Remove.bg – A service to remove the background of any photo | Hacker News Reader

Remove.bg – A service to remove the background of any photo

remove.bg

221 pointsAugust-Garcia52 comments

Open articleView on HN

Good marketing too, having made it to the front page here several times in several weeks :)

Previously featured on Show HN: https://news.ycombinator.com/item?id=18697601,

Really useful project, I think.

We do the same as Mobile apps:

iOS: https://itunes.apple.com/us/app/cut-copy-paste-photos/id1126...

Android: https://play.google.com/store/apps/details?id=com.km.cutpast...

We use Image Segmentation with Deep Learning. Our Hair extraction is not as good as theirs. We support Pets, Furniture, Vehicles Etc though.

This is really cool but pricing for commercial images seems high. At work we use pixelz.com which is an API for human image processing. They clear cut/resize/rotate and recolor for around $0.60 per image. Results are fantastic, we have around 1 in 1000 images that have to be sent for reprocessing.

The only advantage I could see to using this is the instant turn around.

Edit: I was looking at the one off pricing. It looks like subscriptions are really good value.

Isn’t that more than what this service charges?

That's 10 times the price of this service.

Somewhat important to note, I think:

> The user may use the platform, which is available at remove.bg, exclusively for non-commercial purposes. This means that all results of the platform (in the following: the platform) may only be used for private purposes. Any use for (direct or indirect) commercial purposes is excluded.

They do have several paid plans that allow commercial use.

If you want to make money out of their work through commercial use, it seems fair to pay them back.

Source: https://www.remove.bg/pricing

Pricing seems fair though. I’d pay 1.99 for an image I needed to use this for commercial use. To do this manually might take a tad longer and may not come out as nice. I haven’t tried it yet but if it’s anything like the examples seems great.

I agree, the single image pricing seems very fair. You could easily try it using the non commercial service to check your happy with the outcome and simply pay the 1.99 to license the image for one off use. Not forcing you to sign up for a monthly plan or pay for multiple images at once as the lowest tier is very smart in my opinion.

I find it interesting that the first image they choose to show on their page to represent their product has an obvious flaw in it. For me, the image of the woman in black overalls with a grey tshirt underneath is the first to show. If you look under her right armpit, you can see that the grey lake is still there and not the expected grey-and-white checkerboard. Maybe they're just trying to be honest about the limitations of their product, but I assume this was an accidental oversight. Obviously this is a part of the image that could be easily fixed by (computer-aided) hand, but it might make more sense as a third or fourth image if that is the point.

There's a pretty bad issue with the hair too, near the top of her head. That was the very first thing I noticed about the image and it gives a poor impression.

The image on the front page changes all the time, so I am not sure which you are talking about.

Here's the image: https://www.remove.bg/images/samples/combined/s1.jpg

They do mention higher res output here Even higher quality with 4000 × 2500 pixels, or any other dimensions with up to 10 megapixels.

https://www.remove.bg/pricing

Image https://www.remove.bg/assets/comparison-hd-9b9708387758e42f0...

What does that have to do with the flaw in the image on the home page?

Tried it on a family picture right in front of the castle of Disney's Magic Kingdom, my family posing and tons of people walking around. The result is incredible, this is really great!

How well does it work with hair? Hair is notoriously hard to select well in Photoshop for most retouchers, so it would be amazing if the automation somehow became better than professional humans doing the editing. There are a ton of manual techniques out there for extraction, none of them work incredibly well out of the box depending on the background you're working with.

I found this class on hair retouching to be great btw, goes pretty in depth on the latest techniques used in commercial-grade work: https://proedu.com/collections/available-tutorials/products/...

Just try it! It's free and quick.

This looks phenomenal to me, and I recommended it to a few photographer friends of mine. They all felt the same way: yes this is super cool, but 10 MP is far too low to use in any professional capacity (in particular when dealing with photographs that would eventually be made into prints). It's cool for social media and stuff, but not anything they'd be able to sell, and thus not something they would pay for.

Currently the list of people I know who would pay for something like this is identical to the list of people I know who would require higher resolutions for it to be useful.

I tried out several different pictures and got pretty good accuracy... That's impressive!

Biggest problem encountered was over correction with some foreground image part removed

popping in the resulting image back into photoshop and layer masking the foreground back with the original image seems to be a quick fix

seems like this could save a lot of time.

