ComfyUI Video Anti-Aliasing Pack (TAA + DLAA)
These ComfyUI nodes provide advanced anti-aliasing solutions for videos and images.
The VideoTAADLAA node combines Temporal Anti-Aliasing (TAA) and Deep Learning Anti-Aliasing (DLAA) techniques to deliver sharp and smooth results, while the VideoAdaptiveAA node smooths only jagged (aliasing-prone) areas.
Preview
| TAA + DLAA Anti-Aliasing | Adaptive Anti-Aliasing |
|---|---|
 |
 |
Features
- Temporal Anti-Aliasing (TAA): Stabilizes flickering in video sequences using frame history.
- DLAA (Deep Learning AA): Neural-based cleanup of residual aliasing without losing texture detail.
- Ghosting-Free Logic: Advanced history clamping prevents trailing artifacts in fast motion.
- Halton Sequence Jittering: Improves sub-pixel sampling for cleaner edges.
- Edge-Aware Weighting: Preserves fine details while blending.
- Adaptive Anti-Aliasing (AdaptiveAA): Targets only aliasing-prone regions for localized smoothing.
Compare
AdaptiveAA Comparison
 |
 |
TAA + DLAA Comparison
Parameters
🚀 TAA Parameters
-
taa_strength: Controls the overall impact of past frames on the current frame. Higher values yield smoother results but increase the risk of ghosting on moving objects.- Recommended:
0.9- Provides a good balance for most situations. - Fast motion:
0.7- Reduces ghosting in fast-paced scenes. - Static scene:
0.95- Maximizes smoothness in still/slow scenes.
- Recommended:
-
taa_history_alpha: Determines how quickly the history frames are updated. Lower values update the history slower, providing a more stable image.- Recommended:
0.1- Offers good stability. - Aggressive AA:
0.3- More aggressive smoothing, but might introduce flickering. - Ghost-free:
0.05- Minimizes ghosting trails.
- Recommended:
-
jitter_scale: Controls the amount of sub-pixel shifting (jitter) applied to each frame for superior edge detection.- Recommended:
0.5- Ideal for effective anti-aliasing. - Off:
0.0- Disables jittering. - Maximum:
1.0- Maximum effect, but might cause slight blurriness in fine details.
- Recommended:
✨ DLAA Parameters
dlaa_strength: Adjusts the strength of the deep learning-based sharpening and edge correction neural network.- Recommended:
0.7- Retains TAA smoothness while sharpening the image. - Pure AA (no sharpening):
0.0- Only uses TAA and Edge AA effects. - Aggressive sharpening:
1.0- Provides maximum sharpening.
- Recommended:
🎨 Edge AA Parameters
-
edge_threshold: The threshold value determining which pixels are detected as edges. Lower values will target more fine details.- Recommended:
0.08- Good balance between fine and prominent edges. - Only hard edges:
0.2- Targets only high-contrast, prominent edges. - Almost everything:
0.02- Applies edge smoothing to a massive portion of the image.
- Recommended:
-
blur_radius: The radius of the blur applied to the detected edges.- Recommended:
1- Provides light and effective edge smoothing. - Softer:
2–3- For noticeably softer edges. - Very soft:
4–5- For heavily smoothed, almost "cartoonish" edges.
- Recommended:
⚙️ Control
reset_history: Clears the temporal history memory.- Normal:
False- Memory persists continuously between frames. - On scene change:
True(for 1 frame) - Enable this for exactly one frame during a scene cut to prevent severe ghosting artifacts from the previous scene.
- Normal:
Installation
Recommended: Install via ComfyUI-Manager.
Manual installation:
cd ComfyUI/custom_nodes/
git clone https://github.com/MSXYZ-GenAI/comfyui-msxyz.git