Omegance
A Single Parameter for Various Granularities in Diffusion-Based Synthesis
Key Innovation
In this work, we introduce a single parameter ω to effectively control granularity in diffusion-based synthesis. This parameter is incorporated during the denoising steps of the diffusion model's reverse process, enabling precise control over the level of details in generated outputs without requiring model retraining, architectural modifications, or additional computational overhead during inference.
Key Advantage: Spatial masks or denoising schedules with varying ω values can be applied to achieve region-specific or timestep-specific granularity control.
Prior knowledge of image composition from control signals or reference images further facilitates the creation of precise ω masks for granularity control on specific objects. To highlight the parameter's role in controlling subtle detail variations, the technique is named Omegance, combining "omega" and "nuance".
How to Use Omegance
Adjust Parameter
Use sliders to adjust ω parameter values and see real-time effects
Observe Effects
Compare image detail changes under different ω values
Apply Technology
Apply the technique to your diffusion model projects
Continuous Output Visualization
Drag the sliders below to observe real-time effects of ω parameter on image details
Sample Image 1
Sample Image 2
Sample Image 3
Sample Image 4
Comparison Results
Drag the divider line to compare image effects under different ω parameters
Less Detail (ω +)
More Detail (ω -)
Text-to-Video Results
Demonstrating the powerful effects of Omegance in video generation
Latte Model
"A cartoon panda in a sparkly bowtie performs a cheerful dance in a bamboo forest."
Original (ω = 0)
ω Enhanced
AnimateDiff Model
"A panda surfing in the ocean, realistic, highquality."
Original (ω = 0)
ω Enhanced
Live Demo
Local Control (ControlNet)
Precise control over specific image regions
Global Control
Uniform detail adjustment across the entire image
BibTeX
@inproceedings{hou2025omegance,
author = {Hou, Xinyu, and Yue, Zongsheng and Li, Xiaoming and Loy, Chen Change},
title = {{Omegance}: A Single Parameter for Various Granularities in Diffusion-Based Synthesis},
journal = {International Conference on Computer Vision},
year = {2025},
}