Swap One Detail or Remake the Whole Structure?
You have a video that works, and you want a version of it: a new outfit, a new background, or the same winning structure selling a different product. Those sound like one job. They are not. Pick the wrong tool and you pay twice: once in credits, and again when the result quietly changes something you never asked it to change.
Riffkit's published Swap tests make this easy to see. They show the prompt, the cost, and what drifted. The lesson applies whichever tool you use.
Two different jobs
A swap keeps the original footage, its shots, and its rhythm, and changes one element inside it: the same clip in three outfits, or a living room turned into a beach boardwalk. If the video is already yours and one thing needs to change, a swap is the right tool.
A remake keeps none of the original footage. It studies the structure (the hook, the pacing, the shot breakdown, the script) and rebuilds it with new material: your product, your actor, your words. That is the job when you want to turn a proven format into your own video. Swapping the jacket on a video you do not own does not make it yours. Learning its structure and rebuilding it does.
Name the job first, then pick the tool.
Do the math: one swap is cheap, ten swaps are not
Riffkit's outfit test cost 2,340 credits for a 12-second clip on Seedance 2.5 at 720p: 13 billed seconds at 180 credits per second. Its background test cost 1,620 credits for an 8-second clip on the same setup.
For a single, targeted change, that is reasonable. The problem starts when the change multiplies. Riffkit's clothes-changer test says it plainly: ten garments on the same clip are ten swaps. That is 23,400 credits at Seedance 2.5 720p, or 6,500 at Seedance 2.0 Fast 480p.
Credits are not comparable between products, so convert them. Riffkit's smallest plan is $15 for 3,000 credits, or about $0.005 per credit. At that rate, ten swaps come to roughly $117 on Seedance 2.5, or roughly $33 on Seedance 2.0 Fast.
On Kopi, a 15-second video made with MiniMax H3 is 120 credits. The Starter plan is $29 for 2,900 credits, so about $1.20 each, and ten different videos are about 1,200 credits, or roughly $12. Kopi does not offer Seedance, and these are different jobs at different quality settings, so treat this as a budgeting estimate rather than a head-to-head test.
The useful question is not the headline rate. It is how many paid operations your job needs. One outfit change on your own clip: a swap wins. Ten variants of a format: a remake is usually far cheaper.
The drift checklist: name everything you want to keep
The most useful part of Riffkit's tests is the list of things the model changed without being asked. In the outfit test, a phone moved from her shoulder to her hand, and her shoes turned black. In the background test, a necklace disappeared, a red X sticker appeared, and her hair changed color.
The pattern is worth memorizing: anything you leave undescribed is free for the model to improvise. Before you generate anything, write a keep-list next to your change-list: the product, its color, the on-screen text, the accessories, and the background details that matter. Then name the one thing that changes.
Approve the structure before you pay to render it
With Kopi, you paste a TikTok or Douyin link, or upload an MP4 or MOV, and the agent breaks it down into hook, pacing, shots, and script. Before any credits are spent, you see a plan. You approve it or change it, and only then does generation start. The output is 9:16 vertical video at 768p.
That order matters for two reasons. The most expensive mistake in AI video is rendering the wrong plan: a good-looking video built on the wrong hook or scene order. Reviewing the plan catches that while it is still free to fix. And when one generated scene is wrong, you can redo just that scene instead of paying for the whole video again.
Which tool, when
| Your situation | Use | Budget for |
|---|---|---|
| The video is yours and exactly one element must change | A swap | One charge per clip, per change, plus reruns to fix drift |
| You want a proven structure with your own product and words | A remake | One plan you approve first, plus single-scene fixes |
| You cannot yet approve the hook, pacing, and shots on paper | Neither | Fix the plan first. It costs nothing |
Before you choose a tool, ask three questions. What does this exact change cost, by operation and engine? What does it cost when one scene is wrong? Can I see the plan before credits are spent?
Learn the structure, not the footage
One rule sits underneath all of this. A remake takes the hook, the rhythm, and the shot logic, and rebuilds them with your own material. It should not lift the original clip, its audio, its people, its text, or its logo. Use material you have the rights to, and label AI-generated content where the platform requires it.
If it is your clip and one thing needs changing, swap it, and write your keep-list first. If it is a proven structure you want to make your own, remake it, and approve the plan first.
Try it on Kopi: new accounts get 100 free credits, enough for a 10-second video. Paste a link to a video whose structure you admire and review the plan before you spend anything.