Can an AI Model Stay Consistent Across 20 Lookbook Images? A Repeatable Test

Short answer: Yes, with a reference set and fixed light, most images stay recognisably the same person, but some will drift, especially profiles, extreme angles and close-ups. Score every image and replace the ones that fail rather than accepting drift.

Who this is for: Brands building an AI lookbook or ambassador.

Setup

One reference set (front, three-quarter, profile, full body). One set and light. Twenty prompts that change only pose and garment.

Scoring

Rate each image 1–3 on face, hair, skin tone and body. Anything with a 1 is regenerated.

Where drift usually happens

Shot	Risk	Fix
Front	Low	—
Three-quarter	Low–medium	Use three-quarter reference
Profile	High	Use profile reference
Close-up	Medium	Add skin and feature details
Wide shot	Medium	Face is small; check carefully

Best for / not ideal for

	Use it for	Not ideal for
Scoring sheet	Catching drift before publishing	Single images

Copyable prompt

Replace the words in [brackets].

Same person as references [front, three-quarter, profile]. [Pose], wearing [garment].
Set: [fixed set]. Light: [fixed light]. 50mm lens.

Where to go next

• Read the AI lookbook guide hub.
• Keep one model across every look with the free Lookbook Consistency Helper.
• Pay per image or clip, no subscription, in the studios.
• Learn the full method in the consistency workshop.

Questions

Why does the face drift? The model regenerates the face each time; references reduce but do not remove drift.

How many references do I need? Four views cover most shots.

Can video stay consistent? Short clips from a consistent still usually hold well.