RTX Batch Upscaler — Image & Video

What's New in v1.1

Version 1.1 introduces more flexible image output options, a cleaner workflow layout, and a new custom node developed specifically for this project.

🆕 New Custom Node — Save Original Image (JPG / PNG)

We've developed and published our own ComfyUI custom node, Save Original Image (JPG / PNG).

This node allows you to:

• Save upscaled images in either JPG or PNG format.

• Preserve the original filenames of processed images.

• Adjust JPG quality (1–100).

• Adjust PNG compression level (0–9).

• Choose a custom output subfolder.

• Control how existing files are handled: Skip, Overwrite, or Error.

• Preview saved images directly inside ComfyUI.

The node is now available through ComfyUI Registry and ComfyUI Manager.

GitHub Repository:
https://github.com/aiwaifuhen-oss/ComfyUI-SaveOriginalImage

🖼️ JPG vs PNG — Choose Your Preferred Output

You can now choose the output format depending on your needs.

PNG — Lossless Quality

• Preserves image quality without lossy compression.

• Recommended for archiving, further editing, and maximum image fidelity.

• Usually produces larger files.

JPG — Smaller File Sizes

• Uses adjustable lossy compression to significantly reduce file sizes.

• Recommended for sharing, uploading, and saving storage space.

• Default quality is set to 95, offering an excellent balance between visual quality and file size.

Note: PNG compression level affects file size and processing time, not visual quality. JPG quality controls the trade-off between compression and image fidelity.

🎨 Improved Workflow Layout

The workflow has been visually reorganized to make it easier to understand and use.

• 01 — IMAGE INPUT: Dedicated section for batch image processing.

• 02 — VIDEO INPUT: Dedicated section for batch video processing.

• Clearer input selection controls.

• Improved node positioning and connections.

• Color-coded groups for easier navigation.

• Visible reminders for folder configuration and batch settings.

Both image and video processing continue to use NVIDIA RTX Super Resolution.

🔧 Improved Custom Node Installation

The workflow now includes the Registry metadata needed for ComfyUI Manager to identify our custom node.

If the node is missing, open ComfyUI Manager → Install Missing Custom Nodes and install Save Original Image (JPG / PNG).

Installation has been tested successfully with both the 1.1.1 and nightly options shown in Manager.

Description

This is a simple batch upscaling workflow for images and short videos using NVIDIA RTX Video Super Resolution in ComfyUI.

The main goal of this workflow was to create a practical way to process batches of videos without completely saturating system RAM, which was a problem I experienced with my previous upscaling workflow.

Instead of loading and processing an entire video at once, the video pipeline uses batch processing, allowing you to control the number of frames processed at a time. This makes RAM usage much easier to manage, especially when processing multiple videos in a queue.

The workflow supports two separate modes while sharing the same RTX Super Resolution processing stage:

• Image Mode — Batch loads images from a folder, upscales them, and saves the resulting images.

• Video Mode — Batch loads MP4 videos, processes their frames through RTX Video Super Resolution, preserves the original audio and frame rate, and combines the upscaled frames back into MP4 videos.

The included switch selects which source is sent to RTX Super Resolution:

1 = Image
2 = Video

The workflow is currently configured for 3× scaling with ULTRA quality, but you can easily change these settings directly in the RTX Video Super Resolution node.

Requirements

You will need:

• ComfyUI

• A compatible NVIDIA RTX GPU

• ComfyUI NVIDIA RTX Nodes — for RTX Video Super Resolution

• ComfyUI-VideoHelperSuite — for loading, batching and combining videos

• ComfyUI-FolderBatch — for folder-based image/video batch queues

• ComfyUI-Impact-Pack — for the Image/Video switch

• rgthree-comfy — for the Fast Groups Bypasser used to enable/disable the Image and Video sections

How to Use

First, set your input folder in the corresponding Image or Video section.

For Images

Enable the IMAGE INPUT group and disable/bypass the VIDEO INPUT group.

Set:

1 = Image

The workflow will load the images from the selected folder, process them through RTX Video Super Resolution, and save the upscaled images.

Supported image formats include:

PNG, JPG, JPEG, WEBP and BMP

For Videos

Enable the VIDEO INPUT group and disable/bypass the IMAGE INPUT group.

Set:

2 = Video

The workflow will load MP4 videos from the selected folder, process their frames in batches, upscale them using RTX Video Super Resolution, and combine the resulting frames back into video while preserving the original audio and frame rate.

The video queue is currently configured for:

MP4

RAM Usage / Frames Per Batch

One of the main reasons this workflow was created was to avoid excessive RAM usage when upscaling videos.

Video processing is divided into batches using the Meta Batch Manager.

The default setting is:

200 Frames Per Batch

You can lower or increase this value depending on your available system RAM.

As a reference, on my system with 64 GB of RAM, setting this to approximately 500 frames can consume almost all available RAM.

If you're unsure, I recommend starting with 200 and monitoring your RAM usage.

Lower values should reduce memory usage at the cost of potentially taking longer to complete the video.

⚠️ IMPORTANT — Refresh Before Running

Whenever you add or remove images/videos from the input folder, refresh the ComfyUI interface with F5 before starting the workflow.

Make sure the Video/Image Count displayed in the FolderBatch Queue matches the actual number of files inside the folder.

If the numbers don't match:

Press F5 before running the workflow.

Running the workflow with an incorrect file count may cause queue errors or cause batch processing to stop unexpectedly.

This is especially important when processing several files automatically.

Tested Use Case & Long Video Disclaimer

I primarily created and tested this workflow for short-form video upscaling.

My main use cases are:

• Short video clips

• 15-second videos generated with MiniMax

• Processing multiple short videos automatically as a batch

• Batch image upscaling

For these workloads, the workflow has worked very well for me and, most importantly, has helped keep RAM usage under control compared with my previous workflow.

Long Videos

I have not properly tested this workflow with long-form videos.

This includes things such as:

• Full anime/TV episodes

• Movies

• Long recordings

• Other extended video content

It may work, but I cannot guarantee the same performance, RAM usage or stability with long videos.

If you want to experiment with longer videos, I strongly recommend starting with a lower Frames Per Batch value and monitoring RAM and disk usage.

If anyone tests it successfully with long-form content, I would be very interested in hearing about your results.

RTX Super Resolution Settings

My included configuration uses:

Scale: 3×
Quality: ULTRA

These are simply my current settings and are not required.

Feel free to change the scale and quality depending on your source material, GPU, desired output resolution and processing time.

About This Workflow

I'm not a ComfyUI expert.

I originally built this workflow because I needed a better way to batch upscale my generated videos without having the process consume practically all of my system RAM.

The workflow was developed through experimentation and troubleshooting with the help of ChatGPT, and after getting the video batching system working reliably, I expanded it so the same workflow could also be used for batch image upscaling.

I'm sharing it because it has been useful for my own workflow and hopefully it can be useful to someone else too.

Improvements are welcome!

If you know ComfyUI better than I do and see a way to improve, optimize or simplify this workflow, please feel free to modify it.

Suggestions and improved versions are absolutely welcome.

In particular, I'd be interested in improvements related to:

• Better RAM management

• Long-form video processing

• More efficient batching

• Better Image/Video switching

• Automatic mode selection

• Performance optimizations

If you improve it, I'd love to hear what you changed so I can learn from it as well.