Skip to content
IFAI

Generative media

Kling lets you train your own consistent AI video characters

Published Reach 47

Kuaishou's Kling AI just started testing a new Custom Models feature that tackles character consistency in video generation. Users can train personal characters using 10 to 30 short video clips and tag them directly in prompts to keep faces stable across angles. Why it matters: persistent character identity has long been the hardest bottleneck holding back AI filmmaking.

Kling lets you train your own consistent AI video characters
Image: x.com

Anyone who has tried making a coherent story with AI video knows the biggest headache: switch the camera angle, and your main character gets an instant face transplant.

Kuaishou's AI video generator KLING is now testing a feature called Custom Models. The whole point is to fix that nagging consistency problem by letting creators train and lock in specific characters across different scenes.

Training on video instead of still images

Most image and video tools have traditionally relied on static headshots or LoRA fine-tuning to keep a face looking familiar. KLING is taking a different route by training directly on moving video footage.

Early testers on X revealed that setting up a character requires uploading 10 to 30 video clips, with each clip lasting at least 10 seconds. Notably, you don't even need real-life camera footage. Training data can come directly from previous AI clips generated within KLING.

Once the custom character is trained, users just tag the character model right in the text prompt and describe the surrounding scene. Demonstrations shared by creator Halim Alrasihi show the character holding their likeness through full 360-degree camera rotations, sharp perspective changes, and complex facial shots.

KLING launched its international version in July, and its current 1.5 release generates 1080p HD video across standard formats like 16:9, 9:16, 4:3, and 1:1. It already offers a Motion Brush tool to manually control element motion, and gives new users 66 free daily credits upon email signup.

Why it matters

AI video has spent the last year showing off eye candy, but cinematic storytelling needs persistent actors. We think training on actual video snippets is much smarter than relying on a handful of flat images. Video gives the system native insight into how skin folds, shadows stretch, and features deform as a face turns.

For indie filmmakers, advertisers, and digital animators, this is a massive practical leap. If your protagonist looks the same from shot to shot, piecing together a real short film or commercial becomes a viable workflow rather than a lottery ticket.

The catch is the setup cost. Sourcing up to 30 ten-second video clips of a single subject is a noticeable barrier to entry. How KLING prices the compute for training these custom weights will ultimately decide whether it becomes an everyday creative staple or an expensive niche toy.

Sources (2)

  1. [1]
    @techhalla(opens in a new tab)Originaltwitter.com
  2. [2]

Topic trackerEvolution of Kling AI

Every story on this topic, newest first.

  1. Kling lets you train your own consistent AI video charactersThis story
  2. Kling AI upgrades to 1.5 with 1080p HD and motion brush
  3. Kling AI goes global with free video credits

Best of AI, daily