Why does character consistency break across AI images?

Image models regenerate everything from scratch each time, so small wording changes drift the face, hair or outfit. The fix is to keep an identical, detailed description of the unchanging traits in every prompt and only vary the scene-specific words.

What should go in the locked block versus the variable block?

Locked tokens are anything that defines who the character is — name, age, face shape, hair, eye colour, build, skin tone, signature clothing and accessories. Variable tokens are anything that should change per shot — pose, camera angle, expression, lighting and setting.

Does this work for Midjourney, DALL·E and Stable Diffusion?

Yes. The locked-token approach is model-agnostic. For even tighter consistency, pair it with model features like Midjourney's --cref / character reference, DALL·E reference images, or a trained LoRA in Stable Diffusion.

Is my character description sent anywhere?

No. The sheet is assembled entirely in your browser from the text you type. Nothing is uploaded, stored or logged.

Consistent Character Prompt Sheet

Consistent character prompt sheet

The hardest part of telling a story with AI images is keeping the same character looking like themselves across many shots. Models regenerate every pixel from scratch, so even a reworded prompt can change the face, hair or outfit. This sheet solves that by splitting your prompt into a locked block (identity that never changes) and a variable block (pose, expression and setting that change per image).

How it works

You describe the character once in detail — name, physical traits, clothing and signature features — and the tool freezes that into a reusable locked block. For each new image you only change the variable tokens: the pose, the expression and the scene. Because the identity description is byte-for-byte identical in every generation, the model has the same anchor each time and the character stays recognisable. The tool outputs both the full prompt for the current shot and the isolated locked block so you can paste it into future prompts unchanged.

Tips and notes

Be specific in the locked block. “Brown hair” drifts; “shoulder-length wavy chestnut hair, side-parted” holds. The more concrete the identity, the less the model improvises.
Never paraphrase the locked block. Copy it verbatim every time. Reordering or rewording words is the most common cause of drift.
Pair with reference features. Use Midjourney --cref, a DALL·E reference image, or a Stable Diffusion LoRA on top of the locked text for the tightest consistency.
Keep one shot as your canon. Once you get a render you love, treat it as the reference image and feed it back into image-to-image generations.