AI Kiss Video: 7 Prompt Templates That Actually Work (2026)
The AI kiss effect took over social feeds — here are 7 copy-paste prompt templates for realistic, tender couple animations, plus the photo rules that decide whether your video succeeds.
2026-09-28 · Tutorials · 7 min read
AI Kiss Video: 7 Prompt Templates That Actually Work
The AI kiss effect — two people in a still photo leaning in for a gentle, natural kiss — became 2026's most-shared video format. Here's how to get results that look like a film moment instead of a glitch.
Before the templates, the truth about this effect: 80% of output quality is decided by the input photo, not the prompt. The model reconstructs 3D motion from a single frame, so it needs clear faces to work with. Fix the photo first, then choose a template.
The photo checklist (read this first)
- Both faces fully visible, frontal or three-quarter — no profile-only shots.
- Even lighting on both faces; mixed indoor/outdoor lighting confuses motion.
- Faces at similar distances from camera; a huge size difference breaks the lean-in.
- No sunglasses, masks, hands or hair covering mouth area.
- Higher resolution source = better identity preservation. A 150px avatar will not survive.
7 prompt templates
1. The classic gentle kiss (the default that works): “The two people in the photo look into each other's eyes, lean in slowly and share a brief gentle kiss, then part slightly with a soft smile. Faces stay exactly the same as in the photo, clothing and background unchanged, one continuous shot.” This is essentially the tuned prompt behind Slloy's AI Kiss effect — plain, specific about motion, and explicit about what must not change.
2. The forehead kiss (wholesome mode): “The taller person gently kisses the other person's forehead; the other person closes their eyes and smiles. Slow, warm, tender motion, one continuous shot, background unchanged.” Far more reliable than lip-sync-adjacent motion and perfect for family photos.
3. The almost-kiss (romantic suspense): “The two people lean toward each other as if about to kiss, pause with faces an inch apart, then both smile. Gentle suspenseful motion, no actual lip contact.” Avoids the hardest motion class entirely and looks intentional and cinematic.
4. The kiss on the cheek: “The person on the left turns and kisses the other person's cheek; the other person laughs softly. Natural speed, one continuous shot, faces consistent with the photo.” Works even with slight face-angle differences.
5. The wedding photo revival: “The couple in wedding attire lean in for a soft kiss; the veil moves slightly in a gentle breeze. Slow cinematic motion, warm lighting preserved from the photo.” Adding one environmental detail (veil, hair, fabric) gives the model a safe secondary motion channel.
6. The family edition (parent and child): “The parent kisses the child's cheek; the child giggles and hugs back. Warm, playful, natural motion, background unchanged.” Identity preservation on children's faces is excellent when the source photo is sharp.
7. The reunion hug-kiss combo: “The two people step toward each other, share a warm hug and a brief kiss on the cheek, then hold the embrace, smiling. One continuous tender shot.” Slightly more ambitious motion — use your sharpest photo for this one.
If you're generating from scratch instead of using the effect
Slloy's AI Kiss effect packs all of this into one click for 20 credits. If you want custom control — camera moves, different pacing, longer duration — take a template above into the video generator with your photo as the first frame: Seedance 2.0 Mini does 5 seconds at 720p for 100 credits, and you only pay for successful renders (failed runs are auto-refunded).
Consent, always
Animate yourself, your partner, your family — with their knowledge. Animating photos of strangers or public figures without consent isn't just against every platform's terms; in many jurisdictions it touches personality and image rights. The sweet spot of this effect is genuinely personal moments, and there's plenty of room there.
Troubleshooting table: what went wrong → what to change
Keep this next to your prompt library. Almost every bad kiss-video result traces to one of six causes:
| Symptom | Likely cause | Fix |
|---|---|---|
| Faces morph or swap identity | Source faces too small or blurry | Crop tighter to the subjects; use the sharpest photo you have; avoid heavy re-crops of group shots |
| One person animates, other is frozen | Uneven lighting or one face in shadow | Even out light on both faces in a photo editor before animating |
| Motion looks robotic / sped up | Prompt asks for too much action in 5 seconds | One action per clip. Cut the prompt in half, not the motion |
| Background warps during lean-in | Busy background with fine patterns | Simplify background with a light edit pass, or add “background stays completely unchanged, static camera” |
| Mouth area glitches | Hands, hair or objects covering the mouth zone | Choose a photo where both mouths are clearly visible |
| Result feels uncanny, not romantic | Camera movement requested | Drop all camera language for this effect — static camera reads most natural |
And the meta-fix: if two regenerations fail the same way, stop prompt-tweaking and change the photo. The model is telling you the source can't support the motion.
Same photo rules, whole effect family
Everything above transfers directly to the neighboring effects, which is why building the habit once pays off across the shelf: the AI Hug effect uses the same lean-in mechanics with more tolerant geometry (hugs are easier than kisses — faces don't need to align as precisely); Muscle Transform cares mostly about a clear single-subject silhouette; and Earth Zoom Out rewards photos with some visible environment around the subject, since the camera pulls back through the scene. All five-second, all 720p, all 20 credits on Slloy — cheap enough to test each with a different family photo in one sitting using your signup credits.
Posting it well
If the clip is destined for social: post the before/after as a two-frame story (still photo, then the video) — the contrast is what makes people stop scrolling. Keep the original photo in the first frame rather than a zoom-in; audiences punish bait-and-switch. And caption it with the moment, not the tool: “Grandma and grandpa, 1974” outperforms “AI kiss effect!!!” by every engagement metric that matters. The technology is the easy part now — the sentiment is the content.