Workflow for Generating Multi-Character Scenes

Two characters holding hands

Overview

A key objective for game studios using AI creative tools is the ability to consistently generate multiple subjects within a single scene. This capability opens up a multitude of mediums and use cases, yet many AI tools fall short in this area.

In this article, we present a workflow that enables you to utilize your trained character models and styles to quickly iterate and construct engaging scenes featuring multiple characters. This revolutionizes the way you create immersive gaming experiences, from multi-character scenes and marketing materials to web toons. The multi-character workflow unlocks a wealth of possibilities.

In the following sections, we will guide you through:

  1. Composing your character and style models.
  2. Generating your base image.
  3. Creating your characters.
  4. Rapidly assembling the scene in Canvas.

Below we will provide you with a fully reproducible workflow. Follow along to familiarize yourself with the steps, or apply it to your own styles.

Step 1: Composing the Character Models

Choose Your Models

In this workflow, we will highlight three different models found in the Platform Model section of both the general model gallery and the Composition Creation menu. The models are:

  • Wyn - a character model depicting a brown-haired cartoon knight.
  • Lizbeth - a character model depicting a white/blonde-haired cartoon princess.
  • Liminal Spaces - an illustrated landscape style model that creates stunning scenes.
Composing multiple characters

Begin Composing

Start by navigating from the homepage to the Compose button at the top of the screen. You will immediately see all the platform models. Use the search bar to quickly find the models for this workflow using the names provided above.

For this workflow, you will need to compose three different models. To create them, select the models and adjust their influence values to the amounts listed below. Simply click the Add button on the model thumbnail and adjust the slider on the left to match the numbers below.

The Model Formulas

Liminal Wyn - Wyn (0.85) + Liminal Spaces (0.6)

Liminal Lizbeth - Lizbeth (0.85) + Liminal Spaces (0.6)

Liminal Group”- Liminal Spaces (0.55) + Lizbeth (0.55) + Wyn (0.55)

The composition of the model blends

Now that your models are composed, it's time to move on to generating.

Step 2: Generate Your Base Image

Prepare for Generation

With your models prepared, it's time to generate your base image. This step is crucial for establishing the ideal design foundation. You can create your base image directly from the style model you're using or use a reference image sourced from elsewhere.

Navigate back to the homepage of the webapp and click Generate. Select the Liminal Group model you just created. This model is useful because it contains elements of both characters, making it easier to recreate the characters later. We're keeping this simple for this workflow, but you can use the same principles to create more complex scenes in the future.

Prompt and Go

Start by prompting 'a knight talking to a princess' and click generate. You should see various scenes showing a blonde-haired princess talking to a dark-haired knight. Choose the one you like best or keep generating until you find one you prefer. Once you’re done, save the image to your computer or click Add to Collection and create a new collection for this project.

The raw base image

Step 3: Refining the Knight

Prep the Image

Once you have your base image, drag and drop it into the Reference Image area of your generation screen. Then, click the model name on the top left of the screen, currently saying Liminal Group. This will open the model gallery. Click Liminal Wyn to change the model, allowing you to refine the knight.

Switching the model.

Use the Sketch Tool

Click the Sketch button above your reference image, resembling a little pencil. You will see a zoomed-in picture of your image. Click the eyedropper tool and use it to match the color of the sketch tool to your background. Use this color to color block out the princess, leaving only the knight in the scene. (If your background has more than one color, you may need to change your color swatch a couple of times.) Once satisfied, click Apply.

Removing the princess.

Prompt and Generate

Shorten the prompt from 'a knight talking to a princess' to just 'a knight'. Because you previously used the Liminal Group model, the knight should already have some characteristics of the knight character you're trying to create. Set the Reference Image Influence to 20 and generate.

The goal of this step is to refine the character, so don't worry about other details in the image. Once you're satisfied with the character you’ve created, download the image or save it in your collection for later.

Raw unedited knight.

Step 4: Refining the Princess

Prep and Generation

Repeat the previous step with a few adjustments. Re-upload the original reference image you'd like to use for your scene, and adjust the model to Liminal Lizbeth. Follow the same color blocking technique as before, but this time, block out the Knight character and leave the princess intact.

Once you've set up your reference image with the Sketch tool, change the prompt from 'a knight talking to a princess to just a princess'. As before, adjust the Reference Image Influence to 20 before generating.

Rough draft of the princess

This step will help you refine the princess character. Once satisfied, you'll have all the building blocks for your project.

Step 5: Finalizing in Canvas

Open a Project

Now, launch a project in the Canvas tool. The easiest way for this workflow is to do so directly from the original reference image. Navigate to Models > Your Models and click on the Liminal Group model. Then click Images and find the reference image you've been working with. If you've saved your images in a collection, navigate directly to that collection from the left-hand menu and quickly find the image there.

Click the image and select > Edit in Canvas from the menu above the image. This will preload your prompt and reference image. Name your project to get started.

Navigating to Canvas

Upload the Second Image

Once in Canvas, click Import icon from the left-hand toolbar.

Upload the refined princess image. This places that image as the current top layer of your canvas. Ensure the image is centered. Click on the image directly and select Remove BG from the tool menu above.

Next, select the Eraser button.

With the refined princess layer selected, erase any areas you don't want to be carried over to the final image that the background removal may have missed.

Removing unneeded details.

Upload the Third Image

Repeat the same steps with the refined knight image, ensuring it's centered. Now you should have three layers: your base image in the back and the two characters superimposed on top. There may be some remaining rough edges, which we'll now fix.

Superimposing new characters.

Mask and Generate

Select the Paintbrush Mask symbol on the left-hand toolbar.


Mask out any seams on the images that look rough or don’t match the overall scene. Go ahead and mask out any areas you notice that need smoothing.

Before generating your improved edges, click Reference Image below your prompt box and adjust the Influence slider to 40. This ensures flexibility while preventing artifacts or unwanted details from being added. Now, click Generate below the prompt.

Optional Tips

After a couple of minutes, you'll see new interpretations of the areas you masked out. If they're too varied from the original image, increase the Reference Image Influence. If they're not varied enough, reduce it. If you want to add something to the scene that doesn't yet exist – a prop, for example – use the Sketch tool to roughly color block new details before masking.

A princess holding hands with a knight.

Step 6: Exporting and Optional Enhance

Save and Export

Once you're happy with the scene, click Export and Download as PNG. You can also download the layers as separate image files by selecting Save All Layers.

Enhance

If your scene uses a zoomed-out perspective and you notice lost details, fix them by navigating back to the homepage and clicking Enhance. Upload your final image and generate. We recommend the Anime - Balanced setting for the style in this workflow.

Upscale example

Final Notes

Now you know how to create easy multi-character scenes with Scenario. While these steps work for the models presented in this workflow, they should also apply to your own styles and characters. We recommend incrementally adjusting settings to find the ideal parameters for your models if you choose to customize the workflow!

Table of content

You'll also like