Portable Topaz Gigapixel AI 7.1.1 (x64) + All Models
Beautiful photo enlargements using machine learning. Enlarging your image without losing detail has always been impossible… until now. Upscale your photos by up to 600% while perfectly preserving image quality.
Gigapixel AI Portable helps you do a single thing very well: increase image resolution in a natural way. This is pretty hard to do. Most upscaling algorithms don’t know how to accurately fill in missing pixels, so you end up with blocky low-quality results.
To solve this problem, Gigapixel AI’s neural networks are continuously trained on a wide variety of sample images to understand how photorealistic detail looks. After learning from millions of images, it can now believably increase resolution on a wide variety of different subjects (portraits, wildlife, landscapes, architecture, etc) and image types (DSLR, web/compressed, CG). The results are often magical.
Improve image quality.
Gigapixel AI can make your image appear like it was captured with a much higher-quality camera. It’s been trained on enough images of different types of subjects (like portraits, landscapes, architecture, and more) to know what photo detail should look like. This allows you to increase your photo’s resolution and quality very naturally with minimal artifacts.
Pixel count isn’t the only thing that affects image quality. Normally, if you start out with a low-quality image, you’ll just end up with a bigger low-quality image if you upscale it. Gigapixel AI’s technology can instead create very believable and natural image detail that can often look amazing.
Add image detail that looks completely natural.
Normally enlarging an originally low-quality image will cause fuzzy square artifacts that visually degrade the image. Because Gigapixel AI Portable was trained on an extensive dataset of mountain textures (among other subjects), it’s able to enlarge the image while simultaneously increasing the resolution. You almost can’t tell it was upscaled.
Enhance the texture of feathers
Feathers are often difficult to enhance correctly; they need to be both sharp and soft in different places to look natural. While this is difficult for normal enlargement methods, Gigapixel AI understands how to increase feather detail in a believable way.
Improve skin texture and sharpness
Portraits are another tricky subject – especially skin texture – because we are all very sensitive to “unnatural” skin or faces. When upscaling portraits, Gigapixel AI adds believable skin detail and texture for the kinds of results you see above.
Higher-resolution landscapes
Landscapes often look unnatural without fine detail. Gigapixel AI can add believable texture from existing detail and dramatically increase the image quality of your landscapes.
More detailed cityscapes
Man-made architecture often have more straight and clearly-defined lines than nature. Gigapixel’s special “Lines” AI model was trained specifically on architecture and objects for better results.
Crystal-clear upscaling with no blocky artifacts
Normal upscaling techniques will create fuzzy and blocky artifacts that you can see above, but Gigapixel AI Portable resolves edges sharply with few artifacts for much more natural results. (Pay special attention above to the fine details around the windows, trees, and mountain edges.)
Fix extreme pixellation in low-resolution images
While Gigapixel AI Portable works best with input images that start with moderate or high quality, it can also work wonders on extremely low-quality and compressed images. In addition to increasing resolution, the “Very Compressed” model will also automatically remove JPEG compression artifacts, apply noise reduction, and sharpen the output in a natural way.
Create better prints
Large prints display every small bit of detail in your photo, so it’s especially important that you start with a high-quality image. Gigapixel AI helps you achieve the extra image resolution to make your print really pop. In fact, Gigapixel AI Portable works even better when starting out with an already high-quality image, so you can get especially good results here.
Enlarge Images by up to 600%
Create a beautiful 80 x 60 inch print from a simple iPhone photo. Or transform your DSLR photos into sharp and ultra-detailed 100MP images.
Use truly intelligent upscaling
After being trained with millions of sample images, Gigapixel can now enlarge your images with a surprising level of quality.
Get crisp and clear results.
Gigapixel performs around 2 million operations per pixel to get you results with as much detail and sharpness as possible.
No more fuzzy enlargements.
Breathtakingly sharp prints. High-resolution cropping in post-production. iPhone photos that look like they came out of a DSLR. No matter why you’re enlarging your photos, you always want the best possible quality for your results. AI Gigapixel is currently the only photo enlargement product available that can actually add detail back to your upscaled photo.
Recover real details for actual sharpness
Increasing edge contrast is the simplest way to make an image look “sharp”, and how most other upscaling software works. Gigapixel is different: it increases the actual sharpness of your upscaled photo by recovering real detail in your image. This is only possible through training and exposure to millions of test images.
Crisp photos that look like they were never upscaled
It’s hard to believe that the crisp output image above was enlarged 4x, but it was. Gigapixel AI successfully determines which image features are stars (detail) and what’s noise. Then, it naturally accentuates and recovers detail while suppressing noise in your upscaled photo.
The highest-quality image upscaling that you can get anywhere
Because of these exceptional results, Gigapixel AI is used by everyone from leading photographers, to video game modders, to major motion studios. If you ever need to upscale your photo while preserving beautiful detail, take advantage of the most modern technology in the space and give Gigapixel AI a try.
Image enlargement: interpolation vs machine learning
Interpolation-based photo enlargement software has been around for decades. Gigapixel’s new approach based on machine learning is a giant leap ahead: state-of-the-art “deep learning” methods allow it to enlarge images while actually adding new details.
Interpolation
Interpolation algorithms work by approximating new pixels from the colors of surrounding ones. This does not add any detail to the image, it just smoothes out the transition between pixels.
