip. Runway is fast and competent but pointed the wrong way: its text-to-video is locked to landscape, so a vertical pipeline has to enter through image-to-video instead. Three tools, three different compromises — the table above is the whole argument.