Skip to main content

Guide

How to keep the same character across AI video shots

Published · By the Draftless team · 2 min read

Build a character sheet first: front, three-quarter, profile views of the same person, plus the outfit. Write one fixed wardrobe line and reuse it word for word in every prompt. Generate each shot from those references or from a start frame, and use the last frame of one shot as the first frame of the next.

Why do characters change between shots?

Each generation starts fresh. With a prompt alone, the model invents a new face, haircut and jacket every time. With one front-facing photo, it guesses how the person looks from the side, and the guess drifts as the head turns. Consistency comes from giving every shot the same visual evidence.

Step by step: how do I lock a character?

  1. Make a character sheet. Save the person in Characters with front, three-quarter, profile views. Draftless counts those three as core angles and recommends up to 5 photos; it can generate the missing angles from your first photo.
  2. Write a locked wardrobe line. One sentence that never changes, for example “olive field jacket, white T-shirt, dark jeans, silver ring on the left hand”. Paste it into every prompt unchanged.
  3. Generate stills first. Make the key frame of each shot as an image with the character attached. A Nano Banana Pro still starts at 45 credits, against 240 for a 5-second Kling 4.0 clip, so fix the face and outfit here.
  4. Animate from the still. Use image-to-video so the clip starts on the frame you approved, or reference-to-video (Seedance 2.5) so the model sees several angles at once.
  5. Join shots with frames. Export the last frame of shot one and use it as the start frame of shot two. With Kling 4.0 or Seedance 2.5 you can also set an end frame, so a shot lands exactly where the next begins.
  6. Steer the middle when needed. Kling 4.0 image-to-video accepts up to 8 ordered middle keyframes through the API and MCP server, to hold a pose or outfit through a longer take.

Which settings matter most?

Consistency techniques and where they live in Draftless
TechniqueWhat it fixesWhere in Draftless
Character sheet (3 core angles)Face drift when the head turnsCharacters
Locked wardrobe lineClothes and accessories changingYour prompt, reused word for word
Stills firstPaying for video before the look is rightImage models in Studio
Start frameThe opening of the shotImage-to-video (Kling 4.0, Seedance 2.5, Veo 3.1)
End frameCuts that do not matchKling 4.0 and Seedance 2.5 image-to-video
Reference imagesIdentity across a whole shotSeedance 2.5 reference-to-video
Middle keyframes (up to 8)Drift in the middle of a long takeKling 4.0, API and MCP

What should I avoid?

  • Rewriting the description of the character in each prompt. Small wording changes produce small face changes.
  • Changing lighting, lens and outfit in the same shot. Change one thing per shot.
  • Long takes with fast head turns on a single front photo. Add the profile and three-quarter views first.
  • Re-rolling the same prompt hoping for a match. Fix the references instead; see why credits run out.

Frequently asked questions

How many reference photos do I need for a consistent character?
Three core angles (front, three-quarter, profile) cover most shots. Draftless recommends up to 5 photos per character, adding a full-body and a detail shot.
How do I make two AI video clips join smoothly?
Use the last frame of the first clip as the start frame of the second. Kling 4.0 and Seedance 2.5 also take an end frame, so you can make a clip land on a chosen image.
Do start and end frames cost extra?
Not on Kling 4.0: start, end and middle frames use the same per-second rate as image-to-video.

Every Draftless generation shows its exact credit price before it runs.