I built Rain Dance AI to turn one or two portraits into a 15-second video set on a wooden pier at sunset. You choose who appears. The site supplies the scene: a singer facing the camera, a partner waiting further along the pier, and a closing moment together toward the sunset.
The output is a vertical 9:16 MP4 at 768p. Solo and Duet are both 15 seconds. There is no prompt box, and new orders don't require an account.
That sounds straightforward until someone has to decide whether to pay for a video they haven't seen. A sample can show the setting and motion, but it can't tell them how their own face will look. A personal preview helps with that question, though it introduces a different problem: people may reasonably assume the paid video will continue where the preview stops.
The four-second preview and the paid video are separate generations. That needs to be obvious on the page, especially when someone likes their preview enough to buy a pack.
Choosing who appears in the scene
The first choice is Solo or Duet. I prefer describing the jobs of the two characters because "person one" and "person two" leave too much to guess.
| Mode | What you upload | What each person does |
|---|---|---|
| Solo | One adult portrait | Your photo supplies the singer; the other character is a fictional adult partner. |
| Duet | Two separate adult portraits | The first photo supplies the opening singer. The second supplies the partner waiting on the pier. |
These are story roles, not gender labels. If you want the other person to lead the scene, change which portrait goes into the singer's slot. You don't need a photograph of both people standing together.
The distinction matters before upload. Someone choosing Solo might otherwise expect a video containing only one person. Here, Solo means personalizing one character in a scene that still includes a partner. Duet gives you a portrait input for both characters.
I kept the setting fixed because it makes the choice easier to explain. Someone who wants a different location or a custom sequence will need a different workflow. This page is for a particular scene, and the sample should make that scene clear before anyone pays.
Picking photos that give you a useful starting point
Each upload accepts JPEG, PNG or WebP, up to 10 MB. Use one consenting adult per image. A sharp, front-facing portrait in even light is the simplest starting point; keep the whole head in frame and avoid sunglasses or heavy beauty filters.
Clothing deserves a moment of attention too. A close crop of a face provides little information about the rest of the outfit. If the clothes matter to the result, a waist-up or full-body photograph gives you more of that information in the input. It still doesn't guarantee an exact match. A jacket can change, a sleeve can move oddly, or details can disappear as the character turns.
For Duet, check each image on its own. One clear portrait doesn't compensate for a second portrait where the face is tiny or hidden. Before continuing, make sure each face is recognizable and belongs in the intended role.
This is also a useful moment to decide what would make the video disappointing. If a particular shirt or an exact facial detail is essential, the possibility of variation matters more than it would for a casual experiment. I would rather spell that out on the upload screen than leave someone to discover it after payment.
What four seconds can tell you
The free preview covers the opening of the scene. It gives you a moving result with your input, which is more useful for judging appearance than a still thumbnail. You can look at the face as it moves and see whether the general treatment appeals to you.
Four seconds cannot show how a 15-second clip holds together. It won't demonstrate the whole approach to the partner or the closing sunset moment. Even a convincing opening leaves most of the video unexamined.
There is another limit: buying a video doesn't extend the preview. The paid result is generated separately and can differ in facial appearance, clothing and movement. A preview you like is a reason to consider the format, not a promise that the next result will reproduce it.
The free allowance has limits as well. A workspace saved in your browser can use one four-second preview. There is also a limit of one per network per day, plus shared daily availability. The shared daily limit resets at midnight in Asia/Shanghai; the used allowance in that browser workspace does not reset. The page shows whether a preview is currently available. Allow about 15 minutes or longer rather than expecting an immediate playback.
I included the preview because somebody else's sample leaves you guessing about your own portrait. But I also want the explanation beside it to be hard to miss. A visitor shouldn't have to dig through a help page to learn that the paid clip will be a fresh attempt.
Watching the full example
The site also has a complete 15-second example made from fictional adult portraits, with the input portrait shown beside the result. It gives visitors a way to compare the starting image with what appears on screen.
Watch the whole clip once at normal speed. Then replay the parts where a face turns or the camera changes its view. A face that looks convincing in one frame may drift later. Check whether the outfit remains recognizable and whether the partner is present when the scene calls for them. Hands and transitions are worth another look if either draws your attention during playback.
The page describes the intended sequence in approximate five-second sections: the singer on the pier, the approach to the partner, and the ending together toward sunset. Those timings help explain the story. They aren't an exact shot schedule promised for every result.
I'd watch the complete sample even after trying a personal preview. The ending is part of what you're buying, and a four-second result doesn't show it. One sample also doesn't establish how often future videos will turn out well; it gives you one full result to inspect.
Sound and the finished file
The download is a vertical MP4, and the site doesn't add a watermark. Sound may be generated with the video. The sample's audio is generated sound, not the original Raindance recording, and the purchase doesn't include that recording or guarantee exact lip-sync.
That matters if the music is the main reason you want the clip. Someone expecting a finished edit with the familiar song already attached would be buying a different thing from what this site supplies.
After downloading, you can add audio from a TikTok or Instagram library where its use is permitted. Check the rules for your account and intended post. Choosing a song afterward also doesn't repair a mouth movement that fails to match it. Listen to the clip and watch the mouth before deciding how to share it.
Payment, delivery and coming back later
The current USD packs are $9.90 for one video, $24.90 for three, and $49.90 for ten. They are one-time purchases with no subscription or automatic renewal. Unused video allowances don't expire.
The ten-video pack reduces the price per attempt to $4.99, but it doesn't make any individual attempt more predictable. If you're deciding whether the format suits you, look at the example and the available preview before comparing pack sizes.
New orders use a delivery email and a private order link instead of an account. That link gives access to the order's videos and remaining uses. You can return to check progress and download completed files; you don't have to keep the original browser tab open throughout generation.
The email address therefore needs the same care that a login would usually get. Check it before checkout and keep the order message. If you lose the message, the site's "Find my videos" form can email links for your five most recent paid guest orders. Support handles older orders or previous accounts.
Keep the private order link out of social posts. It grants access to the order and its remaining uses. Share the downloaded video rather than copying the page address from your own order screen.
A failed generation and a disappointing video need different handling
A confirmed technical failure restores the allowance used for that attempt. It doesn't automatically send money back to your payment method. Cash refunds require review.
A completed video can still disappoint. You might dislike the expression or notice that the clothing changed. Another generation uses another allowance, and choosing a different photo doesn't guarantee an improvement.
For a serious quality problem, such as the wrong identity, a missing Duet partner or severe visual corruption, contact support with the receipt reference, the affected timestamp and a screenshot. Those details make the problem easier to locate than a general description of a bad result. Quality complaints receive manual review; a retry or refund isn't automatic.
If a request is still processing or under review, keep that request and check its status. Starting another one while the first is unresolved can leave you with two separate attempts to track.
Uploaded photos are retained in our storage for seven days, with scheduled cleanup removing expired files. An interruption can delay cleanup. External processing has its own retention terms, so our seven-day rule isn't a promise about every copy outside our storage. Completed private videos have no automatic expiry, and you can request deletion through support. Download the files you want to keep instead of treating an online order page as your only copy.
Rain Dance AI is an independent project, unaffiliated with the artists behind the song. I want the page to give someone enough information to decide against buying, too. If the preview leaves them needing a guarantee of exact likeness, adding a larger pack won't resolve that concern.
What would you need beside a four-second preview to judge whether a separately generated 15-second video is worth buying?
You can see the current flow and playable example at Rain Dance AI.
Top comments (0)