How to Make a Raindance AI Video

A Raindance AI video takes two photos and returns a short clip of those two subjects dancing on a pier at golden hour, with light rain starting to fall. On our platform that is a reference-to-video job: 194 credits for eight seconds at 720p, or 54 credits for a draft to check the setup first. The rest of this post is the setup, a prompt to start from, and what the clip will not do.

Prices and specs verified 2026-10-10 against our production model catalogue.

What is the Raindance AI trend?

It is a remix format. Kapwing's walkthrough describes it as two photos turned into a short music-video scene inspired by "Raindance" by Dave and Tems, and Media.io's page on the trend describes the same two-image setup. The scene is fixed and only the subjects change, which is why it works as a template: friends, couples, family and pets all get dropped into the same shot.

The trend is new and still moving. Google Trends data for "raindance ai" shows it appearing on 26 September, peaking on 3 October, and sitting at roughly 60 to 90 percent of that peak through 9 October. Treat the window as weeks, not months.

How do you make a Raindance AI video?

Use reference mode with two single-subject photos. Each photo should show one person or one animal, well lit, face visible, with nothing else competing for the frame.

1. Open reference-to-video and pick Seedance 2.0.
2. Upload the left subject first and the right subject second. The order matters, because the prompt refers to them by number.
3. Set the frame to 16:9 and the length to 8 seconds.
4. Paste a prompt that fixes four things: where each subject stands, the pier and the light, the rain, and how they move.
5. Run a cheap draft before the final.

Here is a prompt to start from. It is our own wording, it contains no real names and no song title, and we have not yet published a sample made from it, so treat it as a first draft to iterate on, not a tested recipe:

A cinematic 16:9 music-video scene. The two subjects from @Image1 and @Image2 stand side by side on a weathered wooden pier at golden hour, then move together in a relaxed, rhythmic dance as light rain begins to fall. Keep both subjects exactly as they appear in the reference photos: the same faces, hair or fur, markings and clothing, with no substitutions. Warm sunset light, gentle handheld camera drifting in, shallow depth of field, soft film grade.

The @Image1 and @Image2 tags are how our reference mode ties each photo to a part of the scene. Details are on the Seedance 2.0 page. If you want more control over wording, our text-to-video prompt guide explains the subject, action, camera, light and look structure used above.

What does a Raindance AI video cost?

Credits come straight from upstream cost (one credit is one cent of compute), and the same function the generator calls before submission produced these figures for a 16:9 frame with two reference images.

Seedance bills by pixels, not by the clip. A 16:9 frame costs more than a square crop and the same as a 9:16 frame, because only the pixel count matters. The mechanism is explained in our MiniMax H3 vs Seedance cost breakdown.

Reference mode on Seedance 2.0 tops out at 720p. It runs on the fast tier, so there is no 1080p or 4K option for this kind of job.

In currency, our credits cost roughly $0.03 to $0.05 each depending on the plan or pack, which puts the 194-credit take at about $6 to $10. Current rates are on the pricing page.

Subscribers also get a one-click version: the Raindance AI template uses fixed settings and a built-in prompt, and charges a flat 240 credits for an 8-second 720p clip. It is a convenience, not a different model, so if you are happy to write your own prompt, the manual route above is cheaper.

How should you test before spending 194 credits?

Run a 5-second draft on Seedance 2.0 Mini at 480p for 34 credits. Look for the failures that waste a final render:

• Faces merging or swapping. The two subjects blend into one person, or the left subject turns into the right.
• Side switching. Subjects change places mid-clip.
• Hands and feet. Extra fingers, feet that disappear into the planks.
• Outfit drift. Clothes change colour between the first and last second.

Use the draft to judge composition and who ends up where, not final quality. We have no quality comparison between Mini and the standard tier to share, so do not read a draft's polish as a prediction of the final. If the draft is wrong, fix the prompt rather than raising the resolution, because a higher resolution does not repair a wrong layout. If you are making a different trend, the same workflow is used in our hotel lobby AI video guide.

What will a Raindance AI video not do?

A few honest limits, because some of what you find online promises more.

• It will not reproduce the original video. Our generator creates new clips from reference images and a prompt. It cannot take existing footage and replace the people in it, so you will not get the original choreography or camera work.
• It will not include the song. We do not generate or license the original track. For a specific song, add it afterwards in TikTok, Instagram or CapCut.
• It will not check who is in your photo. Moderation on our side screens for explicit content, not for identity. Use your own photos or ones from people who agreed to appear, and skip celebrities and minors.
• It is not free. A local model would be, but every render here costs credits.

Which model should you pick for a Raindance AI video?

Pick Seedance 2.0 at 720p for the final, with Seedance 2.0 Mini at 480p for drafts. Both are available without a subscription and both accept up to nine reference images, far more than this trend needs.

Seedance 2.5 accepts up to 30 references and longer clips, which is overkill for two subjects and needs a subscription. Reserve it for group versions with many subjects. The Seedance 2.5 page covers when it is worth the extra credits.