"Shout-out to the creator of this workflow - shikasensei (https://civitai.com/articles/3582). While it's been helpful"

I've encountered some performance issues on my mid-end hardware. A 1024 x 786 image takes over 10 minutes to process on a T4 GPU.

Here are my modifications to address these challenges:

• Separate VAE model: This has increased tile processing time.
• Switched to DPM++ 2m Karras: Provides broader model compatibility.
• Reduced image upscaling from 2x to 1.5x: Improves performance.
• Added Image Compare (by rgthree): Allows for visual quality assessment.
• Saving image node: Preserves the output.

with this modification, time for the process has toke less than  2 min on T4 GPU

Key Challenge: The workflow currently decodes and encodes the image three times, leading to some data loss. My goal is to refine the process to avoid this and achieve better results on mid-range hardware. I'm working on optimizations for this purpose.