Updated Workflow
The updated workflow removed all the audio processing nodes to be more efficient. This speed things up and use less resources. I have also switched the video output node to use the native ComfyUI one (VHS node has an issue dealing with video with no audio). Another changes to the workflow is the addition of an optional subgraph that you can enable. This allows you to convert the frame rate of your video back to the original. There is also another optional subgraph to quickly stretch the video to a very high resolution.
Update: Version 2
The new lora and workflow are updated. The processed video will look more natural and will be free from ghosting artifacts. You need to use the new workflow with the lora.
LTX 2.5 - CQ Video and Image Enhancer LoRAs
This is a LTX 2.5 Lora with 2 versions, one for image and the other for video. The goal of these loras are to improve, enhance, and restore the quality of old, low-resolution, or poor-quality images and videos. Please be aware that this is a generative enhancement lora, it is not a regular upscaler. The result is better than many commercial models and also faster to process.
- Prompt
- V2 Enhanced Video Example
- Prompt
- V2 Enhanced Video Example
- Prompt
- Minimax H3 Enhanced
- Prompt
- Minimax H3 Enhanced
- Prompt
- Enhanced Image Example
- Prompt
- Enhanced Image Example
- Prompt
- Enhanced Video Example
- Prompt
- Enhanced Video Example
- Prompt
- Enhanced Video Example
The full V2 demo video can be view here:
https://www.youtube.com/watch?v=grME8VMMwAU
The full demo video can be view here:
https://www.youtube.com/watch?v=uVa66lSCI_U
Please use the included workflows to get started:
https://huggingface.co/CQdesign/LTX-2.5-CQ-Video-and-Image-Enhancer-LoRAs/tree/main/Workflow
Prompt is not needed for this lora. It is important that you use "ltx-2.5-video-vae-conv-bf16.safetensors" for VAE (especially for image generation). This convolution version produced a smoother and higher quality image. If you use the normal VAE, it will output an over-sharpened look that degrades the image.