Scenario

Hi, how can we help you today?

Generate Multi-Character Scenes

This guide walks you through creating scenes with multiple distinct characters in a single generated image, using Scenario's Multi-LoRA feature.

By merging character models together, you can generate consistent and unique subjects interacting within one cohesive composition. Whether for game assets, or concept art, this approach ensures each character maintains their individual traits while appearing together naturally.

Why Use Multi-LoRA for Character Generation?

Scenario allows you to train custom models, including both style models and character models. Character models are particularly valuable for maintaining consistency in a specific character's design across multiple images. 

By merging two separately trained character models into a Multi-LoRA composition, you can generate dynamic interactions while ensuring each character retains their unique features.

This technique works effectively with both Flux and SDXL models.

Creating a Multi-Character Model

Step 1: Merge Character Models

To generate two distinct characters, you'll need to merge their individual models into a single composed model. Navigate to Models and click "New Model," then select "Compose Models" to open the model merging interface. For detailed information about merged models, see our Multi-LoRA section.

Choose two character models to merge. In this example, we'll combine a "Red-Haired Superheroine" (Character A) and a "Green-Skinned Zombie" (Character B), each trained separately with their own unique traits.

When prompting these characters later, describe each using language similar to the captions they were trained with for best results.

For more generic information about merged models (Multi-Lora), please refer to our Multi Characters Scene video.

Step 2: Test the Composed Model

Once the merged model is created, it will appear in your model gallery, ready for testing:

A 3:2 image format often works well for character placement, though you can experiment with other ratios based on your needs.

Create a structured prompt that describes each character's appearance and actions separately. For example:

"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, a dark crimson cape, and silver forearm protection, is sitting on a table playing chess with a green-skinned zombie with scruffy brown hair and wide yellow eyes. His face is smeared with blood, and he wears ragged clothes with a backpack."

Generate several test images and evaluate the results.

Be aware that features from “Character A” and Character B may occasionally blend together. In our example below, the red-haired woman might show "red scars" similar to the zombie's face. If this happens, you can adjust your prompts with more descriptive details or adjust the influence of each model component.

Step 3: Refine Character Consistency with Canvas

For more precise control over each character, use Scenario Canvas to adjust details separately for each character.. First, open your generated image and click "Retouch" to load it in Canvas.

Mask (select) the part of Character A that needs editing, then adjust the model component influences in the top-left panel:

  • Increase the influence of Character A's model (the heroine)

  • Lower Character B's model influence (the zombie) to zero

Use a simple prompt focusing on Character A, such as "A female character with short bob red hair," and click generate to update the masked area.

Repeat this process for Character B, masking their portion of the image and adjusting influences accordingly.

This method gives you precise control over each character's appearance, ensuring they remain distinct while sharing the same scene.

Step 4: Finalize and Enhance

Once you're satisfied with your multi-character composition:

  • Save the image to your gallery via the "Export" option in the top-right corner.

  • Use the Enhance tool to upscale your image for higher resolution and finer details.

  • This process can be extended to three or more characters, though it requires careful balancing of model influences and prompts.

Below is an example of three character models that were trained independently before being merged into a multi-character composition, using the workflow described above. Each character retains its unique design and does not lose defining details when placed in the same scene.

Final Notes

Multi-LoRA in Scenario provides a powerful approach for generating consistent, multi-character scenes with minimal effort. Rather than drawing multiple characters separately and compositing them together, you can create natural interactions while maintaining each character's unique identity.

Start by merging character models, generate compositions, and refine in Scenario Canvas as needed. The more you experiment with this technique, the better you'll become at controlling how characters appear and interact within your scenes, creating compelling visual narratives for games or illustrations.

Check out our Multi-Character tutorial (Part 1) for a more in-depth look as well as Part 2 (train Dual Character Models). You can also refer to this tweet.

Access This Workflow Via API

Resources:

Was this helpful?
Quentin

Quentin