How to Keep One Consistent Character Across All Your AI Clips (Full Guide)

So how do you solve character consistency when you create ads with AI?

Many “AI experts” show all sort of solutions, the one that works for me without fail is the character sheet or turnaround sheet method.

In this guide you’ll learn how to prompt it, see it in action on the workflow canvas and MCP and get a free skill that does it for you.

What a character sheet is

Character sheet locks the entire look of the character from several directions to make sure that in any scene it will always look the same.

The point of the sheet is that the model sees the character from several angles instead of one. When using only one image the AI model needs to infer how the character looks from other sides or angles which leaves you open to mistakes, the AI in many cases will get certain details wrong completely and will ruin your shot, ending in more wasted credits.

How to create your character sheet

You need two things: an image and the prompt from this guide.

Take the image of the actor or the character you want.

Give it to your Claude or ChatGPT with the instructions to fill in the specific elements from your reference image in the master prompt, or do it manually. Make sure the details of the entire look make sense, if your original actor image shows only upper half or bottom half.

Generate with your character sheet, attaching your actor image as reference. We use GPT Image 2.5 Flare for this. Nano Banana 2 is also very effective for these, it comes out really good, and it’s cheaper and faster.

Play with both and see which one works best for your character.

Important conditions for this to work

Generate at 16:9. The 4-panel row needs the width.

If you have multiple characters you should create the sheet for each.

The prompt

Using the person in the reference image as the exact identity source, create a character reference sheet on a pure white background (#FFFFFF), studio lighting, no shadows on the background. One horizontal row of exactly four panels, side by side, all the same height.

From left to right:

1. Far left: large close-up of their face and head, front-facing, filling the panel.

2. Full body shot, standing, facing the camera straight on, arms relaxed at their sides, head to shoes fully visible.

3. Full body shot, standing, body rotated 45 degrees so they face the LEFT edge of the image. Their nose points toward the left edge. The viewer sees more of the RIGHT side of their face and body.

4. Full body shot, standing, body rotated 45 degrees the OPPOSITE way so they face the RIGHT edge of the image. Their nose points toward the right edge. The viewer sees more of the LEFT side of their face and body. Panels 3 and 4 are exact mirror images of each other — they must face in opposite directions, never the same direction.

Every panel: neutral expression with mouth slightly open, relaxed lips.

Preserve their identity with 100% fidelity in all four panels — identical face across every panel: [FACE]. [HAIR]. [ACCESSORIES]. Do not beautify, do not change their apparent age, do not stylize. If the reference is cropped or partial, reconstruct the unseen parts fully consistent with the visible features and natural proportions.

Outfit in all full-body panels: [OUTFIT].

Photorealistic, white cyclorama, soft even lighting, 85mm look, no text labels, no watermarks, no grid lines, no borders between panels.

How to use the sheet

From here on the sheet replaces the original photo as your reference.

When you want to create any clip, you take the character sheet as the first reference, and the scene shot (if you have shots) as the second reference. You put them together with your clip prompt and generate. The character stays consistent across every clip, and if the sheet is in a specific style, the style carries over with it.

This works on any model that accepts image references.

The Claude skill that runs all of it

You can run all of this manually using the UI or the workflow canvas, or you can use it directly with the MCP inside Claude, using a fine-tuned skill that was built especially for this purpose.

With the skill, you drop the actor’s photo and Claude analyzes the face, hair, and outfit, fills the prompt slots for you, and generates the sheet through the Maxfusion AI MCP without leaving the chat.

There is also an already made canvas workflow. You drop the actor’s image into the input node, run it, and get the character sheet back.