Enlarge images in the browser, select restoration modes and preview results. Developers can submit 2× or 4× jobs through an API.
Used this tool? Rate it
Used this tool? Rate it
Upscayl is a browser-based image upscaler and photo-enhancement service operated by PixelForge LLC. Its workspace accepts a picture or image URL, lets the user choose an enhancement mode and enlargement setting, and offers a preview and download. It is part of a broader collection that includes prompt-based image editing, restoration and background removal. The service also describes an API for applications that need to submit upscaling jobs without using the interactive screen.
PixelForge identifies its service with upscayl.app; a similarly named project at upscayl.org describes an open-source desktop application for Linux, MacOS and Windows. The desktop project's features and licence should not be assumed to describe PixelForge's site. Upscayl's terms say it is independent of the developers or owners of its listed third-party models unless expressly stated otherwise. That disclaimer concerns the listed models; it does not establish the legal relationship between the similarly named projects.
The workspace displays Standard, Portrait, Anime / Art, Face Restore and Scratch Restore. The operator assigns different models to the modes, and the interface shows different credit amounts. A mode does not guarantee exact reconstruction of a missing detail.
The operator says models and routing may change; its description is not an independent test of quality.
The uploader lists JPG, PNG, WEBP and ZIP up to 20.00 MB each, a maximum of five images, file or URL input and examples. It displays 1×, 2×, 4×, 6× and 8×, with JPG, PNG and WEBP output. These are visible choices, not a guarantee that every account can export each one free.
The batch page describes testing five images free and paid batches up to ten, with 6K and 8K as paid options. Pack descriptions separately state larger batch limits. No checkout or paid processing was performed for this description.
A separate AI photo editing workspace uses written prompts to replace backgrounds, remove distractions or refine portraits. Unlike simple enlargement, it changes image content. The operator identifies Nano Banana and GPT Image among models used for generation and editing.
The editor page says it accepts up to eight JPG, PNG or WEBP reference images of at most 10 MB each and that free edits are resized before upload. This is a different input boundary from the homepage upscaler's five-file, 20 MB display. A captured editor configuration shows aspect-ratio choices, an output count from one to four and a total of four credits. The displayed total is a configuration at the time of capture, not a price for every model or combination of settings.
A linked watermark-removal workspace asks the user to brush over an area. Its availability does not establish permission to alter a protected image; the terms require image rights and prohibit infringement.
Choose the file or URL, select a mode and scale, run the job and inspect its preview before saving. The uploader displays one credit for Standard and Portrait, three for Face Restore, six for Anime / Art and fifteen for Scratch Restore. The research did not execute an authenticated job or verify output for a specific image.
The documented API is asynchronous: a server submits a public HTTPS image URL, receives a task identifier and retrieves status and billing state through a later request. An accepted response means the task was received, not that a finished image is already available. Its documented standard scales are 2× and 4×, narrower than the interactive selector.
Access is activated after a first successful credit purchase. The documentation says the secret key is shown once when created and belongs in server-side authorization, rather than client-side browser or mobile code. API keys can be created separately and revoked independently. The image URL must end in a supported JPG, JPEG, PNG or WEBP extension; the documented output format is PNG.
A request includes an idempotency key. The API guide says a retry with the same key and identical parameters returns the original task without another charge; reusing that key with different parameters returns a 409 conflict. The guide describes polling for task status, starting after three seconds and gradually increasing the interval up to ten seconds. Customer webhooks are not currently available, so the documented integration cannot rely on a callback when a job finishes.
For successful standard upscaling, the API lists 0.7 credit per image. The operator says it reserves credits when accepting a task, charges on success and returns them after a confirmed failure. This API price is a separate documented rate from the interactive mode labels and the pack table. A result URL is a delivery link: the API contract does not currently promise a minimum public URL retention period.
The batch page identifies e-commerce images as one use case: increasing image dimensions and making product textures, labels and finishes easier to inspect. A seller can use the workspace for a small set of source photographs or consider the server-to-server API for an existing catalog pipeline. Those paths have different upload, scale and billing limits; the existence of both does not mean that a browser batch purchase overrides API task limits.
For a different background rather than more pixels, the editor's examples include requesting a studio setting while preserving a product's label, proportions and material. This is an intended workflow, not evidence that each output preserves its label.
The provider describes an anime and illustration mode for line-based artwork and a face-restoration option for degraded faces. The batch page also identifies generated artwork prepared for display, print or further editing as a use case. These descriptions help distinguish the selectable modes, but they are not measured comparisons of how well each model preserves a particular illustration or person.
Scratch Restore is presented for photo restoration of damaged images and uses a different model. Reconstructed detail is a generated result, not an independently verified record of the original scene; the preview does not establish accuracy.
The batch page also names documentation screenshots and client-facing assets; their text and visual accuracy have not been tested here.
The upscaler runs through a web interface, with third-party processing where needed. Browser, API and MCP access do not expose identical controls or prices.
The API documentation lists JPEG, PNG and WebP input files up to 20 MB. It also gives scale-dependent pixel limits: up to 1.5 megapixels for 2×, or 0.8 megapixels with face enhancement; up to 0.75 megapixels for 4×, or 0.5 megapixels with face enhancement. These limits apply to the API as documented and should not be silently transferred to a browser upload or to the pack-table resolution figures.
