DEV Community

zz583
zz583

Posted on

Steal My Character-Lock Block — the Exact Template Behind Consistent AI Video Characters

#ai #promptengineering #video #tutorial

If your AI video character's face, hair or outfit quietly mutates between shots, the problem is almost never the model. It's that you re-describe the character from memory in every prompt — and every re-description drifts.

After shipping 30+ AI shorts (original-character PVs, a 35s photorealistic slice-of-life film, film-look replications), I stopped re-describing. I write the character once, as a fixed block, and paste it verbatim into every segment prompt. That's the whole trick — here it is, free.

The Character-Lock Block (copy this structure) 

[Lock block example] The same girl: long mint-green hair in loop-shaped side buns
with thin braids, amber-gold eyes with star-shaped highlights, white-and-navy
off-shoulder sailor top, blue-cyan chest bow with a golden star, star-pattern
pleated skirt, a large blue bow on her back, one long dark-blue tail tipped
with a blue heart.
Enter fullscreen mode Exit fullscreen mode

That block carried a character through a full 14-second PV, shot after shot, without her outfit swapping colors or her accessories vanishing. The rules that make it work:

  1. Cover everything, in a fixed order: hair color/style · eyes · top · bottom · signature accessories · special traits. Miss one item and that item is exactly the one that drifts.
  2. Paste it word-for-word. Never paraphrase it per segment. Consistency comes from identical tokens, not from describing the same thing "well" twice.
  3. Change only action + scene per segment. The block stays frozen; the segment prompt around it carries the movement, camera and lighting.
  4. Pair it with a reference image. Generate every segment independently from the same character sheet — in my tests, independent generation with a shared reference beats chaining from the previous segment's last frame, which drifts the worst.

The negative-constraint tail (attach to every segment) 

no subtitles, no text, no watermark, no logos, normal hands (five fingers),
no distorted face, no face change
Enter fullscreen mode Exit fullscreen mode

One full segment, assembled 

[lock block], standing at the center of a rain-slick neon street, breeze moving
her hair, slowly raising her head to camera, determined eyes, slow push-in,
cinematic light, clean frame, no subtitles, no text, no watermark,
no distorted face, no face change
Enter fullscreen mode Exit fullscreen mode

Where this comes from

This is Method 2 of my AI Short-Video Prompt Pack Vol.1 — the full system behind those 30+ shorts: segmented generation for 15–35s films, image-to-video style replication, the frame-check routine I run before delivery, and 36 fill-in templates (character PVs, duels, slice-of-life, old-photo revival, product shots and more), each annotated with the tool, aspect ratio and segmentation it was verified with. Bilingual EN / 中文 edition.

If the lock block saves one of your characters, I'd love to hear about it in the comments.

Top comments (0)