How to Make a Rumpelstiltskin AI Video From Two Photos

How to Make a Rumpelstiltskin AI Video

Making a Rumpelstiltskin AI video comes down to two jobs. The template supplies the straw room, the candlelight, the dance and the camera move. Your job is to supply two clear photos and pick a length. Here is the full process, including the mistakes that waste the most time.

Step 1: Prepare two photos, not one

The clip has two performers, so it needs two uploads. One slot is for the seated woman, the other for the little man who dances in front of her. The two slots are separate, so a single group photo will not work even if both people are in it.

For each slot, use a recent photo with even lighting and one person in frame. Avoid sunglasses, hats, heavy beauty filters, or anything covering the face. A front-facing portrait with the head and shoulders clearly visible gives the model the most to work with.

You do not need to match the pose. Sitting, standing, smiling or looking serious in your photo is fine. The seated posture, the arm movements and the timing all come from the template, not from your uploads.

Step 2: Check what the template keeps

The template is one continuous shot, and it is deliberately fixed. Everything below comes from the source clip and survives the swap untouched: the straw-filled spinning room and the stacked bales; the candle-lit colour grade and the wide two-shot framing; the camera move and the timing of the dance.

Only two things change: the identity, hair and wardrobe of each figure. That is why the result stays consistent between runs, and why you can repeat a batch of posts without the scene drifting.

Step 3: Pick 10 or 15 seconds

10 seconds captures the core gesture and approach. 15 seconds includes the longer walk and camera move. Watch the free preview before you pay. The finished file is a vertical 9:16 clip labeled AI-made.

Step 4: Generate and review

Submit the task and keep the page open. Most tasks finish within a few minutes, but video models queue under heavy load, so a task can take longer during peak hours.

When the file is ready, watch the entire video before downloading or publishing it. Look at faces, hands, motion, background details, and anything that could make viewers believe the clip records a real event.

What to avoid

Two different people in two slots. Uploading the same photo twice produces the same face in both roles and wastes the run.

Covered faces. Sunglasses, masks and heavy shadows all reduce how much identity the model can carry across.

Group photos. One person per upload. A crowd gives the model nothing to bind to.

Extremely low resolution. A small, blurry source photo cannot produce a detailed face.

Posting the result

The clip works on any feed, and the motion already matches the source beat, so it does not need extra editing. If you want to use it as a green screen source instead, export the render and composite your own background around the two figures.

A failed paid generation requests a Stripe refund when Stripe can process it. Each paid file includes one free redo.