Upscayl

What is Upscayl?

Upscayl is a free and open-source AI tool for enlarging and enhancing images; it can be used on Windows, macOS, and Linux. It utilizes the Real-ESRGAN neural network along with Vulkan acceleration to restore pixel details, and is suitable for enlarging photos, illustrations, animations, low-resolution images, as well as AI-generated images.

The project also offers Upscayl Cloud as a paid service; it allows users to take advantage of cloud computing power through a browser, without the need for a compatible graphics card on the local device. The desktop version emphasizes local use, free access, and scalability, while the Cloud version provides additional features such as higher resolution outputs, cloud storage, as well as image creation and editing tools.

Upscayl’s main functions

• AI image upscaling: Utilizing pre-trained models to increase resolution and reconstruct textures;
• Multiple model selection: Different models are chosen for general photos, digital art, images with clear details, and smooth visuals;
• Double Upscayl: Perform two amplifications in succession to achieve a higher magnification level;
• Batch mode: Process multiple images in an entire folder at once;
• Custom scaling and width: Generate results suitable for publication or printing based on the desired dimensions;
• Output control: Select formats such as PNG, JPG, WebP, etc., as well as the compression quality;
• Side-by-side preview: View details, sharpness, and artifacts by comparing them side by side;
• Custom models: Import compatible NCNN models to handle specific styles;
• Local processing: The desktop version does not require uploading images to third-party servers;
• Cloud enhancements: It offers face enhancement, larger output sizes, image generation, natural language editing, and cross-device access.

Differences between the desktop version and Upscale Cloud

The desktop version is suitable for users who have a GPU compatible with Vulkan and who value local privacy; there is no limit on the number of processing tasks that can be carried out. The cloud version is intended for older computers, mobile devices, or teams that need larger images, cloud-based history storage, and commercial-level features.

How to choose an AI model

Different models produce varying levels of sharpness, texture, and noise reduction; there is no single model that works well for all images. The most reliable approach is to extract elements such as faces, text, hair, leaves, and gradient areas, and then use multiple models to create samples for comparison.

Universal photo model

General-purpose photo models are suitable for portraits, landscapes, and products, and they enable a balance between clarity and natural textures. If low-quality JPEG images are over-sharpened, it can amplify compression artifacts, pores, or noise.

Digital art and anime models

Models in the field of digital art place more emphasis on lines, color blocks, and edges; they are suitable for use in illustrations, anime screenshots, icons, and game assets. When used with real photographs, they may give a plastic appearance or an overly smooth texture.

High-fidelity and sharpening models

High-fidelity or sharpening models can enhance local details, but they also tend to create textures that are not present in the original image. Archives, evidence, as well as medical and scientific images cannot consider the details added by AI as genuine information.

Custom models

Upscayl allows the import of compatible NCNN models, and the community also maintains an additional set of models. Before importing, it is necessary to check the model’s license, scaling factor, file pairing, and the appropriate images; third-party models are not guaranteed to be of quality or safety by the main project.

System requirements and compatibility

The core requirement for the desktop version is a Vulkan-compatible GPU that functions properly; many standalone NVIDIA and AMD GPUs can be used, and some integrated GPUs may also work, though the manufacturer does not provide any guarantees in this regard. Drivers, video memory, operating system, image size, and models all affect speed and stability.

If the GPU cannot be recognized, if the screen goes black, or if processing fails, you can first update the graphics card drivers, and then use the Vulkan detection tool to check for compatibility. On computers with multiple graphics cards, it is possible to select the specific GPU ID in the settings; for very large images, it is advisable to reduce the scaling factor or adjust the size of the image segments.

Upscayl Cloud prices

The desktop version can always be downloaded and used for free, while the Cloud version operates on a points system. The options displayed on the official website currently are those corresponding to 300 points per month; other quota options, one-time purchases, annual discounts, and regional taxes shall be as indicated on the payment page.

The official website allows users to choose between monthly credit amounts of 100, 300, 500, 1000, 2500, and 5000, and it shows a discount of up to 40% on an annual basis. Once the subscription is canceled, the accumulated credits are not carried over, and the cloud usage history may be lost after one month.

Integration and storage

The cost of Cloud tasks is calculated based on the processing parameters and the files involved; the interface returns the estimated amount of credits that will be deducted. Pro offers 6 months of storage, and the privacy policy states that the generated results are typically retained for 180 days or until the user deletes them. The actual deletion process may still be affected by backups and system caching.

Upscayl usage guide

Zoom in on a local image

1. Download the version appropriate for your system from the official website or the GitHub release page;
2. After installation, verify that the graphics driver and Vulkan support are working properly;
3. Open Upscale and select an image for which you have usage rights;
4. Select a model based on the type of photo or illustration;
5. Set to 2 times, 4 times, or the desired width and output format;
6. Select the output folder and start the processing;
7. Examine faces, text, edges, and noise by comparing views;
8. When unsatisfied, switch to a different model instead of using the same one repeatedly;
9. Save the best result while keeping the original image.

Scale up a set of materials in bulk

1. Copy the original image to a separate input folder to avoid overwriting unique files;
2. Group the materials according to video quality, illustration quality, and compression quality;
3. Representative images are selected from each group to test the model and the magnification level;
4. Estimate the size of the pixels after scaling, as well as the file size and disk space required;
5. Open Batch Upscale and select the input and output directories;
6. Set unified models, formats, compression, and scaling factors;
7. Start the task and monitor the GPU temperature, video memory usage, and error logs;
8. Randomly inspect blur, faces, thin lines, transparent edges, and text;
9. Use other models to process abnormal images separately;
10. Delivery or release is carried out only after confirming the quantity and quality.

Use Upscale Cloud for enlargement.

