Does background removal upload the photo? No
Run Lift the subject in ReadyIMG and the work happens in the browser you already have open. Pixels do not pass through a ReadyIMG server. There is no sign-in, no sign-up, and no account server waiting to receive an image. You press start, this device computes a new file, and you download it yourself.
So the photo is not uploaded to ReadyIMG. What you send to a client afterwards is the copy you downloaded. The original is still in the folder it came from, unless you move, replace or delete it yourself.
The photo stays on this machine. What you download is a transparent PNG this device just produced.
The first run fetches a model
The first time you open this step, the browser downloads a detection model from the model publisher and puts it in this device cache. That file exists to tell foreground from background. It is not your photo. The computation still runs here, and the model only decides where the edge is.
A model download and your picture are two separate things. The download goes from the publisher to your browser; the image does not travel the other way. That is why a first run that touches the network is not the same event as a photo being uploaded. Doing the detection locally costs you that one download, and buys you every later image being computed on the same device without handing over an original first. What sits in the cache is the model file, not your frame.
Only two things touch the network
When the page opens, the visit is counted by 51.LA. What gets counted is that someone opened a page, not an image file, and the counter does not learn which picture you dropped in. The same boundary is written up in How files are handled.
The other one is the model above. Beyond those two, background removal has no reason to send the frame anywhere to be computed. The claim that the whole computer stays offline would be false. The claim that the image is not uploaded to ReadyIMG holds.
The traffic is a page count and a model. The image file is not part of it.
What you get back is a transparent PNG
The output of this step is always PNG, because PNG is the format that keeps an alpha channel. Once the background is gone, the space around the subject is empty rather than white. Drop that file on a dark page or over another photo and the edge will not carry a pale halo with it. These PNGs are often larger than the JPG you started from, since the empty area still takes up room. Before you send one, run it through Shrink file and keep the format as it is, so the transparency is not filled in. Do not finish the job with a JPG conversion; that step paints colour into the empty pixels.
This step works on the one picture in front of you. Loading forty files at once belongs to the batch steps; a cutout needs its edge checked one at a time, so do not expect it to finish a folder in one pass.
When the edge is messy, brush it
An automatically detected boundary is not right every time. Loose hair, a translucent scarf, a cuff that sits close to the background colour, all tend to keep a rim of the old backdrop. Use the brush to paint that leftover background away. You are painting on the result you downloaded, so the original is untouched.
Sometimes the cutout simply fails. The page does not pretend otherwise; it points you at the paint-over option so you can mark the subject yourself. Anyone promising a clean result on every image is overselling. This step gives you a starting edge, and the last few strokes are yours.
Three looks before you send
- Is there a rim of the old background around the subject? A dark page shows it fastest.
- Is the area that should be transparent actually empty, rather than filled with white?
- What you downloaded is a new file. Is the original still where it was?
One more thing gets forgotten: the queue lives only inside this one visit. Refresh the page or close the tab and any result you have not downloaded is gone. Download first, then go do the rest.
Order matters
Set the framing before you cut, and the job gets easier. With the subject already in the middle of the frame there is less area to classify and the edge is quicker to inspect. Resize after the cut instead of before, so the transparency is not flattened and the edge is not smeared by a second resample.
If nothing is loaded yet, go back to the bench and drop the picture in, then open Lift the subject. To check where a file actually goes, read the page on how files are handled.