I burned six weeks and a few hundred generations on one lesson. A video prompt is not a longer image prompt. It’s a leaner one plus motion. Every fat image prompt I dropped into a video model came back like a slideshow having a seizure. Faces sliding, a third hand out of a hip.
Dense prompts melt
An image model renders once, so it can afford 40 descriptors: skin texture, freckles, rim light, film grain, lens. A video model holds all of it stable across a hundred plus frames. Overload it and it negotiates with itself. Frame 12 favors the freckles, frame 40 the lens, the face drifts in between.
So I cut. Subject, one or two physical anchors, wardrobe, setting, light. Fifty words for the scene, tops. What I save there, I spend on motion.
Common fail
Pasting your image prompt into the video box, all 90 tokens, then wondering why she’s a different woman by second three.
Fix: strip it to the five things that must stay true, then add one action and one camera move. If a word doesn’t say who she is, where she is, or how she moves, cut it.
One action she can hold
This is the whole game. These models render one continuous, sustainable action beautifully. Ask for three and you get a stutter, because it starts all three at once.
Sustainable means it loops, no obvious start or end. Breathing. Rocking. Riding. Hair falling. Weight shifting. “She undresses, turns around, then lies back” is three scenes, and five seconds can’t serve three.
Example prompt
“Adult woman, late twenties, lying back on white sheets, topless, one hand in her hair. Slow push-in on her upper body. Chest rising and falling with slow breathing, hair falling forward across one shoulder. Warm window light, consistent lighting, smooth motion, 4 seconds.”
Animate it on PornWorksPaste the prompt as written. One motion, four seconds.→
The words that make it move
Motion cues are their own vocabulary. My whole list, basically.
| Layer | Words that work | Rule |
|---|---|---|
| Camera | slow push-in, slow pull-back, locked camera, slow pan left, handheld drift | One per clip. Never two. |
| Body | weight shifting hip to hip, arched back, shoulders rolling, hips rocking, thighs tensing | One action, held. It has to loop. |
| Small stuff | hair falling forward, chest rising and falling, fingers tightening, lips parting | One or two. This is what sells it. |
| Quality glue | gentle movement, smooth motion, consistent lighting, no cuts, single continuous shot | All of it, every prompt. |
“Gentle movement” and “smooth motion” look like filler. They are not. They’re the two phrases that most reliably talk a model out of jerky, sped up nonsense. “Consistent lighting” stops the room strobing. Fuller glossary for the scene half is in the NSFW prompt dictionary.
Example prompt
“Adult couple, both mid thirties, she straddles him on a low couch, both nude. Locked camera. One continuous motion: her weight shifting hip to hip, back arching, hair falling forward. Warm lamplight, consistent lighting, gentle movement, no cuts, 5 seconds.”
Drift and the four second rule
Identity drift is math, not bad luck. Every frame gets predicted from the one before it, so small errors compound. By second eight her nose changed and the tattoo migrated.
You can’t prompt that away. You outrun it. I shoot three to five second clips, then stitch. Six clean 4 second clips look like footage. One 20 second attempt is a fever dream. Reuse the subject description word for word, same seed if you have one. For generators that hold a face best, see image to video AI in 2026.
Common fail
Asking for ten seconds because the slider allows ten seconds.
Fix: half the length, twice the takes. Length is where hands go feral.
The negative block nobody writes
Your image negatives still apply. Video needs its own set on top. Mine never changes.
“stuttering, flickering, warping, morphing limbs, inconsistent face, changing hair color, jittery camera, sudden cuts, frame skipping, melting hands, duplicated body parts”
Paste it into every generation. Half the “this model sucks” complaints I read are a missing negative block.
Eight clips ready to run
Same skeleton every time. A lean scene, one action she can hold, one camera move, four or five seconds. Paste the negative block under each one, then swap the details and keep the bones.
Shower, solo
“Adult woman, late twenties, wet dark hair slicked back, nude under a rainfall shower. Slow pull-back from her face to her waist. Water running down her shoulders and chest, head tilting back, eyes closed. Steam, soft overhead light, consistent lighting, smooth motion, no cuts, 4 seconds.”
Cowgirl, side view
“Adult couple, both early thirties, she rides him on a leather sofa, both nude, side view. Locked camera. Her hips rocking forward and back, his hands on her waist, her hair swaying. Dim amber lamplight, consistent lighting, gentle movement, no cuts, 5 seconds.”
POV, kneeling
“POV looking down at an adult woman, mid thirties, kneeling on a white rug, topless, hands on her thighs. Slow push-in. She looks up into the lens, shoulders rising with each slow breath. Warm side light, consistent lighting, gentle movement, no cuts, 4 seconds.”
Close-up, just breathing
“Extreme close-up of an adult woman’s face, early thirties, flushed cheeks, lips parted. Locked camera. Slow heavy breathing, eyelids fluttering, a loose strand of hair moving with each breath. Warm candlelight, shallow depth of field, consistent lighting, smooth motion, no cuts, 4 seconds.”
Doggy, from behind
“Adult couple, both late twenties, doggy style on a bed, both nude, three-quarter view from behind. Locked camera. His hips moving in a slow steady rhythm, her back arched, hair falling over one shoulder. Low warm light, consistent lighting, gentle movement, no cuts, 5 seconds.”
Two women, slow kiss
“Two adult women, both early thirties, kneeling face to face on a bed, topless, kissing slowly. Slow push-in. Hands moving up each other’s backs, heads tilting into the kiss. Soft daylight through sheer curtains, consistent lighting, gentle movement, no cuts, 4 seconds.”
Anime, on the futon
“Anime style adult woman, long silver hair, open kimono, lying on her side on a futon. Locked camera. Chest rising and falling with slow breathing, hair sliding across the pillow. Flat cel shading, warm paper lantern light, consistent lighting, gentle movement, no cuts, 4 seconds.”
Satin, slow sway
“Adult woman, mid thirties, curvy, red satin slip dress, dim hotel room at night. Handheld drift, waist up. Her hips swaying slowly side to side, the satin catching the light. City lights through the window, consistent lighting, smooth motion, no cuts, 5 seconds.”
Start from a still you already like
Don’t invent the frame. Hand an image you love to the image to prompt page and it returns the prompt that would recreate it, in your browser, nothing uploaded. Trim to fifty words, then add one action, one camera move, the negative block. You’re animating a proven frame.
Example prompt
“Anime style adult woman, black bob, unbuttoned white shirt, kneeling on a bed. Slow push-in from the waist up. She leans forward, hair falling across one eye, shoulders rolling gently. Flat cel shading, consistent lighting, gentle movement, no cuts, 4 seconds.”
One note on tools. Kling and the mainstream majors do gorgeous motion and a flat no on explicit output, so you need a generator that allows adult content. Mine is PornWorks. It takes a text prompt or animates a still, so the lean scene and the motion line go in exactly as written. And if writing the scene is the slow part, pick VIDEO in my free prompt generator. It keeps the motion words an image prompt would drop.
Lean scene. One motion. Four seconds. That’s where it stops looking like AI and starts looking filmed.
