Best Photos for AI Kiss Videos: A Complete Input Guide (2026)

2026-05-01 · 6 min read

Photo input quality matters

Let's Mack's AI is good but it's not magic. The quality of your output is bounded by the quality of your input. This guide breaks down exactly what makes a photo a great AI kissing video input — lighting, resolution, composition, expression — and which mack styles forgive bad inputs vs which expose them.

The four input quality dimensions

Every photo can be evaluated on four axes. Optimize at least three for great output.

1. Lighting

The most important factor. Skin tones, edge details, and motion all degrade in bad lighting.

Best: Natural daylight (window light, golden hour, even outdoor light), soft indoor lamp light, diffused studio light, ring-light front-on selfies.

Acceptable: Bright fluorescent (works but slightly flat), midday sun (too harsh on its own but workable), candle light or warm-amber rooms (cinematic but can wash out faces).

Avoid: Harsh phone-flash (creates red-eye and flat shadows), backlit photos where face is silhouetted, mixed-color-temperature lighting (warm + cool clashing), under-exposed dark photos, blown-out over-exposed photos.

2. Resolution

Higher is better, but most modern phone cameras produce sufficient resolution.

Minimum acceptable: 1080×1080 pixels (most Instagram square posts).

Ideal: 1920×1080 or higher (most modern phone photos exceed this).

Avoid: Low-resolution screenshots (Twitter / X images, low-quality Instagram resaves), pixelated images, heavily-compressed JPEG with visible artifact blocks.

3. Composition

How the subject is framed in the photo matters.

Best: Front-facing or 3/4-angle portraits where the face is unobscured, both eyes visible, face takes up at least 30% of the frame.

Acceptable: Profile shots (work for Two-Photo Mode but limit single-photo options), group photos cropped to two-people (the AI handles cropping but cleaner inputs work better).

Avoid: Back-of-head shots (no face visible), faces obscured by hair / hats / sunglasses, faces partially cropped out of frame, faces too far away (less than 15% of frame).

4. Expression

The expression in your source photo influences the output mood.

Best: Neutral or slightly smiling expressions. The AI motion can flow into different expressions but a neutral baseline gives it the most flexibility.

Style-dependent:

• Awkward and NPC Energy actually work *better* with photos where the subjects have neutral or surprised expressions — the contrast amplifies the joke.
• Tickle Fight needs smile / laughing photos as input — the AI can't generate believable laughter motion from a frowning face.
• Plot Twist and Enemies to Lovers prefer intense / serious expressions — softness breaks the tension.

Avoid for most styles: Closed eyes, mid-blink photos, fully-laughing tongue-out photos for sincere romantic styles.

Single-photo mode vs Two-Photo Mode

Two different input requirements depending on which mode.

Single-photo mode

Upload one photo with two people side by side. Both faces need to satisfy the four quality dimensions independently.

Common single-photo sources that work well:

• Couple photos (engagement shoots, vacation candids, dinner-party photos)
• Friend photos (casual selfies, group photos cropped to two friends)
• Sibling photos
• Photos at events (weddings, birthdays, parties) cropped to two subjects

Avoid: Group photos with two subjects far apart in the frame, photos where one subject is significantly smaller than the other, photos where one subject is partially obscured.

Two-Photo Mode

Upload one photo per person. The AI composes them into a single scene before motion generation.

Best practice: Use two photos with similar lighting conditions when possible — if both are golden-hour photos, the AI's composition step is cleaner. If one is daylight and one is night, the AI handles it but output quality is slightly lower.

Two-Photo Mode forgives more: Because the AI is composing two separate photos anyway, it can handle larger differences between the two source images than single-photo mode can.

By-style input recommendations

Different mack styles have different input forgiveness levels.

High forgiveness (works on basically any input quality):

• NPC Energy — the glitchy aesthetic masks input issues.
• FBI Open Up — the chaos masks input issues.
• Glitch in the Matrix — the chromatic aberration masks input issues.
• Awkward — bad photos make it funnier.
• Photobomb — the third-subject implication distracts.
• Nature Doc — the absurdist framing makes any setting funny.

Medium forgiveness:

• Telenovela, Bollywood, Italian Chef — maximalist styling masks moderate input issues.
• Anime, Fairy Tale — particle effects mask moderate issues.
• Rain Kiss — the rain effect smooths inputs.

Low forgiveness (need clean photos):

• Classic — no effects to mask input issues.
• Movie Kiss — clean Hollywood styling exposes input quality.
• Sunset — warm-tone styling exposes blue-cast inputs.
• Slow Mo — slow motion exposes any softness.

Common input mistakes

Things people upload that produce bad output:

1. Heavy filters / Instagram-Lightroom-edited photos. The filters are baked into the pixel data; the AI struggles to generate motion that respects the filter look. Use unfiltered originals when possible.
2. Screenshots of photos. Always re-compress and lose quality. Use the original photo file.
3. Photos with hands in front of the face. Hand-on-chin / hand-near-face poses confuse face detection.
4. Photos with one closed eye / mid-blink. The AI generates motion but the closed-eye starting state propagates into the kiss.
5. Photos with strong color casts (heavy blue, heavy red). Most styles assume neutral lighting; strong casts clash.
6. Photos where someone is wearing a face mask / cloth covering. The AI can't render motion through fabric.

Quick optimization checklist

Before generating, ask:

• [ ] Is the lighting good (window / golden hour / soft indoor)?
• [ ] Is the resolution at least 1080px on the short side?
• [ ] Are both faces front-facing or 3/4-angle, fully visible?
• [ ] Does the expression match the style I'm picking?
• [ ] Is the photo unfiltered / un-edited (or at least lightly edited)?

If at least 4 of 5 check, you're in great shape.

TL;DR

• Lighting > resolution > composition > expression.
• Single-photo mode needs cleaner inputs than Two-Photo Mode.
• High-forgiveness styles (NPC Energy, FBI Open Up, Awkward, Glitch in the Matrix, Photobomb, Nature Doc) work on any input.
• Low-forgiveness styles (Classic, Movie Kiss, Sunset, Slow Mo) need clean photos.
• Most common mistakes: heavy filters, screenshots, hand-on-face poses, mid-blink photos.

Try Let's Mack free — 3 videos on signup, no credit card. Get started at letsmack.com.