Gigapixel AI is the fourth of our core apps to gain native Apple M1 support, joining DeNoise AI, Sharpen AI, and Video Enhance AI. Building Gigapixel AI to natively support Apple’s integrated “System on a Chip” architecture allows us to provide Apple M1, M1 Pro, M1 Max, and M1 Ultra users with even faster performance. Everything from app load times, single image editing, and batch processing will be noticeably faster.
In our first benchmark test, I compared the time it took to load Gigapixel AI v6.0 from a cold boot using both the native M1 and Rosetta versions. As you can see, the native M1 build provides a 47% faster load time when compared to the Rosetta emulation version.
Next, I compared processing speeds for a resource-intensive task such as batch processing. I opened a mix of 12 JPEG files of varying resolutions and configured a batch processing job that would use the Standard model at a 6x (600%) upscale factor, auto settings, and had both the Reduce Color Bleed and Face Refinement options enabled.
When comparing the batch job performance between the native M1 build and the Rosetta emulation build, the processing time is almost 44% faster. Even more impressively, when comparing the native M1 build to a comparable Macbook Pro with the Intel chipset, the batch process job was almost 55% faster.
We’ve improved Gigapixel AI’s model download method in an effort to streamline your experience while using the app while also reducing the installer file size. In previous versions, model downloads would kick off only when a user selected the particular model and upscale factor. If the model was not already stored locally, a manual download would begin.
Now, when you first launch Gigapixel AI v6.0, you will be presented with a new dialog boxes asking if you prefer to batch download all of the models for your computer configuration. This streamlined approach makes it faster and less obtrusive to switch between models and upscale factors.
One important point to make is that you won’t be able to process images offline until you open Gigapixel AI v6.0 and download at least some models. Allowing the batch download of all corresponding models to complete will be your best bet and should make using the app smoother, especially while offline.
We’ve also made significant quality improvements to the Very Compressed model. In prior versions, using this model could have resulted in a visible pixel shift as illustrated below:
In Gigapixel AI v6.0, we improved the quality of the Very Compressed model to eliminate virtually any pixel shift from occurring.
Along with the recent updates to DeNoise AI and Sharpen AI, you will now benefit from a refresh to Gigapixel AI’s user interface. The improved colors and increased contrast will make using Gigapixel AI clearer and easier. We’ve also added optimized 16-bit backends to the Standard and face refinement models, which should lead to faster processing for some Intel CPU and Windows GPU users. To round things off, we also fixed a number of annoying usability bugs and improved memory handling during batch processing.