Consistent Character Variants (Character Reference)

This guide walks you through using Scenario's Character Reference feature to generate consistent character images from a single reference. This approach provides a quick, streamlined way to create variations of the same character without the time commitment of full model training. 

While it doesn't offer the same level of customization as a dedicated character model, Character Reference is perfect for fast iterations when speed and simplicity are the priorities.

Why Use Character Reference?

Character Reference offers several advantages for quick character variations:

It's significantly faster than full model training since you only need one reference image instead of a complete dataset. The feature maintains consistency in key facial features and structural elements across all generations. It's ideal for quickly exploring variations in expressions or poses while keeping your character recognizable. The process is simple and immediately accessible, making it perfect for rapid concept exploration.

This approach works best with a clear, high-quality reference image as your starting point.

Step 1: Upload Your Character Reference

Begin by selecting a strong reference image of your character:

Choose one clear, high-quality image that clearly shows your character's defining features. For optimal results, use an image that was generated using a custom model that matches your desired style, ensuring better consistency in generations.

For best results, use the same model throughout all generations to maintain a cohesive look across variations.

In the left panel of the generation interface, select 'Character Reference' from the dropdown menu, then if your image is not already pre-loaded as a Reference Image, you can upload a high-resolution image of your character (1024x1024 square format recommended for optimal results).

Step 2: Use Prompts for Character Variations

With your reference loaded, craft prompts that explore different aspects of your character:

Focus your prompts on the specific variations you want to see, such as changing poses, expressions, or outfits while maintaining core character features.

Example prompts include: "The character is smiling with a joyful expression" "Wearing a punk outfit with a leather jacket and spiked collar" "Sitting on a futuristic throne with a determined expression"

The key advantage is that core character features like facial structure and proportions remain consistent, while only the specified details adjust based on your prompt.

Step 3: Adjust Influence for Flexibility

Adjust how closely the output adheres to your reference using the influence parameter:

• Higher influence (80-100): Keeps facial structure and core details virtually unchanged, focusing variations only on elements explicitly mentioned in the prompt.

• Lower influence (40-60): Allows for more creative freedom in expression, attire, and lighting while maintaining recognizable character traits.

• Mid-range influence (60-80): Creates a balanced approach that preserves key character features while permitting natural variations.

Avoid setting the influence too high (95-100), as it can lead to overly similar results with minimal variation between generations.

Step 4: Refine with Scenario Canvas

For more precise control over specific details:

• Open Scenario Canvas by selecting 'Retouch' on your generated image.

• Use masking and inpainting tools to modify facial details, adjust colors, or fix any minor inconsistencies.

This step ensures precision while retaining the original style and quality, allowing you to perfect elements that might not have generated exactly as intended.

When to Use Character Reference vs. Full Model Training?

Use Character Reference when you:

• Need quick, consistent character images from just one reference image

• Want to explore different poses or expressions, and outfits without extra training time

• Prefer a fast, lightweight approach without long setup times

• Are comfortable with slightly less consistency than a fully trained model

Use Full Model Training when you:

• Need a fully customizable character model with fine-tuned consistency

• Want to train the AI on multiple angles, expressions, and poses for greater variation

• Require a high degree of control over style and details

• Can invest more time into gathering images and training

Final Notes

Character Reference provides an excellent balance between speed and consistency, making it a valuable tool for rapid iteration and concept exploration. You can even use Character Reference to generate initial variations, which can then serve as a dataset for training a more advanced character model later.

Experiment with different prompts, influence levels, and refinements to find the perfect balance for your specific character design. This approach streamlines your workflow and reduces the time between concept and execution, allowing you to focus more on creative exploration and less on technical setup.

Take a look at our Character Reference tutorial for a step-by-step.

Access This Workflow Via API

• Upload an Image: Scenario API Documentation - POST /assets

• Character or Style Reference: Scenario API Documentation - POST /generate/txt2img-ip-adapter

• Inpainting Generation: Scenario API Documentation - POST /generate/inpaint