Creating multiple distinct characters in a single image opens up endless possibilities for industries ranging from gaming & entertainment to advertising and marketing. This workflow leverages Scenario’s model merging feature (Multi-LoRA), making it easier to blend custom-trained models while ensuring consistency across scenes.

In this tutorial, we’ll guide you through:

- Merging custom-trained models

- Generating a set of base images and refining them

- Finalizing each character’s details in the scene using Scenario Canvas

Check out our video, before going through the tutorial:
‍

‍

Step 1: Merging Character Models

For multi-character scenes, one of the most efficient methods is to start with custom-trained character models, trained on individual characters. Navigate to the Models section and select New > Compose Models. Here, you can merge multiple custom models, ideally trained with distinct captions to ensure unique features for each character. For more details on how to train character models, please refer to this tutorial.

For example, combine the "red-haired superheroine" and the "green-skinned zombie." Start by setting both models to equal weights before adjusting their mix as necessary.

‍
For more details on how to merge model models, please refer to this tutorial. Once merged, save the composition to proceed.

Step 2: Generate Base Images

Typically, we recommend setting the image format to 3:2, but other formats can also work depending on your scene. When creating a prompt, follow these elements:

1. Description of Character A.‍

2. Action or scene that the two characters are part of.

3.. Description of Character B.

For instance:

• “A female character with vibrant messy short bob red hair and bright blue eyes, wearing a dark navy blue tactical suit with silver accents on the shoulders with a dark crimson cape, silver forearm protection
• Is playing chess on a table again
• a green-skinned zombie with scruffy brown hair, wide yellow eyes. His face is smeared with blood”.

You can also craft a prompt by starting with the action, followed by descriptions of both characters. This will help the AI distinguish each character clearly. Generate a few versions and, if characters blend too much, tweak the prompt or the weight of each model until you get a solid base image.

‍

Step 3: Refine Individual Characters

For finer details and to enhance consistency with the original model, switch to Canvas by clicking on Edit Detail in the top menu. Use Inpainting (mask) to refine each character individually without modifying the rest of the image:

- Mask Character A and regenerate only that character using a prompt focused on its features.

- Repeat the process for Character B.

You can select the entire character or specific parts, such as the face. Adjust the prompt accordingly based on your selection, including details like expression, pose or outfit details. Don’t forget to adjust the Influence slider in the left column for optimal results.

For more information on how the Scenario canvas works, please check out these articles.

Step 4: Finalize with Scenario Enhance

Once the characters are refined, use Enhance to further polish the image. Save the Canvas image to your gallery or collection and import it into Enhance. You can use 2x or 4x upscaling with the Precise presets, or high Image Fidelity settings. For additional refinement, you can even use some standalone images from the original characters’ training sets as Style Images.

Conclusion

This workflow provides flexible, precise control over how multiple characters appear and interact in a scene. It’s all about blending custom models and refining images in Canvas to achieve the desired result. This approach is adaptable across use cases, offering a powerful tool for generating consistent multi-subjects visuals with ease.

‍