IO tried a pretty complex group picture, and it did great, except for a couple of spots...a lot of time saved.

You can also press the edit button then remove/restore parts that it missed.. super handy.

Magic Wand tool in Macromedia/Adobe Fireworks could do something like this quickly years ago.

Is this really a big problem still today?

I'm not an avid photo editor these days.

The magic wand tool (or really any of the "modern" extraction tools in Photoshop) can't handle hair the way they claim to do in those examples. If this is truly done with non-manual work, it's incredibly impressive.

Did you try it? I put in an image and got back a result way too fast for someone to have done it manually.

None of those tools result in what the service has done to several photos I threw at it.

I gave it a photo of Daniel Craig with a black suit on a black background and it did a damn good job of isolating the person.

Even if those tools were right for the job (aren’t, and fireworks doesn’t exist anymore)... to pay our lowest paid art guy to do this is well above this service’s per photo cost.

Incredibly useful tool. Very good domain name. Instant value delivered with free plan. Technology looks great: took 5 seconds to get the output.

All the best.

Nice!

But when I click Edit to change the background color, then hit Save, the pop-up window says "blob:" in the URL and that Safari Can't Open The Page (that's with the pop-ups allowed option). If I go back to the initial results page, it still shows the checkered background.

Any clue at all how much processing this takes? I don't doubt you could build a salable product out of that that worked in real time, even if it took a big rackmount box of FPGAs.

You can upload an image. It spits it out in seconds.

Never mind. I threw some chromakey tests at it and there were way too many artifacts for any sort of broadcaster.

Cool product though. I can see the sense of the pricing and the delivery method. No doubt there's an Adobe plug-in version of this somewhere, and probably more competitors than you might think.

I have seen veteran videographers praise this tool. I can't wait until we get more of these tools for optimizing photos/videos.

This is so nostalgic, looking at this makes me remember the days when i manually used to remove backgrounds from photos with Photoshop.

Impressive if that treatment with the hair is realistic as shown.

Some tests:

Easy: https://i.redd.it/dwi7jrmrwri21.jpg -> https://i.imgur.com/bQZXrAJ.png (Okay, but doesn't mask shadows around the feet correctly)

Medium: https://500px.com/photo/195143459/ximena-by-agata-serge -> https://i.imgur.com/kUdcvFo.png (Okay; it could have taken out a bit more background in some semitransparent sections of hair)

Hard: https://i.redd.it/i7fn0fy89ma21.jpg -> https://i.imgur.com/WGM67XL.png (Fooled with some false-positives, but still a good try)

Certainly not perfect, but definitely usable.

Weirdly I've found it works a lot better on real photos and less well on CG or doctored photos.

okay HN so what doesn't it work on?

From their about page: "Our AI is trained to detect persons as foreground and everything else as background. That's why it only works if there is at least one person in the image."

Think they're referring to one of the top 5 comments on any Show HN post being about how it doesn't work in mobile Safari or Firefox etc.

I tried it on a bee image. The bee was not a person, so it did not work.

I tried a few cartoon images. Realistic styles (most of Avatar: The Last Airbender, most of Ghost in the Shell) work well. Semi-deformed styles (Avatar gag faces, Samurai Jack, The Simpsons) usually didn't work or resulted in artifacting. Non-human cartoons (Spongebob, Adventure Time) never worked.

Some photographs which had low contrast between subject and background were not as cleanly cut out. It was good enough that I wouldbe able to import the mask as a starting point and clean up the edges manually.

Photographs in which faces were obscured where handled surprisingly well!

Would love to see this as a native feature of an image editor.

Yeah, make a (PS/other image editors) plugin that shove it into the API and return the mask that you can correct when needed, that seems the way to go.

I tried it with a photo of a crowd of people and it chose around 20% of the people spread randomly around the photo and removed the others.

Is this just a high-res implementation of Deeplab?

i fucking love this

Not very interesting if not open source

This is pretty cool. What's the guestimate on number of jobs killed?

"No persons found: At the moment remove.bg only works for photos with at least one person in them. Sorry – please select an appropriate image."

Might it be harvesting images for facial recognition?

It’s another way to say their models only segment human subjects at the monent.

Reply on news.ycombinator.com