⚠️ Already have an image? These prompts describe only what moves — not the scene. If you're starting from a written scene with no source image, use our 20 NSFW AI video prompts instead.
An image-to-video prompt has a different job from a text-to-video prompt. The model can already see your character, her outfit and the room, so the prompt only has to say three things: what moves, how the camera behaves, and what must stay the same. Prompts that re-describe the picture waste the words that should control motion — and they are the most common reason a clip warps.
Below are 20 copy-paste prompts in five groups, ordered from safest to riskiest, plus a settings cheat sheet and a fix-by-symptom list.
🎞️ Need a tool that accepts a source image? Compare every option — prices, clip length and free tiers — on the NSFW AI video generator comparison, or start with Joi AI ($2.38/mo) or CandyAI ($3.99/mo), which animates your companion right inside the chat.
How to Write an Image-to-Video Prompt
Every prompt below follows one formula:
[what moves] + [how it moves] + [camera] + [what stays unchanged] + [speed / length]
Say "the woman in the image." Refer to your subject as already visible. Don't describe her hair, outfit or room again — the model can see them.
One action per clip. A single slow, sustained movement is the most reliable output. "She turns, then stretches, then walks" produces cuts and melted limbs.
Name what must stay fixed. "Face, hair and outfit unchanged" is the single most useful line in this guide. It tells the model to hold identity instead of improvising.
Use slow words. "Slowly," "gently" and "subtle" pull motion down. Most bad clips come from too much movement, not too little.
Keep it to 15–30 words. Shorter than a scene prompt, because the picture already carries the detail.
For the settings side (motion strength, source-image framing, clip length) see the full NSFW image-to-video guide. One rule from it applies here too: only animate AI characters you generated yourself or images you own outright — never a real person's photo.
The 20 Prompts
Each prompt is written to paste straight into the prompt field next to your uploaded image. Keep to a single subject; multi-person clips are the weakest output on every tool.
Group 1: Subtle Life (Lowest Risk)
Start here to test whether your tool holds a face. Best run on OurDream AI — the face you build carries into the clip ($9.99/mo).
1. Slow breathing
The woman in the image breathes slowly and naturally, shoulders rising gently, soft blink, subtle eye movement. Camera static. Face, hair and outfit unchanged. 4 seconds.2. Glance to camera
The woman in the image slowly turns her gaze to the camera and gives a faint, relaxed smile. Only her eyes and lips move. Camera static, lighting and framing unchanged. 5 seconds.3. Playful look-up
The woman in the image lowers her eyes, then looks back up with a slow, playful smile. Minimal head movement, camera static, same lighting and framing. 5 seconds.4. Hair tuck
The woman in the image tucks a strand of hair behind her ear with one hand, then lets the hand rest. Slow, small movement. Camera static, background unchanged. 4 seconds.Group 2: Hair, Fabric and Wind
Best run on Secrets AI — skin, fabric and lighting hold up in motion, not just in a still ($19.99/mo, currently 50% off).
5. Breeze
A gentle breeze moves the woman's hair and the loose fabric of her clothing. She stays in position, looking off-frame. Camera static. Face and pose unchanged. 5 seconds.6. Robe strap
The woman in the image slowly lets the strap of her robe slide off one shoulder, the fabric falling naturally. One continuous motion. Camera static, face and background unchanged. 5 seconds.7. Sheer curtain
A sheer curtain behind the woman drifts in a light breeze while she stays nearly still, window light flickering slightly. Camera static. 6 seconds.8. Steam and droplets
Steam rises slowly around the woman in the image and light droplets run down her shoulder. She holds her pose. Camera static, face and pose unchanged. 5 seconds.Group 3: Slow Body Movement (Medium Risk)
Best run on CandyAI — it animates your companion mid-chat, with no export-and-reupload step ($3.99/mo).
9. Look over the shoulder
The woman in the image slowly turns her head and shoulders to look back over her shoulder at the camera. One smooth turn. Camera static, outfit and lighting unchanged. 5 seconds.10. Stretch
The woman in the image stretches her arms overhead and arches slightly, then relaxes. Slow, natural movement. Camera static, same room and lighting. 6 seconds.11. Lean toward camera
The woman in the image leans slowly toward the camera, eyes on the lens, hair falling forward slightly. Camera static, face and outfit unchanged. 5 seconds.12. Two steps forward
The woman in the image takes two slow steps toward the camera and stops, weight shifting naturally. Camera fixed at eye level. Same location and lighting. 5 seconds.Group 4: Camera Moves (Subject Nearly Still)
Best run on Joi AI — the cheapest way to test camera prompts before paying more ($2.38/mo on the annual plan).
13. Slow push-in
Slow, steady push-in on the woman in the image from medium shot to close-up. She stays nearly still, breathing subtly. Face, outfit and lighting unchanged. 5 seconds.14. Gentle orbit
The camera arcs slowly around the woman in the image by about 20 degrees, keeping her centered. She holds her pose. Consistent lighting and face. 5 seconds.15. Tilt up
The camera tilts slowly upward from her hands to her face, ending on her eyes. Subject barely moves. Face, outfit and background unchanged. 5 seconds.16. Handheld feel
Subtle handheld camera movement, as if filmed close up on a phone, small natural sway. She looks at the lens and smiles slightly. Same framing and lighting. 4 seconds.Group 5: Edit and Retouch Prompts
These change something about the image while it animates. Best run on GeneratePorn AI — built around generate, edit, then animate in one workspace ($14.99/mo, 3 free images).
