The cohort
The 50 clips break down as roughly: 18 social ads (5–15 second hooks), 14 product B-roll cuts (5–10 sec, no character), 9 narrative sequences (15–30 sec with character continuity required), and 9 explainer cutaways (mixed). Clients ranged from DTC e-commerce to B2B SaaS to one music video. We did not cherry-pick the providers — every job started with a head-to-head test on a single shot before we picked a tool for the project.
Sora wins: the prompt-fidelity job
When the brief is a paragraph and you need the output to match the brief, Sora has the highest hit-rate. We measured "first-prompt acceptable" rates roughly: Sora ~52%, Kling ~31%, Runway ~28%. That gap is the difference between "ship after one render" and "ship after six."
Sora also handles physics that the others fluff — water, fabric in motion, hair against wind, things tipping over and bouncing. If the brief includes any of those, Sora is the default.
Where Sora lost: API throughput. Bundled with ChatGPT means the production-pipeline use case is awkward. Two of our four agency-style clients (rendering 30+ variants of the same hook for A/B tests) gave up on Sora for that workload because we couldn't automate it cleanly. They went to Runway.
Kling wins: the long shot, the continuity shot
Kling's longer maximum clip length and stronger frame-to-frame continuity is the right pick whenever the shot has a character or a recognisable product that must look like the same character or product across the duration. For our music video client this was the only model that didn't morph the lead's face mid-shot.
Kling's continuity advantage held up especially well on the 9 narrative sequences. We used Kling for 7 of those 9. Sora held the other two — but only because the shots were short enough that morphing wasn't an issue.
Where Kling lost: Western compliance. Two B2B SaaS clients had a hard "no Chinese-hosted models for marketing assets" rule from their legal team. The clips never made it to production for non-technical reasons. If you're delivering for an enterprise buyer, factor this in early.
Runway wins: the production pipeline
Runway rarely produced the best individual clip in our tests. It produced the most consistent pipeline. The agency client running 30+ ad variants ended up on Runway because the API was the most reliable, the editor inside the product cut down on round-trips to After Effects, and the multi-shot session made it easy to keep style consistent across a campaign.
Two of our DTC e-comm clients also went all-in on Runway because their workflow was "render this product clip with these 12 variations of camera motion" — that's editor work, and Runway is built for editors.
Where Runway lost: pure quality on long takes. The 30-second narrative shots were not its game. Use Kling.
The decision tree we use today
- Is this an enterprise-bound asset where Western compliance matters? → Drop Kling. Choose between Sora and Runway.
- Is the shot longer than 10 seconds with a character or product that must remain consistent? → Kling, unless rule 1 fired.
- Does the brief have specific physics (water, fabric, breakage)? → Sora.
- Are we rendering 10+ variants of one shot for A/B testing? → Runway.
- Default: do a 1-shot bake-off across all three. Pick the one with the best first-render result.
The bake-off step matters. Even after 50 clips we'd still get surprised — we'd predict Sora would win a particular shot and Kling would. The cost of testing all three on shot 1 is low; the cost of committing to the wrong one for the project is high.
The cost story
Cost-per-finished-clip varied a lot more than the sticker subscription price suggests. Sora's bundling with ChatGPT made it hard to attribute, but on Pro at $200/mo we comfortably finished ~25 acceptable clips/month. Kling Premier at $92 finished about 18. Runway at $76/seat finished ~22. Per-clip cost lands within the same band — the difference is reliability and pipeline fit, not raw price.
We keep our video generation cost calculator updated with these numbers; if your ratio of variants-to-final-clips is high, plug it in and the ranking can flip.
The summary card
- Sora — best per-shot quality, best physics, worst pipeline. Use for hero shots.
- Kling — best continuity, best long-take, blocked by some enterprise legal teams.
- Runway — best workflow tool, best for variant rendering and integration with the rest of an editor's stack.
We use all three on most months. If forced to pick one, our agency clients pick Runway and our brand clients pick Kling. Sora rarely survives as the only choice — but it's almost always the right choice for one shot in the project.