The API lists 60 create requests per minute, 20 active tasks and 2,500 tasks per UTC day; its recommended client concurrency is five to ten tasks. These are published limits, not measured throughput.
For compatible AI clients, the remote MCP service exposes four operations: list products, check the credit balance, start a 2× or 4× upscale from a public HTTPS URL, and retrieve a task. The provider says MCP uses the same paid account, API key and billing ledger as the REST API. The MCP interface does not imply that other listed image-editing features are already available through those four operations.
The captured pricing page displayed four one-time credit packs. The amounts below are the figures displayed on that page on November 12, 2026; the page shows a dollar symbol without identifying a currency code in the captured price display. The pricing page also has monthly and yearly controls, but the current capture showed the credit-pack state, so subscription amounts are not inferred from another display.
| One-time pack | Displayed price | Displayed credits | Batch / upload figure |
|---|---|---|---|
| Mini | $10 | 200 | Up to 10 images; 1000 × 1000px uploads |
| Creator | $30 | 800 | Up to 50 images; 4000 × 4000px uploads |
| Pro Pack | $50 | 2000 | Up to 50 images; 4000 × 4000px uploads |
| Studio Pack | $100 | 5000 | Up to 100 images; 4000 × 4000px uploads |
The pack comparison says credits never expire. The cards advertise all enhancement modes, watermark-free downloads and editing, with priority support for Mini and Creator and dedicated support for Pro Pack and Studio Pack. Checkout, taxes and account entitlements were not tested.
The table calls the unit one credit per image for each listed pack, while the uploader displays one credit for Standard and Portrait, three for Face Restore, six for Anime / Art and fifteen for Scratch Restore. The two interfaces therefore do not support treating every listed conversion count as a confirmed count for every mode. The editor's captured four-credit total is another feature-specific display; the API has its own published 0.7-credit successful-job rate.
The refund policy covers subscriptions, renewals and one-time packs. Requests can be made at any time and are calculated proportionally from unused purchase value; used credits or paid actions are deducted. A refund may go to the original method, subject to payment-provider rules and law.
The policy says subscriptions renew automatically until cancelled, and describes requesting a proportional refund of unused current-period value. A confirmed service or provider error may cause an automatic credit return, separate from a cash refund.
A similarly named desktop product and a separate browser-based image service illustrate two different comparison points. They are alternatives to the PixelForge-operated website for image enlargement; features or prices of either should not be attributed to the .app service.
The project at upscayl.org describes a free, open-source desktop upscaler for Linux, MacOS and Windows, and advertises a cloud offering. Its desktop features and licence are not established features of PixelForge's website.
Bigjpg describes an image upscaling and enhancement service and states it is not a text-to-image platform. PixelForge's service also describes a separate prompt-based editing and generation workspace. The scope difference does not establish which produces better results for a particular image.
The workspace's 20 MB file limit and five-image uploader differ from API megapixel and active-task limits. Pack batch sizes and upload resolutions also differ from the batch page's general paid-workflow description; paid-account combinations were not tested.
The API requires a public HTTPS URL, supports the stated image extensions and PNG output, and does not currently provide customer webhooks or a minimum public result-URL retention period. These integration limits do not describe all browser workflows.
The privacy policy says uploaded images are stored temporarily and automatically deleted within 30 days after processing, with an option to delete them from the account dashboard. The provider says it does not sell the images or use them to train its own AI models. That phrase is limited to its own models: the terms separately state that uploads, prompts and related instructions may be transmitted to relevant third-party model or infrastructure providers as needed for the selected feature. The operator's statement is not a blanket no-training promise on behalf of every external provider.
When analytics is permitted, the policy says a pseudonymous account identifier can be sent to an analytics provider; text fields, uploaded previews and results are masked in session recordings. Network traffic, deletion and masking were not independently tested.
The terms permit personal or commercial use when the user has rights to the image. They state that users retain their rights in uploaded images and grant a limited licence for processing to provide the service. The tool's ability to remove a mark or reconstruct part of an image does not, by itself, establish permission to process the underlying work or a person's likeness.
The service prohibits sexually explicit or other NSFW content and infringement of copyright, trademarks, privacy, publicity or other rights. Reported or flagged content may be reviewed by trained personnel and processing or outputs restricted. These rules do not independently prove detection or a particular report outcome.
This is PixelForge's online site. The similarly named project at upscayl.org describes its own open-source desktop application, whose features and licence should not be presumed to apply here.
The pack comparison displays one credit per image, while the uploader shows different amounts beside some modes: one for Standard and Portrait, three for Face Restore, six for Anime / Art and fifteen for Scratch Restore. The published displays are not fully reconciled in this research, so the selected feature and current interface matter to the stated cost.
The documented API requires a public HTTPS image URL and uses status polling; customer webhooks are not currently available. The manual uploader separately accepts local files.
The policy excludes training the company's own AI models. The terms also allow relevant third-party providers to process uploads, prompts and instructions as needed; the operator's statement does not establish every provider's policy.
