Free AI Photo Tools: Where the Free Tier Actually Stops

Photo tools are the category where the word free does the most damage. A background remover will process your image in two seconds and show you a perfect cutout. Then you click download and discover the free file is a quarter of a megapixel, which is roughly a five hundred pixel square. The work was free. The usable output was not.

That pattern repeats across every job below. What follows is the version with the numbers in it: for each task, the tool, the published free limit, and the point at which you either pay or move the job onto your own hardware. Where a vendor publishes a limit we quote it. Where a vendor does not publish one, we say so rather than guess.

Making a blurry or low-resolution photo sharp

This is the job with the cleanest free answer, because the best tool in the category is not a service at all.

Upscayl is a free and open source AI image upscaler for Linux, macOS and Windows, released under the AGPL-3.0 licence. It runs entirely on your own machine, which means no upload, no account, no monthly cap and no resolution ceiling other than what your hardware will chew through. The trade is that it is a desktop application you have to install, and that a large upscale on an old laptop takes real minutes rather than seconds.

The hosted comparison is Bigjpg, which works in a browser with no installation. Its published cap is explicit: free users are limited to 3000 by 3000 pixels, paid users go to fifty megapixels. Three thousand square is nine megapixels, which is more generous than most free tiers in this article and enough for the majority of family photos.

The decision rule: one photo, use Bigjpg and skip the install. A folder of photos, or anything above nine megapixels, install Upscayl once and stop thinking about limits.

Restoring an old or damaged photograph

Restoration is really three jobs stacked on top of each other, and the videos tend to show only the flattering one.

The first job is resolution, covered above. The second is damage repair, which means filling in creases, tears and missing corners. The third is face reconstruction, which is where the honest caveat lives: these models do not recover a face, they generate a plausible one. On a heavily degraded photograph the result can be sharp, convincing, and subtly not the person. That matters more for a family archive than it does for a product shot.

Hotpot.ai bundles most of these steps as separate tools, and its own pricing page states the position plainly: most Hotpot products offer free access for exploration, while commercial usage and advanced settings require credits. That is a fair description of the whole category. Free is enough to see whether the tool can help your specific photograph. It is not enough to process an inherited shoebox.

Practical advice: keep the original scan untouched, work on a copy, and if the restored face looks wrong to someone who knew the person, trust them over the model.

Colourising a black and white photo

Palette.fm is the tool named most often here, and its free tier is unusually well documented, which makes it a good illustration of how these limits are structured.

The free plan gives you unlimited colour previews. Those previews are resized to a maximum of 500 by 500 pixels and carry the Palette logo watermark in the bottom right corner. You also get one free trial credit, which produces a single watermark-free colourisation at full resolution, up to 5000 by 5000 pixels. After that it is one credit per photo on a paid plan.

Read that structure carefully, because it is the template for the entire category. The free tier is not a smaller version of the product. It is a demonstration of the product, deliberately capped below the resolution at which the output becomes useful. That is a reasonable business model and there is nothing dishonest about it, but it is the exact detail a twenty second video leaves out.

The other thing worth saying about colourisation: the colours are inferred, not recovered. The model is guessing that a jacket was probably brown. For a family photo that is a lovely effect. For anything documentary or historical, the output is an illustration, not evidence.

Removing a background

Remove.bg is the most-cited tool in this category and it publishes its limit clearly. It is free for low-resolution images, up to 0.25 megapixels, processed on the website for personal use. You also get fifty free previews per month for use with the desktop app or the API. Higher resolutions, up to fifty megapixels, and any commercial use require a paid plan or pay-as-you-go credits.

A quarter of a megapixel is about 500 by 500 pixels. That is fine for a profile picture or a small web thumbnail and immediately too small for a product listing, a print, or anything you intend to composite. This is the specific case that generates the most complaints, because the preview looks perfect and the download does not.

There is also a change worth knowing about before you build a workflow on it. Remove.bg's own API documentation states that from 1 December 2026, background removal moves to Leonardo.Ai, which is also part of Canva. If you are automating anything against that API, that is a migration on your calendar rather than a surprise.

Removing an object or a person from a photo

Generative fill and object removal are the newest of these jobs and the ones where free tiers move fastest, so specific numbers here go stale quickest.

The general shape is consistent with everything above: a free allowance sufficient to try it, a credit system beyond that, and a resolution or watermark cap on the free output. Check the vendor's own pricing page on the day you use it rather than trusting a number in an article or a video, including this one.

The more useful point is about the result rather than the price. Removing a person from a photograph does not reveal what was behind them, it invents it. On a plain wall or a patch of grass that is invisible and excellent. In front of detailed architecture, repeating patterns or another person, the invented region is usually obvious at full size even when it looks clean in a thumbnail. Always check the output at one hundred percent before you use it.

Turning a phone snapshot into a portrait

This one is worth separating out because it is not a single tool, it is four of the jobs above run in sequence: upscale, sharpen, relight the face, then replace or blur the background.

That matters for two reasons. The first is that each step carries its own free-tier cap, so a workflow that looks free in a demonstration can cost credits at three separate stages. The second is that the errors compound. An upscaler invents a little detail, a face enhancer invents a little more on top of it, and by the fourth pass the person in the photograph is noticeably smoothed into someone slightly different. If the result has to look like a specific human being, do fewer passes, not more.

How to decide, in one rule

Across every job on this page the free tier is capped in the same place: resolution. Not features, not speed, not quality. The model does the same work either way; the free file is just too small to use. Once you see that, the decision gets simple.

• One photo, viewed on a screen — the hosted free tier is genuinely enough. Take it and move on.
• Anything printed, listed or composited — you will hit the resolution cap immediately. Budget for credits or install a local tool before you start.
• A whole folder — go local and open source now. Per-image credits and a batch job are a bad combination.
• Anything commercial — read the licence, not the price. Several free tiers on this page permit personal use only, which is a separate restriction from the resolution cap.
• Faces and history — remember that restoration and colourisation generate detail rather than recovering it. Keep the untouched original.

The short videos are accurate about what these tools can do. They are simply quiet about the size of the file you get to keep, and that is usually the only detail that decides whether a tool survives contact with your actual photographs.

Frequently Asked Questions

What resolution can I download for free from remove.bg?

Remove.bg states it is free for low-resolution images up to 0.25 megapixels processed on the website for personal use, which is roughly a 500 by 500 pixel file. It also gives fifty free previews per month for the desktop app or API. Higher resolutions up to fifty megapixels and any commercial use require a paid plan or credits.

Is there a free AI upscaler with no limit at all?

Yes, if you run it yourself. Upscayl is a free and open source upscaler licensed under AGPL-3.0 that runs locally on Linux, macOS and Windows, so there is no account, no upload and no monthly cap. The cost is a one-off install and your own processing time.

What does the Palette.fm free plan actually give you?

The free plan offers unlimited colourisation previews, but those images are resized to a maximum of 500 by 500 pixels and carry the Palette logo watermark in the bottom right. It also includes one free trial credit for a single watermark-free colourisation at full resolution, up to 5000 by 5000 pixels.

How large an image can Bigjpg enlarge for free?

Bigjpg publishes a limit of 3000 by 3000 pixels for free users, with paid users able to process up to fifty megapixels.

Are AI-restored and colourised photos accurate?

No. These models generate plausible detail rather than recovering lost detail. A reconstructed face is the model's best guess at a face, and colourisation infers colours that were never recorded. Keep the untouched original, and treat the result as an illustration rather than a document.