AI Video Generation

Ch10 AI Video Generation: Kling/Jimeng/Pika/Runway Compared

AI images are static — short dramas need motion. AI video generation went through a qualitative leap in 2024, moving from distorted artifacts to smooth, photorealistic 5-10 second clips, fundamentally reshaping short drama production economics. This chapter compares the four leading tools in detail, teaches motion control, and explains the editing logic for assembling scattered AI clips into coherent episodes.

Four-Tool Detailed Comparison

Image-to-Video vs Text-to-Video

Image-to-Video (I2V) — Primary Mode for Short Drama

Start with a static image as the first frame; AI generates motion from that image. This is the primary workflow for short drama production because it locks character appearance from the source image, works seamlessly with LoRA-generated character images, and delivers predictable, repeatable results for batch production.

Text-to-Video (T2V) — For Non-Character Shots

Use T2V for shots that don't need your specific character: aerial city views, nature scenes, abstract transitions, background atmosphere shots, and title sequences.

[WARNING] Never use T2V for character shots: Text-to-video cannot reproduce your specific character's appearance. Any shot requiring your defined cast must use image-to-video mode.

Motion Control: Avoiding AI Distortion

The key strategies to prevent "melting faces" and limb distortion:

• Limit motion amplitude: Describe small, localized movements rather than full-body action

• Prefer static camera shots: Fixed camera + subtle character movement is more stable than moving camera + still character

• Use 5-second clips: Shorter clips are more stable; generate multiple 5s clips rather than forcing one 10s clip

• Motion intensity at 40-60%: Higher settings cause more distortion

[Kling — Video Prompt Examples]

-- Close-up, subtle motion --
"A beautiful East Asian woman in a white suit near a window,
slight smile forming, gentle breeze moves her hair,
camera slowly pushes in on face, cinematic, warm lighting"

-- Emotional reaction shot --
"Close-up of woman's face, tears beginning to fall,
slight lip trembling, static camera, soft natural light"

-- Avoid these (cause distortion) --
"running, jumping, turning quickly, fighting, dancing"

Editing Logic: Building Episodes from 5-Second Clips

[Scene Shot Structure Template]

Scene: Office confrontation (CEO fires assistant)

Shot 1 (3s): Wide shot — office environment established
Shot 2 (4s): Medium — CEO walks toward assistant, cold expression
Shot 3 (2s): Close-up — CEO's eyes, contemptuous downward look
Shot 4 (3s): Reaction medium — assistant rises, trembling
Shot 5 (2s): Close-up — assistant's eyes reddening, biting lip
Shot 6 (4s): Wide — CEO turns to leave, assistant watches
Shot 7 (3s): Close-up — assistant clenches fist, resolve forming

Total: ~21 seconds, 7 shots
Rhythm: slow→slow→fast→slow→fast→medium→medium

Cost Reduction Strategies

[Tiered Tool Strategy]

A-tier shots (lead close-ups, emotional peaks):
  Kling Pro or Runway Gen-3
  ~$0.30-0.50/clip × 25 shots = ~$10-12

B-tier shots (dialogue, medium shots):
  Kling Standard or Jimeng
  ~$0.10-0.15/clip × 35 shots = ~$4-5

C-tier shots (backgrounds, transitions):
  Jimeng free quota or T2V
  ~$0 × 15 shots = $0

Total: ~$14-17/episode (vs. thousands for live production)

[TIP] Chapter Action Checklist:

1. Register for Kling and generate a 5-second test video from your Ch08 character image;
2. Compare outputs at motion intensity 30/50/70 to calibrate the setting;
3. Plan 7 shots for one scene using the shot structure template;
4. Generate those 7 clips and practice assembling them in CapCut/Premiere;
5. Track your per-episode generation cost to build a budget baseline.

← PreviousCh09 LoRA Character Consistency Next →Ch11 Storyboarding