If you are creating comics, webtoons, or short drama series with AI, you know the pain: your protagonist looks like a different person in every panel. One frame has a sharp jawline, the next has a soft chin, and somewhere along the way, their jacket changes color. A new tutorial published on DEV.to by user deepfox offers a practical fix. The solution is building a character reference sheetβ€”a single image containing front, side, and back views that serves as the ground truth for every subsequent generation. This method stops the model from hallucinating new features and forces consistency.

Start With a Single Photo or Text Description

The guide begins with the simplest entry points. If you already have a character portrait, upload it to CVY.AI’s Image to Image tool with a 16:9 aspect ratio. The prompt instructs the AI to create a sheet with three full-body views on the left and face close-ups on the right, keeping hair and clothing identical to the source. For creators who only have an idea in their head, the tutorial suggests a Text to Image approach using a 9:16 vertical layout. This generates a four-view turnaround (front, left, right, back) in two rows, ideal for vertical scrolling formats like webtoons. The key is using a pure white background and soft studio lighting to keep the focus entirely on the character model.

Modular Design: Splitting Face, Body, and Style

As you advance, the tutorial recommends a modular approach to avoid starting from scratch every time you want a change. Instead of one giant prompt, you upload three separate images: one for the face, one for the body build, and one for the art style. The prompt explicitly tells the AI what to take from each image and, crucially, what to ignore. For example, if the style reference image contains a character, you must instruct the AI to ignore that character's face and clothing, copying only the line art and shading. This separation allows you to swap out a body type or outfit without losing the specific facial features you’ve already perfected.

Advanced Props and Poses

For complex scenes involving props, the workflow expands to include four or five reference images. The example provided uses a biker character, requiring separate uploads for the face, body, riding suit, and motorcycle. To prevent face leakage, the body reference is generated with a helmet on, and the suit is shown on a headless mannequin. This ensures only one face exists in the reference set. Finally, for dynamic action shots, the guide introduces a 'pose image'β€”a simple grey 3D figure in the desired position. By adding this as the fifth reference, the AI can map the consistent character onto a specific action, such as crouching beside a bike, without distorting the anatomy.

Key Takeaways

  • Faces hold best: Small details like scars, stubble, and tooth shapes transfer reliably across generations.
  • Side views are tricky: You must explicitly prompt 'only one eye visible' to prevent the AI from turning the head too far in profile views.
  • Isolate elements: Use headless mannequins or helmets in reference images to prevent unwanted faces from bleeding into the final character.
  • Modular prompts work: Separating face, body, and style into distinct reference images allows for easier iteration and customization.

The Bottom Line

Consistency is the biggest hurdle in AI-assisted storytelling, and this tutorial provides the exact prompts needed to clear it. By treating character design as a modular assembly of faces, bodies, and styles, you gain control over the output rather than hoping for a lucky roll of the dice. It is a welcome, practical resource for anyone serious about building coherent visual narratives.