Image to Video AI, No Restrictions on Adult Themes: Make a Still Move

A clip can only be as good, and as allowed, as the picture it starts from. Go through this list first:

1. Who is in the picture. A fictional adult you generated, yourself, or another adult who gave documented consent. Nobody else, and never a minor or anyone who looks under 18.
2. Whether they read as adult. If the still could pass for someone under 18, don’t animate it. Make a new one with an adult age in the prompt instead.
3. The file, for HappyHorse 1.1. A WEBP, PNG or JPEG of 10 MB or smaller, 300 px or larger, and no more extreme in shape than 1:2.5 tall or 2.5:1 wide.
4. The final shape. HappyHorse copies the shape of the still, so crop to 9:16 for phones or 16:9 for screens before uploading. Wan 3.0 can copy the still too, or you choose a fixed ratio.
5. Space to move. If the camera should travel, leave room around the subject; a very tight crop gives the model nothing to reveal.

You’ll find both in Video Lab, and either can start from your picture. These differences decide which one a clip needs:

Rule of thumb: if the clip needs audio, a planned ending or more than 15 seconds, it has to be Wan 3.0. One picture you love that only needs to come alive for a few seconds is HappyHorse territory, since it tends to stay closest to the still.

Say you already have a 2K portrait from Spicy UNCENSORED (10 credits) and want a short clip of it. Two ways to get there:

• With sound, on Wan 3.0: 2 test clips of 4 seconds at 480p while you tune the motion (52 credits), then an 8-second final at 720p (104). Total: 156 credits, about €2.60 at the Basic plan’s rate.
• Silent, on HappyHorse 1.1: one 3-second test at 720p (42 credits), then a 5-second final at 1080p (95). Total: 137 credits, about €2.28 on Basic.

Why test at 4 seconds and not 2? Wan 3.0 charges at least 26 credits per clip, and at 480p that minimum already pays for 4 seconds, so the longer test costs nothing extra and shows more of the movement.

Neither route fits inside the free credits. The Spark pack covers the tests on either route; pricing lists every pack and plan.

The model can already see your image. What it needs from you is what happens next, in three beats: where the movement starts, what changes, and where it ends. Add one camera move and, on Wan 3.0, what should be heard.

On Wan 3.0 every word counts: the prompt goes in exactly as typed, with no expansion. HappyHorse doesn’t require a prompt at all, so begin with only the motion and the camera and add detail when a clip wanders off.

• Wan 3.0, with sound: “The woman in her late thirties lifts her eyes from the wine glass, holds the look, then lets a slow smile spread; slow dolly-in; the low murmur of a hotel bar and a piano somewhere behind her.”
• HappyHorse 1.1, silent: “The man in his forties exhales and sinks back into the leather armchair; firelight flickers across his face; the camera drifts slightly to the right.”
• A loop on Wan 3.0: put the same picture in the first and the last frame, then prompt a movement that returns: “The woman, in her thirties, turns toward the rain-streaked window and back again as the curtain sways; static camera; soft rain.”

• The face drifts away from the picture. The prompt probably describes the person again, so the model redraws them. Describe only the movement; on HappyHorse, try a run with no prompt at all.
• Two actions, both clumsy. Keep one main action per clip and split a sequence into several clips.
• It ends in the wrong place. Give Wan 3.0 a last frame as well as the first.
• Odd cropping. HappyHorse follows the shape of the still, so fix the crop before uploading; on Wan 3.0, check the aspect setting.
• Silence. HappyHorse never makes sound. Wan 3.0 does unless audio is switched off, and it helps to describe what should be heard.
• A series of clips loses the character. Start each new clip from the original still, not from the last frame of the clip before.