ComfyUI SUPIR: New UpScaler for ComfyUI!
TLDRIn this video, the creator guides viewers through the installation of Comfy UI Super, a new upscaler by Kijai, and its integration with EV. The process involves cloning a repository, installing requirements, and configuring the UI. The creator shares their experience using the tool, noting the time-consuming nature of the upscaling process, especially with limited VRAM. They compare the results with other methods and highlight the significant improvement in processing time with a graphics card upgrade to 16GB VRAM.
Takeaways
- 🌟 Introduction of Comfy UI Super, an upscaler with a wrapper by Kijai.
- 📂 Installation of Comfy UI Super can be done via Comfy UI Manager or by cloning a repository.
- 📋 Requirements include running `pip install D requirements.txt` or using a portable version for Windows.
- 🚀 The process may be resource-intensive, as the user experienced slow performance with 8GB of VRAM.
- 🛠️ Necessary libraries include Transformers, Open AI CLIP, FS Spec, Cornea, and PyTorch Lightning.
- 🧠 Model requirements involve downloading specific models from Hugging Face and placing them in the Comfy UI folder.
- 📂 The user demonstrates organizing models by creating a 'super' folder inside the checkpoints directory.
- 🔗 Cloning the 'Comfy UI Superior' repository is part of the custom node setup process.
- 🖼️ The workflow includes a load image node, superior upscale node, preview image node, and image compare node.
- ⏱️ The user shares their experience with processing times, noting that upscaling took 1 hour and 34 minutes with 8GB VRAM.
- 💡 The user suggests that for those with less VRAM, alternative, faster methods might be more efficient.
Q & A
What is the main topic of the video?
-The main topic of the video is the installation and usage of Comfy UI Super, an upscaler tool wrapped by Kijai.
How can you install Comfy UI Super?
-You can install Comfy UI Super via the Comfy UI manager, by installing from a Git repository or cloning the repo to custom nodes.
What are the requirements for running Comfy UI Super?
-To run Comfy UI Super, you need to install the required packages listed in the 'requirements.txt' file and X-Forcers if you are using the Windows portable version.
What models are needed for Comfy UI Super to function properly?
-You need the Transformers version 4.28, Open AI CLIP, FS Spec, Cornea, and other libraries like OmegaConf and PyTorch Lightning.
Where can you download the necessary models for Comfy UI Super?
-You can download the necessary models from Hugging Face and save them in the 'models' folder within your Comfy UI directory.
How does the video creator describe their experience with Comfy UI Super?
-The video creator describes their experience as satisfactory but slow due to their 8GB VRAM, suggesting that it might be more efficient to use other methods for upscaling.
What was the time taken for the video creator to upscale an image using Comfy UI Super with 8GB VRAM?
-It took the video creator 1 hour and 34 minutes to upscale an image with 8GB VRAM.
How long did it take to upscale an MP4 file with Comfy UI Super?
-Upscaling an MP4 file with a 1.2 times scale took just under 7 hours.
What was the time taken to upscale with Comfy UI Super after the video creator got a new graphics card with 16GB VRAM?
-After getting a new graphics card with 16GB VRAM, the upscaling process took only 2 minutes and 44 seconds.
What is the video creator's recommendation for users with less than 8GB VRAM?
-The video creator suggests that users with less than 8GB VRAM might find the process too slow and should consider faster methods for upscaling.
What additional tool did the video creator install to enhance their workflow with Comfy UI Super?
-The video creator installed 'image comparer' by ig3 to compare the original and upscaled images within the workflow.
Outlines
🚀 Introduction to Comfy UI Super and Installation Process
The paragraph introduces the viewer to the Comfy UI Super, a new upscaler released by Kijai. It explains that Comfy UI Super can be installed via the Comfy UI manager from a Git repository or by cloning. The user is guided through the installation process, which includes running pip install on a requirements.txt file or using a portable version for Windows. The video also discusses the necessity of having certain software versions, such as Transformers 4.28, Open AI CLIP, FS Spec, Cornea, and others. The creator shares their experience with the software, noting that it runs slowly on their system with 8GB of VRAM and suggests downloading models from Hugging Face and placing them in the appropriate Comfy UI folder.
🎥 Upscaling Experience and Performance with Different VRAM
In this paragraph, the creator shares their experience of upscaling an MP4 file from a previous video. They discuss the time it took for the process with different VRAM capacities, highlighting that the process was significantly faster with a new graphics card featuring 16GB of VRAM. The creator compares the upscale results and processing times between 8GB and 16GB VRAM, emphasizing the substantial improvement in speed and suggesting that there might be faster methods for upscaling, especially for those with less VRAM.
Mindmap
Keywords
💡Comfy UI
💡Upscaler
💡Transformers
💡Hugging Face
💡VRAM
💡SDXL Base
💡Git
💡Pip
💡Image Compare Node
💡Upscale
💡CFG Scale
Highlights
Installing Comfy UI Super, a new upscaler by Kijai.
Comfy UI Super can install EV and run custom nodes from a Git repository.
The process involves running `pip install` for the `requirements.txt` file.
Portable versions can be run from a Windows folder.
Transformers, OpenAI CLIP, FS Spec, Cornea, and other libraries are required.
Models can be downloaded from Hugging Face and saved in the Comfy UI folder.
An SDXL base model is also needed for the upscaling process.
The video demonstrates cloning the Comfy UI Super repository and installing dependencies.
Config UI is used to set up the upscaling workflow with nodes.
An image compare node allows for visual comparison between the original and upscaled images.
The video creator's experience with upscaling an animation video.
Upscaling an image took 1 hour and 34 minutes with 8GB of VRAM.
Upscaling an MP4 file took under 7 hours, suggesting potential for longer processing times.
The creator's new graphics card with 16GB of VRAM significantly reduced processing times.
The video concludes with a comparison of processing times between different VRAM capacities.
Links for further information are provided in the video description.