realistic-ugc-video（逼真的 UGC 视频）

Realistic UGC Video Production

Create long-form AI videos that look and sound authentically human. This skill orchestrates a multi-step workflow using Nano Banana for realistic base images and Kling AI for video generation, with specific techniques to avoid the "AI look".

Why Videos Look AI (And How to Fix It)

Known Limitation: Hands

AI video models (Kling, Veo, etc.) struggle with realistic hand movement. Fingers morph, gestures look unnatural, and hands are often the biggest tell.

Best practice: Keep hands OUT of frame or completely static.

Options:

1. Head/shoulders framing - Crop base image to exclude hands entirely
2. Arms crossed - Static pose, no finger movement needed
3. Hands below frame - Desk edge cuts off at wrists
4. Cover with B-roll - Cut away during any hand weirdness in post

Complete Workflow

Phase 1: Collect Requirements

Before starting, gather from the user:

1. Character description - Age, ethnicity, features, clothing
2. Setting/background - Office, home, studio, outdoor
3. Script - The full text the character will speak
4. Tone - Conversational, urgent, professional, friendly
5. Video length target - This determines how many clips needed

Phase 2: Generate Base Image (Nano Banana)

Use the Nano Banana skill to generate the character image. Critical: Apply the imperfection techniques from CHARACTER-PROMPTING.md.

Key elements for realistic UGC:

• iPhone capture aesthetic (26mm equivalent lens)
• Available/mixed lighting (NOT studio)
• Visible skin texture (pores, oils, fine lines)
• Minor imperfections (stubble, dark circles)
• Computational depth artifacts
• ISO noise (500-900 range)

Command:

~/.claude/skills/nano-banana/scripts/generate.sh "[enhanced prompt]" --aspect 9:16 --size 2K

Then optionally upscale through Enhancor AI for additional texture.

Phase 3: Chunk the Script (Critical for Pacing)

This is the most important step for natural pacing. See SCRIPT-CHUNKING.md.

The 55-60 Syllable Rule:

• Count syllables, not words
• Each video generation = 55-60 syllables
• Never cut mid-sentence
• Add filler sentences to reach target if needed

Example chunking:

Chunk 1 (58 syllables):
"Hey everyone, I wanted to share something that completely changed how I think about productivity. It's not another app or system."

Chunk 2 (56 syllables):
"It's actually about understanding your own energy patterns throughout the day. Once I figured this out, everything clicked into place."

Phase 4: Generate Video Clips (Kling AI)

For each script chunk, generate a 10-second video clip. Use the Kling AI skill with movement prompts from MOVEMENT-PROMPTING.md.

Key elements:

• Hand "Home Base" Protocol
• Timestamped movement clusters
• Natural blinks, head tilts, micro-sways
• Consistent base image reference

Spawn background agent for each clip:

Task tool:
- subagent_type: "general-purpose"
- run_in_background: true
- prompt: [Include image URL, script chunk, movement prompt, output path]

Phase 5: Post-Production

See POST-PRODUCTION.md for detailed guidance.

1. Assemble clips in CapCut or similar editor
2. Fix audio with Adobe Podcast (minimum) or Resemble AI (for voice swap)
3. Cover jump cuts with B-roll or animations
4. Remove filler sentences if they feel awkward
5. Export in final resolution

Quick Start: Single Clip Test

Before generating full video, test the workflow with one clip:

1. Generate base image with full imperfections
2. Create 10-second test video with first script chunk
3. Evaluate pacing and movement
4. Adjust prompts if needed
5. Proceed with full production

Example Full Prompt for Base Image

A vertical 9:16 UGC-style video frame captured on an iPhone 11 resting on a tripod.
Medium-wide portrait at true eye-level with slightly forward-leaning posture.

[CHARACTER]: A [age] [gender] with [ethnicity] complexion, [eye color] eyes beneath
[eyebrow description], [nose description]. [Jawline/facial hair]. [Hair description].
Expression is [emotion]—[specific expression details].

[CLOTHING]: [Fitted/casual garment], [collar detail].

[SKIN TEXTURE]: Visible pores across T-zone, faint smile lines, natural oils catching
light on forehead and nose. [Age-appropriate details]. No filter, no foundation.

[FOREGROUND]: Hands rest naturally on [surface], fingers relaxed, visible veins and
knuckle texture. Nearby: [everyday objects like water bottle, phone, notebook].

[CAMERA]: Native iPhone 11 lens (26mm equivalent), slightly wide perspective, mild
barrel softness at edges. Only tiny pockets of neural blur around hair edges.

[LIGHTING]: Available light mix—cool overcast daylight from window left, warm tungsten
from desk lamp right. Soft asymmetric shadows, natural falloff. ISO noise 500-900.

[BACKGROUND]: [Realistic home/office elements]—bookshelf, [furniture], clearly visible
not heavily blurred.

[REALITY DETAILS]: Gentle 35mm film grain, light fingerprint smudge on lens, tiny dust
haze in air. No cinematic bloom, no studio finish.

Styling: raw UGC realism, available indoor light, mixed color temperature, minimal
depth blur, visible ISO noise, emphasis on authenticity.

Example Movement Prompt for Video

Hand "Home Base" Protocol: Hands default to Active Idle. Fingers shift, thumbs rub,
wrists rotate slightly while anchored. Gestures only for key emphasis.

[0.0s-0.5s] Pre-roll: Sharp inhale, eyes lock to lens, head still
[0.5s-3.0s] Hands in Active Idle (fingers interlocked), head tilts slightly right,
           brows furrow in seriousness
[3.0s-6.0s] Hands break clasp for quick open-palm rotation then return, head drifts
           forward, natural blink
[6.0s-8.0s] Hands return to Active Idle (loose clasp, thumbs tapping), head nods
           encouragingly, cheeks lift in natural smile
[8.0s-10.0s] Hands anchored (wrist shifts), chin lifts in quick final nod, natural blink

[Script]: "[CHUNK TEXT HERE]"
[Tone]: [Urgent/Conversational/Professional/etc.]
[Pacing]: Rapid fire delivery, high energy, viral UGC style, confident, 2x speed

Reference Files

• CHARACTER-PROMPTING.md - Imperfection techniques for realistic characters
• SCRIPT-CHUNKING.md - The syllable method for consistent pacing
• MOVEMENT-PROMPTING.md - Natural movement choreography
• POST-PRODUCTION.md - Audio fixing and editing tips

Alternative: InfiniteTalk

For simpler long-form videos, consider InfiniteTalk (infinitetalk.ai).

Pros: Single generation for longer videos Cons: Less control over pacing (no timer/duration control), charged by output length

Use the syllable method above when precise pacing control is needed.

Checklist Before Generating

• Character prompt includes skin imperfections
• Lighting is available/natural, NOT studio
• Script chunked into 55-60 syllable segments
• Each chunk is complete sentences
• Movement prompt includes hand base, head movements, blinks
• Post-production plan for audio and jump cuts