How to Make the Rumpelstiltskin AI Tiptoe Dance Video (Photo to Video)

You have seen the clip: a little hunched man in a pointed cap, dancing on tiptoe, captioned as a lost scene from a 1987 fantasy film. There is no such film — the footage is AI-generated. This page is how you make your own version with your own face in it, with the numbers and the exact prompt rather than a sales pitch.

Quick answer

• What you need: one photo of an adult, facing the camera, whole face visible.
• What you get: an 8 second vertical MP4 (9:16), 480p Standard or 720p HD, no watermark, silent.
• What it costs: 220 credits for Standard, 460 for HD. Packs start at $4.99, bought once, never expiring. No subscription, no free tier.
• How long: about three minutes per run; you can close the tab.
• Where: the generator, which runs in the browser on a phone or desktop.

Step 1 — Pick the staging

Two stagings exist, and both are our own AI-generated clips rather than footage from anywhere:

| Staging | What it shows | | --- | --- | | Candle-lit stone room | The classic version: flagstones, an iron candelabra, a heavy door, the little man spinning and hopping in the middle of the room | | Tavern table | The same dance performed on a lamplit tavern table, tankards and candle lanterns around him |

Pick one and the rest of the form follows it. The clip is what your photo gets fitted onto, so the choice decides the lighting, the setting and the camera moves.

Step 2 — Add your photo

Use one photo per dance, and follow these rules — they are the same ones the automatic safety check enforces:

| Works | Fails | | --- | --- | | One adult, facing the camera | Children or anyone under 18 | | Whole face and hairstyle visible | Sunglasses, masks, veils, deep shadow | | Even, natural light | Heavy filters, beauty mode, motion blur | | A sharp, recent photo | Group photos, tiny faces, drawings |

The photo is screened before you are charged. If it is refused, nothing is generated and nothing is taken.

Step 3 — Choose the look, then generate

| Option | What it does | | --- | --- | | Face & hair | Takes your face and hairstyle and keeps the dancer's own little body, cap and outfit. This is the one that looks most like the meme. | | Full look | Also takes your body shape and clothing from the photo, so the dancer is recognisably you head to toe. |

Then generate. Rendering takes about three minutes; the finished video appears in the page and in My videos, where it also waits if you closed the tab.

The prompt we actually use

Nothing here is a trade secret, and seeing it is the fastest way to understand what the model is being asked for. The run is an all-in-one recast: our clip goes in as the reference video (@Video1), your photo as the reference image (@Image1), and the prompt is fixed:

Recast @Video1: replace the dancer with the person in the reference image.
@Image1 replaces the little man dancing on tiptoe in the candle-lit stone room.
Take only the face and hairstyle from the photo; keep the dancer's original little
body, pointed cap, outfit and every pose and motion from @Video1.
Keep everything else in @Video1 unchanged: the same camera moves, cuts, timing,
framing, setting, lighting and every step of the dance.
The face stays sharp, front-facing and consistent in every shot.
No text, subtitles, logos or watermarks.

Two things follow from that shape, and they explain most of the output you will see:

1. The choreography is borrowed, not generated. Because the clip is the reference, every run returns the same dance — consistent, rather than a fresh improvisation. It also means a single bad frame in the source clip shows up in every video made from it.
2. "Face and hairstyle only" is doing a lot of work. That instruction is why the body stays small and gnome-shaped instead of the model trying to fit a full adult into the frame.

Posting it so it performs

• Add the audio inside TikTok, Reels or Shorts. The clip is silent on purpose, and using the in-app sound is what connects your post to the trend.
• Show your original photo for about a second first, the way the trend does — the before/after is most of the effect.
• Turn on the AI-generated content label. The output is synthetic and looks realistic; every platform that offers the label expects it here.
• Keep it under 15 seconds so the loop feels tight; an 8 second clip loops cleanly.
• Do not caption it as a real film, even as a joke. That is the part of the trend that annoys people, and it is factually wrong.

Troubleshooting

| Symptom | What it means | | --- | --- | | "This photo could not be used" | The safety check refused it — usually a minor, a public figure, or sexual content. Try another photo. | | "Safety check temporarily unavailable" | The screening provider could not be reached. Nothing was charged; try again in a few minutes. | | Render takes longer than three minutes | Normal at busy times. You can leave the page; the video lands in My videos. | | Render failed | The credits are returned automatically, with no ticket to open. | | The face looks off in one frame | The output inherits the source clip's weakest moment. Re-running the same setup gives a different result. |

FAQ

Honest notes

• No free tier. Every render costs real provider money, so credits are required. The account is free, and you can watch real renders on the home page before buying anything.
• The 1987 film does not exist. No studio ever made that scene, and the caption was invented. The full explanation is here.
• The templates are ours. Both stagings were generated for this tool; no frame comes from the viral clip or from any film.
• Independent. Rumpelstiltskin AI is not affiliated with any studio, streaming service, artist or platform, and the faces in our examples are our own AI-generated portraits.

Ready to try it? Make your video here — one photo, about three minutes.