Using Reference Images / Image Influence
Fater's AI models can be guided not just by text prompts and control layers, but also by existing image content. This influence manifests in several key ways, allowing for nuanced control over your generations.
1. External Reference Images
Some AI models, particularly certain types in the "Generate" or specialized "Edit" categories, allow you to upload one or more External Reference Images. These images serve as additional visual input to the AI, beyond your text prompt.
Purpose:
Style Transfer: Guide the AI to generate content in the artistic style of the reference image(s).
Content Guidance: Influence the subject matter, objects, or composition of the generated image.
Product Consistency (Advanced): Potentially help maintain the appearance of a product across multiple generations (model dependent).
How to Use:
Select an AI model that supports external reference images (check its parameters in the Left Sidebar or the AI Model Directory).
Locate the "Reference Image(s)" input area in the Left Sidebar.
Upload your desired reference image(s).
Write your text prompt. The AI will attempt to blend guidance from your prompt with visual cues from the reference image(s).
Key Parameters:
The model's control panel might include parameters like "Reference Strength" allowing you to control how strongly the reference image(s) influence the output.
2. Contextual Image Influence (Inpainting)
When using an Inpainting model (typically from the "Edit" category) to fill a masked area, the AI heavily relies on the pixels surrounding your mask for context.
Purpose: The AI analyzes the unmasked area around your selection to:
Understand the existing scene's lighting, textures, colors, and overall style.
Generate new content within the mask that blends seamlessly and consistently with this surrounding environment.
How it Works (Implicit): This is a fundamental aspect of how inpainting models achieve realistic fills. The broader the visible context around the mask (within the Generation Area), the better the AI can understand how to fill the masked region appropriately.
3. Source Image Influence within the Mask (Inpainting with Denoising Strength < 1.0)
For many inpainting models, there's a concept often referred to as "Denoising Strength" or an equivalent parameter (the specific name might vary per model or be tied to "Steps" or "Creativity" in some UIs). This parameter controls how much the AI alters the original pixels underneath the mask versus generating entirely new content based on the prompt and surrounding context.
Full Strength (e.g., Denoising Strength = 1.0 or "Destroy Fill Area Behind Mask" = Yes):
The AI completely disregards the original pixels under the mask.
It generates entirely new content within the masked area based on your prompt and the surrounding (unmasked) context.
Use this for completely replacing objects or filling empty areas.
Partial Strength (e.g., Denoising Strength < 1.0 or "Destroy Fill Area Behind Mask" = No):
The AI uses the original pixels under the mask as a starting point or structural guide and then modifies them according to your prompt and the surrounding context.
The lower the strength (closer to 0), the more the original pixels are preserved, and the AI makes more subtle changes.
The higher the strength (closer to 1), the more the AI overhauls the original pixels.
Use this for:
Enhancing or subtly altering existing elements within the mask.
Applying stylistic changes while retaining the underlying form.
Retouching or refining details.
Key Parameter: Look for parameters like "Denoising Strength," "Image Strength," "Creativity," or the inverse, such as "Resemblance." The "Destroy Fill Area Behind Mask" switch in models like most Edit models directly controls whether the underlying pixels are fully ignored (Yes) or can contribute (No).
By understanding these distinct types of image influence, you can more effectively direct Fater's AI models:
Use External Reference Images for broad stylistic or content guidance on new generations.
Rely on Contextual Image Influence (pixels around the mask) for seamless blending during inpainting.
Adjust Source Image Influence within the Mask (via parameters like Denoising Strength or "Destroy Fill Area...") to control how much existing masked content is preserved or altered during inpainting.
Last updated