Neural Network (Gigapixel AI)
A neural network analyzes thousands of photo pairs to learn how details usually get lost. The algorithm learns to “fill in” information in new images based on what it has learned… effectively adding new detail to your photo.:
What’s NEW:
I’m excited to announce the new major update for Gigapixel! This update features a massive amount of changes under the hood, including an updated AI engine which should be faster and compatible with newer GPUs compared to 6.3.3, a UI refresh across the entire app, and updates to the Standard and High Fidelity models.
For Standard and High Fidelity v2 are have changed the following:
- For Standard we’ve improved the deblur consistency (blurry patch issue)
- For High Fidelity we’ve improved the grain preserving consistency for low denoise/remove jpeg sliders
- For use cases requiring keeping original noise/grain, select this model and tune down the two sliders.
- For High Fidelity we’ve fixed the tile blending artifacts
- For both models we’ve fixed the tile artifacts on Apple Silicon Mac’s Neural Engine
- In Photo we have an option to turn off Neural Engine but it will slow down processing
- For quality in general, the models should be more generative
Another big change on the UI involves how images are selected and how controls are shown for those selected images. You can now use shortcuts like Ctrl/Cmd + A to select all and Ctrl/Cmd + Shift + A to deselect all images in the file list. When multiple images are selected, the controls in the right panel will now show a mixed state if the values between the images differ. For instance, if you select 2 images where one has a slider value of 50 and the other has a slider value of 75, the slider will show a blank box and a new “mixed state” slider design to indicate that the current selections have different values. You can then move the slider to apply a new identical value to both of the selected images if you want.
Important notes:
- All preferences will be reset when updating to this version! Make sure to re-check your preferences (and check out the new ones too).
- It is expected that you will need to re-authenticate with the updated version on first launch.
- On first run models that were updated will re-download automatically. On subsequent runs if we detect invalid models it should auto re-download them.
There really are too many changes to get into the weeds with, so instead of an exhaustive change log I’ve pulled out a few points.
Major changes:
- Updated AI engine which should allow better compatibility and faster processing
- Added Standard and High Fidelity v2 models
- These models will eventually replace v1 models in time. If you have quality issues, please let us know.
- Refreshed the UI with new design and controls
- Updated preferences with new options
- Note: This will fully reset your preferences
Notable additions:
- Added panning and pinch to zoom gestures in preview
- Added updated RAW preview handling (should be closer to Photo AI)
- Added ability to capture before/after in split and side by side views
- Added ability to move the file list up and down via dragging
- Note: There is a known issue where removing/adding images causes it to snap back to the default position
- Changed logic for selections
- Changed controls to show mixed states if multiple images with different values are selected
- New “Recovery” Model (Beta): Recovery is a diffusion-based generative AI upscaling model, which produces different results than our other upscaling models. While this model is slower than the other upscaling models, the model performs very well when upscaling low-quality source images, for example images with low resolution or significant blur.The model also seems to perform well on faces, buildings/structures and other objects. The detail slider allows you to control the level of detail in the image after the initial render in post-processing. We have found that detail settings in the 50-75 range work well for most scenarios.
The Recovery model also allows you to preview sections of an image by rendering small tiles before running a larger recovery, as some of those renders can be lengthy. You can also still render the entire image directly from the export panel if you desire. Note that the recovery preview will be a different generated image than the final result, so it will not blend. However, it should be quite representative of the quality of the final output.
Please Note: We have plans to release significantly more performant and smaller versions over time – but you will need a powerful graphics card for this version of the model. We recommend upscaling images under 1000px by 1000px depending on the size of your graphics card. This model currently requires 8GB+ of VRAM. Renders with the Recovery model may take several minutes even on relatively small images at the moment.
What’s New in V7:
- New Low Resolution v2 Model (Beta): The Low Resolution AI upscaling model has been upgraded into the 2nd generation family of models, performing well out of the box for both low resolution and high compression images. You will still have access to the Gen1 model as well, using the drop-down menu on the top right of the model selection panel
- New “Text & Shapes” Model: We have added a new AI-upscaling model that enables more distinct patterns and shapes during upscaling with man-made objects or defined shapes such as cars or buildings, and even written text and fonts.
- Comparison View: The beloved comparison view makes its return debut with its notable four quad design. We’ve cleaned up how the interface is used, showing exactly which image you’re working on. You can also now configure default settings for which AI models are used in the view and remember those settings for future use. We will continue to improve and support this view over time, and we have some more exciting updates in store for this one.
- Crop Rotation: We have extended the Crop tool within Gigapixel, adding new capabilities into your workflow. Images can now be rotated to create perfect vertical or horizontal lines.
- Face Selection: In previous versions of Gigapixel, face recovery sometimes recognizes faces incorrectly. We have now added Face Selection, which will allow you to see the number of faces detected and manually choose which faces should be recovered.
- UI Changes: New colors and icons across all Topaz Labs apps for a unified, streamlined look. Photo AI, Gigapixel, and Video AI are all receiving a UI refresh to make them feel more cohesive, as many of our users own multiple different Topaz Labs applications.
Latest Changelog:
- Fixed visual issue where toggles would look on when they were off
- Fixed wrong link being used for help article
- Fixed uncommon permissions issue on Mac
- Fixed being able to pull up file list while in comparison view
- Fixed recovery warning always saying “small”
- Fixed comparison view incorrectly keeping auto model flag enabled
- Fixed various button hover states, spacing, and icon sizes
- Fixed auto settings not re-applying when switching between models
Note: Microsoft Visual C++ Runtime 2015-2022 Required (if not already installed on your system)
