How to Keep the Same Character in AI Images: Screen Sets with Character Asset Pack

The hardest problem in AI storytelling is consistency. Generate “a blonde warrior” ten times and you get ten different women. For a film, your hero must look the same in every shot: the same face, the same clothes, the same sword. In Proxima Art this is solved by Character Asset Pack. You give it one image, it draws your character from several angles — a screen set — and you save the result as a character with a short name. From then on, that name works as a tag in your storyboard.

What is a screen set?

A screen set is a single reference board: your character in several poses and camera angles, with a consistent face and outfit. Film and game studios call this a model sheet. For the AI it is a useful reference, because it shows what the character looks like from the front, the side and the back — information a single portrait does not have.

Opening Character Asset Pack

Open Proxima Studio (Create menu). You will see tool cards such as Character Asset Pack, Storyboard and Video Pro Generator. Click Create Character Pack. The form is numbered from top to bottom, and the right column, Asset Preview, shows examples and later your generated results.

1. Asset Type

Character Asset Pack can create five kinds of assets:

• Screen Sets — a clean reference board with selected views and a consistent outfit and face. This is what you need for the storyboard.
• Character Sheet — a hero render with readable story, traits, skills and equipment.
• Character Dossier — a poster combining poses, stats, story and equipment panels.
• Expressions — a grid of consistent emotional expressions and head angles.
• All — the full pack: poses, sheet, dossier and expressions in one run.

2. Story Idea (optional)

A short description helps the model see the bigger picture and can change details of the subject — for example add a weapon or change the outfit. For the first heroine the video uses: “A beautiful 25-year-old blonde warrior with European features. She wears practical nomadic clothing and carries an ornate dagger and a large translucent crystal sword.”

3. Input Source

Add a reference: Upload reference images from your device, or Select from Library to pick an image you generated earlier. One clear image is enough; more images of the same character can make it even more consistent. If you have no image yet, Generate base character from short prompt creates one from a few words.

4. Pose Presets — which poses can you choose?

There are nine poses. Six are selected by default:

• Close up, Front view, Left side view, Back view, Top-down view, Low-angle view (default)
• Right side view, 3/4 front left, 3/4 front right (optional)

The default six cover what a film needs: a face close-up for emotional shots, full-body views from three sides, and two dramatic camera angles.

5. Global Controls: model, style, format and resolution

This is where quality is decided. The model list offers:

The aspect ratio options are Square (1:1), Portrait (3:4), Landscape (4:3), Poster (4:5, not available for Qwen), Wide (16:9) and Story (9:16). Resolution is 1K or 2K; 1K is enough for a storyboard reference.

From screen set to named character

When the screen set is ready, open it full size and check the details: the face, the outfit, the weapons. If you like it, click Add / Create Character. For your first character, the creation form opens right away:

• Character name (required) — short, without special characters. It becomes the tag you use in Storyboard prompts; the dialog shows a live Prompt tag preview, for example {iraya}.
• Primary media — which original reference becomes the main image of the character.
• Notes — optional, only for you.

Once saved, the character appears in your Character Library. When you later write {iraya} in a storyboard panel, Proxima adds her description and reference images to the generation automatically — you never have to describe her again.

Creating a second character

For the companion, Ashara, the process is the same with three important differences:

1. Remove the previous reference before adding the new image. Otherwise the old reference distorts the new screen set.
2. Write a new story idea: “A beautiful 25-year-old Black warrior woman with deep brown skin. She holds a long diamond-crystal spear with a sharp black obsidian tip.”
3. When saving, choose Create New Character instead of Use Existing Character, so the screen set is not attached to Iraya.

Use Existing Character is useful later, when you want to add more assets (for example an Expressions grid) to a hero who already exists.

Tips

• Start from a clean full-body image. A single character on a simple background gives the best screen set.
• Mention props in the story idea. Weapons and accessories you name are more likely to appear in every pose.
• Keep names short and unique: Iraya, Ashara, Kai. Short tags are easy to type and impossible to confuse.
• A storyboard needs at least one character with a screen set, so do this step before you start the storyboard.

Common mistakes

• Leaving the old reference in place — the new character inherits the old one's face.
• Adding a new hero to an existing character — always choose Create New Character for a new person.
• Long names with spaces or symbols — they make clumsy tags. Use one simple word.
• Retrying the same refused prompt — if GPT Image 2 Edit blocks it, switch to Qwen 2.1 Image Edit instead.

With two named characters ready, you can start the storyboard in the next lesson.

How we used this in Sleeping Sands of Kharuun

At 02:32 the tutorial makes the screen set for Iraya, then saves the named character at 03:48. At 04:15 it repeats the process for Ashara. The reference boards provide views of the same face, outfit and weapons, so the storyboard can use the characters repeatedly rather than starting from unrelated portraits.

Iraya carries the crystal sword and Ashara the spear. Those props matter later: when they disappear in a generated video, the continuation prompt must explicitly preserve them. A screen set supports consistency; it does not guarantee that every later frame will preserve every object.