1. Create an account and choose a free trial or an appropriate point plan;
2. Remove sensitive information from the file and confirm the upload authorization;
3. Upload JPG, PNG, or WebP images in the browser;
4. Select the model, magnification level, output format, and face enhancement options;
5. View the points estimated to be required for the task;
6. Submit for processing and wait for it to be completed in the cloud;
7. Download the result and check the newly added textures and color offsets;
8. Delete unnecessary inputs and outputs in a timely manner;
9. Check your points, storage history, and renewal status regularly.

Zooming in on an image does not restore true details.

AI enhances images by predicting new pixels based on the image patterns it has learned; this makes the images clearer, but it is not able to recover information that was never recorded in the original image. In cases of severe defocus, motion blur, very small faces, or unreadable text, the model may generate plausible yet incorrect details.

When dealing with judicial evidence, medical diagnoses, scientific measurements, historical archives, and identity verification, the original images should be retained, with a clear indication that AI enhancement has been applied. The enhanced versions should not be used in place of the original evidence, nor should they be used to draw definitive conclusions regarding key facts.

Which users are it suitable for

• Photography enthusiasts: Enlarging old photos, landscapes, and portraits for display;
• Illustrators and anime enthusiasts: enhanced line drawings, color blocks, and low-resolution screenshots;
• Content creators: Adapt small-sized materials for use as covers, in videos, and on social media platforms;
• E-commerce team: Increase the size of product images and handle them in batches in a uniform manner;
• Game and visual designers: enlarging textures, concept art, and pixel-based materials;
• Archivist: Create enhanced copies that are easy to view while retaining the original documents;
• Users who are sensitive to privacy: use the desktop version to process their personal photos;
• Users without compatible graphics cards: Can make use of cloud computing power via the Cloud service.

The advantages of Upscayl

• The desktop version is free and open-source, with no charging based on the number of images nor any mandatory watermarks.
• Images used locally do not need to be uploaded, providing more direct control over privacy.
• It supports Windows, macOS, and various Linux installation formats;
• It includes multiple built-in models, as well as support for compatible custom models;
• Single image, batch processing, Double Upscayl, and target size setting cover the common tasks;
• Project codes, issue tracking, and release records are transparent and accessible to everyone;
• Cloud provides an alternative for users who lack a GPU or need a solution for mobile use.
• Cloud also offers face enhancement, generation, and natural language editing.

Usage restrictions and precautions

• The desktop version typically requires a GPU that is compatible with Vulkan, and many older graphics cards and integrated GPUs cannot function properly.
• Large images and Double Upscayl significantly increase video memory usage, processing time, file size, and disk space consumption;
• Different models may over-sharpen, blur textures, or create false details;
• Enlarging low-resolution text does not guarantee the restoration of the original content;
• The compatibility, quality, and licensing of custom models must be verified by the user themselves;
• Before batch processing, it is necessary to keep the original images and test a small number of samples;
• Uploading files to the cloud incurs costs related to subscriptions, points, and storage management.
• Cloud prices, point consumption, and features may still change;
• Open-source licenses permit the use of software, but modifications, distribution, and deployment over the network must comply with the AGPL requirements.

Privacy, Commercial Use, and Data Retention

Inference is carried out using the desktop version; normal scaling does not require uploading images, making it suitable for processing private photos and materials that are not intended for public use. Users should still download it from official sources and check whether logs, updates, and custom models might introduce any additional network activity.

The Cloud terms state that users retain ownership of the content they upload, and that the images are not used for training AI systems. The privacy policy specifies that uploaded files are retained for a maximum of about 1 day, while the generated results are kept for around 180 days or until they are deleted.

The service makes use of providers such as Firebase, AWS, and PostHog for authentication, storage, and analysis.

Cloud explicitly supports commercial use, and desktop software can also be used for commercial purposes under an open-source license; however, images, models, and outputs are still subject to their respective licensing terms. When distributing modified versions or using them as web services, it is necessary to have legal experts review the requirements regarding the provision of the AGPL-3.0 source code.

API, GitHub, and open-source information

The Upscayl desktop project has its source code made available on GitHub; the current README indicates version 2.15, and it is licensed under the GNU AGPL-3.0 license. The project also makes available a set of custom models, documentation, a issue tracking system, build instructions, as well as release packages for various platforms.

Cloud Document provides interfaces for task submission, segmented uploading, and tracking task status, and requires an API Key to process protected requests. The public documentation related to these APIs does not mean that unlimited calls can be made free of charge; before integrating them into production environments, it is necessary to verify account permissions, credits, authentication requirements, and the service terms.

Frequently Asked Questions

Is Upscale completely free?

The desktop versions for Windows, macOS, and Linux are free and open-source, allowing unlimited processing on the local device. Upscale Cloud is a separate service based on a points system; a trial of 10 points is available.

What graphics card is needed for Upscayl?

The desktop version typically requires a GPU that supports Vulkan; many standalone NVIDIA and AMD graphics cards can be used, and some integrated graphics cards may also work, though this is not guaranteed. If no compatible GPU is available, Cloud can be an option.

How much can Upscale enlarge an image?

The common model uses a 4x magnification level, but other magnification factors or target widths can also be set; Double Upscayl enables continuous processing to achieve higher magnifications.

The ultimate limit is determined by the GPU, memory, disk space, and image size.

Will Upscale upload my pictures?

When using the desktop version, images are not sent to the Cloud; to make use of the Cloud, it is necessary to upload them, and the input and output data are handled in accordance with its privacy and storage policies.

Can Upscale be used for commercial purposes?

It is possible, but it is necessary to comply at the same time with the AGPL license for the software, the licenses applicable to the models used, and the rights related to the images. Cloud also supports commercial use; nevertheless, companies should check the contract terms and data requirements.