Dzine Consistent Characters: Build One, Reuse Everywhere

Dzine (dzine.ai, formerly Stylar) is a design-led image generator with a large style library, a layered canvas and a dedicated tool for consistent characters. That last part is what sets it apart for anyone making a comic, a children’s book, a brand mascot or a run of social posts with the same face. Instead of fighting the model with long descriptions every time, you create the character once from reference images and reuse it across scenes. This guide walks through building a character that holds, prompting with it, and fixing the drift that still creeps in.

How Dzine’s consistent-character tool works

The tool takes a small set of reference images of one character and learns the identity: face shape, hair, skin tone, build and, to a lesser extent, clothing. Once saved, the character appears in your library and can be selected when you generate new images. Dzine then conditions the output on that identity while your prompt controls everything else.

Think of it as separating two jobs that normal prompting mixes together. The saved character answers "who is this?" and the scene prompt answers "where are they and what are they doing?". When you keep that split clean, consistency goes up and prompts get much shorter.

Menu names and button labels change between updates, so if the control you are looking for has moved, search the tools panel for "character" and you will find it in the current interface.

Step by step: build a character that holds

1. Design the character first, in one sitting. Generate a single hero image in the style you intend to use for the whole project. Iterate until face, hair and outfit are exactly right. Everything downstream copies this image, so spend the credits here.
2. Produce reference variations from the hero. Use image-to-image with a high structure setting to create a front view, a three-quarter view, a profile and a full-body shot. Keep lighting neutral and the background plain. Three to six references is the useful range; more images with inconsistencies hurt rather than help.
3. Create the character. Open the consistent-character tool, upload the references, and give the character a short, distinctive name you will remember when you type prompts later.
4. Test with boring prompts. Before anything ambitious, generate "standing in an empty room, neutral expression" and "sitting on a chair, looking left". If the face already drifts here, rebuild with cleaner references.
5. Lock the style. Pick one style from the library for the project and reuse it for every scene. Switching styles mid-project is the fastest way to lose consistency, because each style reinterprets faces differently.
6. Record what worked. Note the character name, style, and any strength values alongside each finished image so you can recreate the setup in a month’s time.

Prompting with a saved character

Once the character exists, the prompt should stop describing them. Write the scene instead:

• Pose and action. "Leaning on a balcony rail, holding a mug, half smile."
• Setting and light. "Rooftop at dusk, warm street lights below, soft rim light."
• Framing. "Medium shot, eye level" or "wide shot, character small in frame".
• Outfit changes, stated explicitly. If the character normally wears a green coat and you want a raincoat, say "wearing a yellow raincoat" or the model will default to the reference clothing.

Avoid re-describing the face or hair. Contradictions between the saved identity and the prompt produce a blend of both, which looks like a sibling rather than the same person. If you need a close-up, say "close-up portrait" and leave the features to the character.

For the general principles behind reusable characters on other tools, the Midjourney consistent characters guide covers the same thinking with a different mechanism.

Settings at a glance

Credit costs vary by tool and output size, so check the platform’s current pricing page before planning a long series.

Fixing drift

• The face changes between scenes. Usually a style switch or a prompt that re-describes features. Return to the project style and strip facial description from the prompt.
• Clothing keeps reverting. The references all show one outfit. Either add a reference in the new outfit or state the clothing explicitly in every prompt.
• Body proportions shift at full length. Include at least one full-body reference. If the tool only saw head-and-shoulders shots, it guesses the rest.
• Extreme angles fail. Views from directly above or below are hard for any model. Generate a conventional angle, then use the canvas and image-to-image to adjust.
• Two characters merge. Generate each character separately on the canvas and compose them as layers rather than asking for both in one prompt.

Try it

Character: [your saved character name]
Prompt: Sitting at a small cafe table by a window, writing in a notebook, soft morning light from the left, medium shot, eye level, calm expression.
Style: [your project style]
Aspect ratio: 4:5

Keep everything the same and change only the action ("stirring a coffee", "looking up at someone off frame") to confirm the identity holds. Once three or four variations match, you can start changing settings and lighting with confidence.