Keeping the same face across every frame

What it is good at

• Lets a small brand build a recurring 'face' for campaigns without repeated model bookings.
• Character and style reference features noticeably reduce drift compared with plain text prompts.
• Works well for a consistent illustrated or stylised mascot across many scenes.
• Useful for lookbooks and social sets that need one recognisable model across many outfits.

Where it falls short

• Perfect pixel-identical faces across dozens of frames are not realistic yet with any tool.
• Extreme angle or expression changes increase drift the most.
• Different tools handle reference images differently, so switching platforms mid-project resets consistency work.

How to use it for creative work

1. 01

   Choose or generate one hero reference image

   Pick a single, well-lit, front-facing image as the canonical reference for the character. Every future generation should point back to this one image rather than to the most recent output, which prevents slow drift across a long set.

2. 02

   Write a fixed identity description

   Write one sentence describing the character's fixed features, hair, face shape, distinguishing marks, build, and reuse that exact sentence in every prompt. Consistency in the words you type matters as much as the reference image itself.

3. 03

   Use the tool's reference feature, not description alone

   Attach the actual reference image using the platform's character or image reference function rather than only describing it in words. Image references anchor identity far more reliably than text descriptions on their own.

4. 04

   Change one thing per generation

   Vary only the pose, outfit or scene in each new prompt while keeping the identity sentence and reference identical. Changing multiple things at once makes it harder to tell whether drift came from the pose or from the model losing the face.

5. 05

   Cull to the closest matches, not the best single image

   Generate several options per scene and select for facial consistency first, composition second. A visually striking frame with a slightly different face breaks the set more than a plainer, more consistent one.

Ready-to-use prompts

Studio portrait of [character description: age range, hair, face shape, distinguishing features], front-facing, neutral expression, plain grey background, even soft light, 85mm lens, high detail on face. This will be used as a reference image.

[Attach hero reference as character/image reference]. Same person, unchanged face and hair. New scene: [describe setting and pose]. Keep facial structure, hair and build identical to the reference.

[Attach hero reference]. Same person, identical face and hair, unchanged. Now wearing [describe outfit], standing in [setting], same lighting style as the reference.

[Attach hero reference as style/character reference]. Generate five frames of the same person in these settings: [list five scenes], each keeping face, hair and build unchanged, varying only pose and background.

[Attach hero reference and the drifted frame]. The face in the second image has drifted from the reference, the jaw and eyes are different. Regenerate the same scene, matching the reference face exactly.

Questions people ask

Will the face ever be 100% identical across frames?

Not reliably with current tools. Plan for small variation and select the closest matches, the way a casting director would pick the best takes rather than expecting an identical frame every time.

Does this work for real people, like a founder's face?

It works better for consistency across marketing scenes than for producing new, materially different photos of a real identifiable person, which raises consent and likeness concerns you should clear first.

Which matters more, the reference image or the written description?

The reference image does most of the work; the written description is what stops the tool reinterpreting features when the pose changes drastically. Use both together for the most reliable result.