When you buy through marked links (†) on PickaSi, we may earn a commission — it never changes a score. How we make money →
Same brief, same prompt, word for word — in every tool. Credits, retries, queue time and watermarks logged. What we haven’t run is labeled vendor claim, never scored.
Brief T1 — the scare scene: a convenience store at night; the CCTV grid shows one figure too many; she turns around — the store is empty.
The storyteller of the three — held back by a watermark its own plan page claims to remove.
Axes, clockwise from top: story fidelity · native vertical · on-screen text · watermark-free · cost · speed · terms
The clean export — best faces, zero text artifacts, no watermark — paid for in queue time.
Axes, clockwise from top: story fidelity · native vertical · on-screen text · watermark-free · cost · speed · terms*
Fast, clean and commercially frictionless — enter via image-to-video, never ask its text-to-video for vertical.
Axes, clockwise from top: story fidelity · native vertical · on-screen text · watermark-free · cost · speed · terms
Serialized shorts, story-first. Why this order →
Verified on a real product photo, Nov 1, 2026. The runs behind this →
None of these three. A 10-minute talking-head video is an avatar-tool job, and we won’t rank tools we haven’t run on it. Our current evidence is the Sprint B free-tier test — Synthesia, HeyGen and ElevenLabs on the same frozen script — and the paid-tier shootout is still ahead of us.
Where did the single score go? We retired the weighted total on Nov 2, 2026. One number hid the only trade-off that matters here — the cheapest tool watermarks its downloads — so every tool now gets seven written verdicts and a shape instead. The old provisional score is kept in the shootout’s appendix, formulas and all, for the record.
Brief T2 — the life scene: rainy night; a delivery rider hands a bag of steaming buns to a customer under a shop awning. One generation per tool, zero retries.
Hand-over beat: complete in both — Kling framed it steadiest; Veo gave it a process (wave → receive) and the strongest neon mood. Kling's panel above renders its real export artifact: the “KlingAI 4.0” watermark, frame-verified. Timings on this brief: Kling ~6 min; Veo ~30 min, almost all queue.
The full battery: We Ran the Same Six Prompts Through Kling, Veo, and Runway. Here Are the Seven-Dimension Verdicts.
The Sora Migration Center: the honest map of what happened, what to save before it disappears, and where each job actually goes now, based where possible on tools we have run ourselves, on the same prompts, with the costs logged.
The evidence article behind the comparison frame in the Sora Migration Center: six briefs, three tools, the same prompt, word for word, in every tool — plus the Nov 1 verification runs.
The Sprint B free-tier test: ElevenLabs, Synthesia, and HeyGen run the same frozen 228-word explainer script on $0 plans — one clean finish, one watermarked demo, one 60-second wall. Paid tiers explicitly not tested.
What each program pays according to its own current terms, how long it pays, what we could not verify from public pages, and which widely-cited figures are stale.
the published protocol and weights behind every score.
how PickaSi makes money, plainly.
what we do when we get something wrong.
New hands-on test results, tool price changes and ranking updates — sent by email, free. Hosted on beehiiv; unsubscribe anytime.