17. Relight
Keep the woman, her pose and the room exactly the same. Gradually shift the lighting from cool daylight to warm golden-hour light. She breathes subtly. Camera static. 5 seconds.18. Background swap
Keep the woman and her pose unchanged. Replace the background with a softly blurred candlelit bedroom. Slow, subtle breathing only. Camera static. 5 seconds.19. Outfit colour change
Keep the woman, her face and pose unchanged. Change the colour of her dress from white to deep red, keeping fabric folds and shading natural. Minimal motion, camera static. 4 seconds.20. Clean-up pass
Keep the same woman, pose and framing. Reduce flicker and warped edges, keep natural skin texture without over-smoothing, stabilise loose hair strands. Very little motion, camera static. 4 seconds.Group 5 only works on tools that accept an edit instruction alongside the image. If yours doesn't, make the edit first as a still — our NSFW image-to-image guide covers how — then animate the finished image with a prompt from Groups 1–4.
Settings Cheat Sheet
Slider names differ between tools, so read "motion" here as whatever your generator calls its movement or intensity control.
| Group | Motion setting | Clip length | Main risk |
| 1. Subtle life | Low | 4–5 s | Frozen clip — nudge motion up slightly |
| 2. Hair, fabric, wind | Low to medium | 5–6 s | Fabric morphs into skin or background |
| 3. Slow body movement | Medium | 5–6 s | Face drift, duplicated limbs |
| 4. Camera moves | Low for the subject | 4–5 s | Background warping on wide arcs |
| 5. Edit and retouch | Low | 4–5 s | The edit bleeding into the face |
How to Edit or Retouch an AI Video With a Prompt
There are three routes, and which one you can use depends on the tool.
Edit during generation. Some generators accept an instruction alongside the source image (relight, change the background, change a colour). Use the Group 5 prompts and start with one change per clip.
Edit first, animate second. The most reliable route on any tool: fix the still with an image-to-image edit, then run a Group 1–4 prompt on the result. You control the edit before motion can distort it.
Regenerate rather than repair. Most tools can't patch a finished clip. If the face drifted or a limb melted, adjust the prompt (lower motion, add "face unchanged") and generate again instead of trying to retouch the output.
Support for editing an uploaded video clip with a prompt is much thinner than support for animating a still. The video generator comparison lists what each platform actually accepts.
What to Leave Out
Don't re-describe the image. "Beautiful woman with long brown hair in a red dress in a bedroom" is already in the picture. Spend the words on motion.
Don't stack actions. One movement per clip; stitch clips together afterwards.
Don't ask for fast or large motion. Spins, dancing and running are where faces and limbs fall apart.
Don't add a second person. Identity across two moving figures is the weakest output on every current tool.
Don't ask for sound or dialogue. Most NSFW video tools don't generate synchronised audio, and the words only confuse the motion model.
Fixing Common Problems
The face changes partway through. Add "face unchanged", lower the motion setting and shorten the clip.
The clip looks frozen. Add a clear verb ("slowly turns", "gently sways") and raise motion one notch.
Clothing melts or changes colour. Cut fabric-movement language, or use a wind prompt from Group 2 instead of a body-movement prompt.
The background warps on a camera move. Reduce the arc (10–20 degrees is plenty) or switch to a static camera with a subtle subject movement.
An edit spreads to the whole image. Name the region ("only the background") and repeat what must stay unchanged.
Frequently Asked Questions
What's the difference between an image-to-video prompt and a video prompt?
With a source image, the prompt only needs to describe motion, camera and what stays fixed. A text-to-video prompt has to invent the whole scene, so it also describes the subject, setting and lighting. Scene prompts live in our 20 NSFW AI video prompts.
How do I edit or retouch a video with a prompt?
Either give the tool an edit instruction alongside your image (Group 5), or fix the still first and animate it afterwards. See the edit and retouch section above for the three routes.
How long should an image-to-video prompt be?
About 15–30 words. Scene prompts run 20–40 because they carry more description.
Can I reuse the prompts from the video prompts guide?
Partly. Delete the parts that describe the subject and setting, keep the action, camera and lighting words, and add "face and outfit unchanged."
Which tools accept a source image?
OurDream AI, Secrets AI, GeneratePorn AI, CandyAI and Joi AI all do. Our image-to-video guide compares how each handles it, and the video generator comparison covers prices and free tiers.
Can I animate a photo of a real person?
No. That's non-consensual intimate imagery once the output is sexual, and it's illegal under the US TAKE IT DOWN Act and equivalent UK and EU law. Animate AI characters you generated yourself.
Why does the face change during the clip?
Usually too much motion for too little source detail. Lower the motion setting, use a clear front-facing source, and add "face unchanged."