NSFW AI video from an image: animating an adult still

Making an adult clip from scratch with text-to-video is a gamble: the model has to invent the person, the room, the light and the motion in one go, and the provider's moderation judges all of it at once. Animating a still you already like is the calmer route. The look is settled, and the model only has to decide how the picture moves. These are our working notes on doing that with adult images on Genlaxy, including the parts that do not work.

Start with a still that already passed

Make the image first, on a model that you know accepts adult prompts (our guide to NSFW image models has a cheap way to find one). Then press Animate on it in your feed or library. The studio switches to the Video tab with that image attached as the start frame, and you pick a model that animates images.

A frame that animates well has the same qualities as any good start frame, and our general image-to-video walkthrough covers them: room in the direction of motion, one clear subject, the aspect ratio you want the clip to have. For adult frames, add two more:

• One person, or two in a simple pose. Two bodies touching and moving are the hardest thing for any video model. Limbs merge, hands vanish. A single figure, or a couple in a still embrace, holds together far better.
• Nothing that makes age ambiguous. The frame is judged by the provider too, not only your words. A clearly adult subject in the still is the best insurance that the clip is not refused.

Expect more refusals than with stills

Be ready for this: video moderation at the providers is stricter than image moderation, and it looks at the start frame and the motion prompt together. We see frames that rendered without complaint as stills get refused as clips. That is the provider's verdict, not ours, and it is not a sign you did something wrong.

When it happens, the render ends as failed and the price returns to your balance automatically. Then either try another model that animates images, or make the still a little less explicit (implied nudity instead of full) and animate that. The clip often ends up more sensual for it, not less.

Write the motion, not the body

The model can see the frame. Describing the anatomy again spends words on nothing and gives the moderation more to object to. Spend the prompt on what moves, how slowly, and what the camera does.

Slow words work better than fast ones in adult clips, for a plain technical reason: fast motion is where video models warp skin, hands and faces. "Slowly", "gently", "barely" and "slow motion" are not only a mood choice. Our text-to-video prompt guide has more camera terms that read clearly.

Length, resolution and the price

Clips run up to 30 seconds, depending on the model, and the price scales with length and resolution; the Generate button shows the exact total before you press it. For adult clips there is an extra reason to start short and small: a refusal is refunded, but it still costs you the wait. Test at the lowest resolution and a 5-second length, and only render the final version once the model has accepted the frame and the motion looks right. The video model comparison helps with picking where to render the final.

What never works, on any model

• Photos of real people. You may upload a start frame only if it shows you or you have the person's permission, and even then the terms forbid sexual or intimate imagery of a real, identifiable person. Animate a generated character, not a photo of someone.
• Anything involving minors, age-play or non-consent. Our screen refuses these before anything is charged, on the motion prompt as on any other.
• Rewording to get past a refusal. If a clip is refused for what it shows, change what it shows or change the model. Rephrasing the same thing to slip past moderation breaks the terms.