· reAPI Team
Photo Resolution Enhancer: Remini vs Free Open-Source Upscalers
A photo resolution enhancer guesses new detail with AI. Remini, free tools like Upscayl and Real-ESRGAN, an API at $0.002 per image, and where upscaling fails.
A photo resolution enhancer enlarges an image with an AI model that fills in detail the original never contained. Upscayl, a free tool with more than 51,000 GitHub stars, describes the process plainly: it enhances images "by guessing what the details could be".[1] That guess can look convincing on a landscape or a product shot and can quietly change a face, which is the most useful thing to know before choosing a tool.
There are three practical routes. Remini is a one-click web and mobile app. Upscayl and Real-ESRGAN are free, open-source programs you run on your own computer. And for developers, Real-ESRGAN is available as a hosted API at about $0.002 per image. Each is covered below with what its own documentation says, checked on November 4, 2026.
What an AI photo resolution enhancer actually does
Classic resizing, such as bicubic or Lanczos scaling, spreads the existing pixels over a larger grid, so a 4x enlargement looks soft. An AI upscaler such as Real-ESRGAN is a model trained to predict what a sharp version of a low-resolution image probably looked like, then paints that prediction in.[2] The result is sharper, but the new edges and textures are generated, not recovered.
Faces get special treatment. Real-ESRGAN integrates GFPGAN, a separate face-restoration model, as an optional face-enhancement step.[2] Remini asks users to consent to the use of "information extracted from faces" in their media to upscale them.[3] In both cases the face in the output is partly reconstructed, so do not treat an upscaled photo as evidence of what someone looked like.
remini: the one-click option
Remini is an AI photo and video enhancer that promises to "improve, sharpen and unblur" files in one click, on the web at app.remini.ai. Its web footer names the operator as AI Creativity S.r.l. of Milan, a company under the management and coordination of Bending Spoons S.p.A.[3]
Before uploading, note the consent line on the upload page: you agree to the collection and use of facial data to upscale your media. Remini says it does not use that data to identify individuals and points to its privacy policy for retention details.[3] The web page does not list prices.
Free open-source photo upscalers
| Tool | How you use it | License | What it needs |
|---|---|---|---|
| Upscayl | Desktop app for Windows, macOS and Linux | AGPL-3.0 | A Vulkan-compatible GPU; most integrated graphics will not work |
| Real-ESRGAN | Command line or Python; portable executables for Windows, Linux and macOS | BSD-3-Clause | Portable builds run on Intel, AMD and NVIDIA GPUs |
| chaiNNer | Node-based image-processing app | GPL-3.0 | Depends on the models you load |
Sources: each project's GitHub repository.[1][2][4]
Upscayl is the most approachable of the three because it is a regular desktop app rather than a command-line tool. It runs Real-ESRGAN under the hood through a Vulkan backend, and its FAQ says it will not work on most integrated GPUs or CPUs.[1] It was still being updated in October 2026.
Real-ESRGAN is the model most of these tools build on. It ships general models such as RealESRGAN_x4plus and RealESRGAN_x2plus, plus models tuned for anime images and anime video. Its --outscale option allows any output scale, which it achieves by resizing the model's output with Lanczos.[2] The repository's last code push was in August 2024, so treat it as stable rather than actively developed.
chaiNNer started as an upscaling app and grew into a general node-based image-processing tool. It suits batch jobs where upscaling is one step among several, such as denoise, upscale, then sharpen.[4]
Choosing 2x or 4x
Pick the scale from where the photo is going, not from the largest number on offer. Real-ESRGAN ships separate 2x and 4x general models, and any other scale is produced by resizing the model's output with Lanczos, so a 3x setting resizes a model's output rather than using a dedicated 3x model.[2]
Print size is a good way to decide. At a common print density of 300 pixels per inch, a 1,024 x 768 photo prints at about 3.4 x 2.6 inches. Upscaled 2x to 2,048 x 1,536, it reaches about 6.8 x 5.1 inches; at 4x, 4,096 x 3,072, about 13.7 x 10.2 inches. For a phone screen or a web page, 2x is usually enough, and the smaller enlargement means less invented detail.
Match the model to the content as well. Real-ESRGAN provides RealESRGAN_x4plus_anime_6B for anime and illustration, which its authors describe as optimized for anime images with a much smaller model size than the general one.[2]
Photo upscaling by API
For apps that need upscaling without running GPUs, Replicate hosts Real-ESRGAN "with optional face correction and adjustable upscale" at $2 per thousand output images, or $0.002 per image, on NVIDIA T4 hardware.[5] At that price, 10,000 product photos cost about $20 to process, before any retries.
reAPI does not currently offer a still-image upscaler. It does offer video upscaling, through Enhance Video 1.0 and Topaz Video Upscaler, and our guide to upscaling video with AI compares those with local tools.
FAQ
Can a photo resolution enhancer recover real detail?
No. It generates plausible detail based on its training. Upscayl's own description is that it works "by guessing what the details could be".[1] Text, faces and fine patterns are the most likely to change.
Is Remini free?
The Remini web page does not list prices, so we cannot confirm a free tier from it.[3] Upscayl, Real-ESRGAN and chaiNNer are free and open source.
Do I need a GPU to upscale photos?
For Upscayl, yes: it needs a Vulkan-compatible GPU.[1] If you do not have one, a hosted API or an app like Remini does the processing for you.
Choosing a photo resolution enhancer
For an occasional family photo, a one-click app is the least effort, as long as you are comfortable with how it handles facial data. For regular use on a machine with a decent GPU, Upscayl gives the same Real-ESRGAN models for free and keeps the files on your computer. For a product or pipeline, a hosted Real-ESRGAN endpoint at about $0.002 per image is the cheapest photo resolution enhancer to integrate.
References
- Upscayl. upscayl/upscayl repository and README (FAQ). Retrieved November 4, 2026 from github.com/upscayl/upscayl
- Xintao Wang et al. Real-ESRGAN repository and README. Retrieved November 4, 2026 from github.com/xinntao/Real-ESRGAN
- Remini. Remini Web photo enhancer. Retrieved November 4, 2026 from remini.ai
- chaiNNer. chaiNNer-org/chaiNNer repository. Retrieved November 4, 2026 from github.com/chaiNNer-org/chaiNNer
- Replicate. nightmareai/real-esrgan model page and pricing. Retrieved November 4, 2026 from replicate.com/nightmareai/real-esrgan