the video quality is reduce after convert .

[endofD]

Any idea how to improve this
I do some test
1.change inference_steps to 35
2. guidance_scale to 1.5 from1

it doesn't help

look like a sd 1.5 vae encode and decode loss .

screenshot .

[theblackbeansoup]

I'm having the same issue. Here is what deepmind suggested, but I haven't tested:

The issue of quality degradation in LatentSync when rendering videos in ComfyUI is a common challenge, but there are several strategies you can employ to overcome this problem. Below are detailed solutions based on the search results and best practices:

1. Optimize Video Input Settings
   Frame Rate and Resolution: Ensure your input video is at 25 FPS and has a high resolution (e.g., 720p or 1080p). LatentSync works best with clear, frontal face videos, and lower frame rates or resolutions can lead to fuzzy outputs 23.

Face Visibility: Make sure the face is clearly visible throughout the video. Obstructions like hair, hands, or poor lighting can degrade the quality of the lip-sync output 3.

2. Use High-Resolution Output Nodes
   ComfyUI-LatentSync-Node: This advanced node builds upon the basic LatentSync implementation and supports high-resolution output. It ensures smoother and clearer lip-sync results by leveraging ByteDance's LatentSync model with enhanced processing capabilities 5.

Ultimate SD Upscale: If you're working with low-resolution inputs, use the Ultimate SD Upscale node in ComfyUI. This node allows you to upscale images and videos while preserving details, reducing fuzziness in the final output 8.

3. Adjust Denoising and Sampling Parameters
   Denoising Strength: Lower the denoising strength in your workflow to avoid over-processing, which can introduce artifacts and fuzziness. A value between 0.3 and 0.5 is often optimal for balancing detail preservation and noise reduction 14.

Sampling Steps: Increase the number of sampling steps to improve the quality of the generated frames. However, this will increase rendering time, so find a balance based on your hardware capabilities 14.

4. Leverage Temporal Consistency Tools
   TREPA Module: LatentSync's Temporal Representation Alignment (TREPA) module is designed to enhance temporal consistency in lip-sync animations. Ensure this module is active in your workflow to maintain smooth transitions between frames and reduce fuzziness 2.

Video Length Adjuster: Use the Video Length Adjuster node to synchronize video and audio lengths accurately. This prevents frame loss or stretching, which can degrade quality 3.

5. Post-Processing Enhancements
   Upscaling with ControlNet: After rendering, apply ControlNet-assisted upscaling to enhance the resolution and clarity of your video. This method preserves the original style and details while reducing fuzziness 14.

Frame Interpolation: Use tools like DAIN or Flowframes to interpolate additional frames between existing ones. This can smooth out transitions and improve the overall quality of the video 14.

6. Hardware and Software Optimization
   GPU Requirements: Ensure your GPU has at least 8GB of VRAM for optimal performance. If you're using a lower-end GPU, consider reducing the resolution or using cloud-based solutions like Runpod for rendering 513.

FFmpeg Settings: Use FFmpeg to fine-tune the final output. Adjust parameters like bitrate and codec to ensure high-quality video compression without significant loss of detail 3.

7. Test and Iterate
   Workflow Adjustments: Experiment with different workflows and settings in ComfyUI. For example, try combining LatentSync with Ultimate SD Upscale or ControlNet to see which configuration yields the best results 814.

Community Feedback: Check forums like Civitai or GitHub for user-tested workflows and tips. Many users share their optimized settings and workflows for LatentSync 513.

Summary
To overcome the fuzziness issue in LatentSync:

Optimize input video settings (25 FPS, high resolution, clear face visibility).

Use high-resolution output nodes like ComfyUI-LatentSync-Node or Ultimate SD Upscale.

Adjust denoising and sampling parameters for better detail preservation.

Leverage TREPA and Video Length Adjuster for temporal consistency.

Apply post-processing techniques like ControlNet upscaling and frame interpolation.

Ensure your hardware meets the requirements and fine-tune FFmpeg settings.

Test and iterate with community-shared workflows.

By following these steps, you can significantly improve the quality of your LatentSync outputs in ComfyUI.