ComfyUI Video Anti-Aliasing Pack

Advanced anti-aliasing nodes for videos and images. VideoTAADLAA combines TAA with a DLAA-inspired pipeline for sharp, stable results, while VideoAdaptiveAA targets specific aliasing-prone regions. The system utilizes temporal accumulation, jittered sampling, and CNN-based refinement for superior edge quality.

Important

Disclaimer This node does not use NVIDIA's closed-source SDKs or native DLAA/DLSS binaries. Instead, it is a custom adaptation of Temporal and Spatial Anti-Aliasing techniques commonly found in modern game engines, rebuilt entirely from scratch using PyTorch tensor architecture specifically for ComfyUI video post-processing.

Preview

TAA + DLAA Anti-Aliasing	Adaptive Anti-Aliasing

Features

Temporal Anti-Aliasing (TAA): Blends frame history with motion information to eliminate flickering and sub-pixel jitter.
Neural Reconstruction: (MSE delta: 0.000023) Lightweight, DLAA-inspired model trained to restore edges with near-lossless accuracy.
Ghosting-Free Technology: Motion-aware suppression logic to minimize trailing artifacts in high-speed sequences.
Deterministic 4-Offset Jitter: Improved sampling stability for sharper, more consistent sub-pixel reconstruction.
Edge-Preserving Clarity: Sobel-based edge detection ensures fine textures stay sharp while aliasing is eliminated.
Frequency-Aware Refinement: Optimized through Frequency and Perceptual loss for "crystal clear" results without artificial blurring.

Comparison

AdaptiveAA Comparison

TAA + DLAA Comparison

🚀 How to use

Test the effect by increasing dlaa_strength to 0.8 or 1.0.

Parameters

taa_strength: Controls the impact of past frames. (Default: 0.6 | Fast motion: 0.7 | Static: 0.95)
taa_alpha: Determines how quickly history updates. (Default: 0.4 | Aggressive AA: 0.3 | Ghost-free: 0.05)
motion_sensitivity: For high-speed videos, try decreasing. (Default: 0.08 | Flicker-prone: 0.0 | Ghosting Risk: 0.3)
jitter_scale: Amount of sub-pixel shifting for edge detection. (Default: 0.02 | Off: 0.0 | Maximum: 0.08)
dlaa_strength: Strength of the deep learning sharpening. (Default: 0.3 | Pure AA: 0.0 | Aggressive: 1.0)
edge_threshold: Determines which pixels are detected as edges. (Default: 0.2 | Hard edges: 0.35 | All: 0.05)
blur_radius: The radius of the blur applied to edges. (Default: 0 | Softer: 2–3 | Very soft: 4–5)
reset_history: Clears temporal history memory. (Default: False | Enable on scene change: True)

Installation

Recommended: Install via ComfyUI-Manager.

Manual installation:

cd ComfyUI/custom_nodes/
git clone https://github.com/MSXYZ-GenAI/comfyui-msxyz.git

License

This project is licensed under the MIT License - see the LICENSE file for details.

4.0 KiB Raw Blame History Unescape Escape