rembg or a hosted background removal API?

What rembg is

rembg describes itself as a tool to remove image backgrounds that can be used as a command-line tool, a Python library, an HTTP server or a Docker container. It is published under the MIT licence by Daniel Gatis, and the code is on GitHub.

• Install: with pip, choosing one backend: pip install "rembg[cpu]" for the library, or "rembg[cpu,cli]" to add the command line. The README lists Python 3.11 up to, but not including, 3.14.
• GPU option: rembg[gpu] for NVIDIA cards, which may need onnxruntime-gpu, CUDA and cudnn-devel, and rembg[rocm] for AMD cards once onnxruntime-rocm works. If the GPU setup does not work, the README suggests falling back to the CPU backend.
• Command line: rembg i for a single file, rembg p for a whole folder (with a watch mode for new files), rembg b for a stream of raw frames, for example from FFmpeg, and rembg d to download models ahead of time.
• HTTP server: rembg s --host 0.0.0.0 --port 7000 starts a local server with its API documentation at /api and a web interface you can switch off with --no-ui.
• Models: local models are downloaded automatically the first time you use them and stored under ~/.rembg/models/, or wherever the REMBG_HOME variable points. The README puts the default model at about 1 GB, and offers several smaller or more specialised ones.
• Licences of the models: the README states that model weights carry their own licences, independent of rembg's MIT licence, and that the default model's licence requires a paid agreement for commercial use. It asks you to check each model's source before commercial use.

When self-hosting rembg is the right choice

• No per-image fee: the software is free. If you already have a machine with spare capacity, the marginal cost of one more image is close to zero.
• Offline work: once the model is on disk, rembg runs without an internet connection. Good for air-gapped networks, field laptops and build pipelines with no outbound access.
• Privacy and data residency: images never leave your infrastructure. If a contract or a regulation forbids sending images to a third party, this settles the question.
• Control: you choose the model, pin the version, read the code, change it, and add it to your own product. Nothing changes unless you upgrade.
• Learning and research: the library, the CLI and the server are a good way to understand how background removal works end to end.

When a hosted API is the right choice

• No GPU and no server to run: nothing to install, no CUDA versions to match, no model files of up to a gigabyte to download and store, no memory sizing. You send an HTTPS request and get an image back.
• Scaling: a traffic spike is the provider's problem. Self-hosting means sizing machines for your peak, keeping them up and paying for them while idle.
• Plugins for non-developers: with LassoCut, the same account and credits work in Photoshop, Figma, WordPress and WooCommerce, GIMP 3, Zapier, Make, n8n, a CLI and the browser (platforms). Designers on your team do not have to touch Python.
• Consistent results across tools: the web app, the API and every plugin call the same service, so a product photo cut in Figma matches the one your script processed overnight.
• Support and upkeep: someone else watches the service, updates it and answers email (contact@lassocut.com). With rembg you rely on the open-source community and your own team.
• Commercial licensing in one place: you buy credits under one set of terms instead of checking the licence of each model you download.

Side-by-side comparison

We have not run a quality benchmark of rembg against LassoCut, so this table makes no claim about which cuts better. Run both on twenty of your own images before deciding.

The same task in Python, both ways

rembg

From the rembg README, after pip install "rembg[cpu]". The first call downloads the model.

from rembg import remove

with open('input.png', 'rb') as i:
    with open('output.png', 'wb') as o:
        input = i.read()
        output = remove(input)
        o.write(output)

For many files, the README recommends creating one session with new_session() and passing it to each remove(input, session=session) call, so the model is loaded once.

LassoCut

From the LassoCut Python guide, after pip install requests and with your key in LASSOCUT_API_KEY. size=preview is covered by the free previews; use full for full-size images.

import os, requests

r = requests.post(
    "https://api.lassocut.com/v1.0/removebg",
    headers={"X-Api-Key": os.environ["LASSOCUT_API_KEY"]},
    files={"image_file": open("photo.jpg", "rb")},
    data={"size": "preview"},
)
r.raise_for_status()
open("no-bg.png", "wb").write(r.content)
print("credits charged:", r.headers["X-Credits-Charged"])

To process a folder in parallel with retries, see the thread-pool example in the same guide.

What it costs

rembg has no licence fee and no per-image price. What you pay for is the machine that runs it, the time to install and update it, and, for commercial work, whatever the licence of the model you pick requires.

LassoCut sells one-time credit packs, from the pricing page: 1 credit is 1 full-size image.

No subscription, and purchased credits do not expire while your account is open. A preview costs 0.25 credit, and every account gets 50 free previews a month. An API call that returns an error is not charged. Credits are sold to businesses only.

A rough way to decide: if you already own a suitable machine and someone on the team is happy to maintain it, rembg is cheaper at almost any volume. If you would have to rent a GPU server, or pay for a person's time to keep it running, compare that monthly bill with the pack that matches your volume.

Using both

The two are not exclusive. Some teams run rembg for internal or confidential images and use a hosted API where convenience matters: designers in Figma or Photoshop, a WordPress shop, a Zapier or n8n automation. Others prototype with rembg, then move to a hosted API when traffic grows, or the other way round when volume makes owning a GPU worth it.

For a wider look at hosted options and their prices, see the background removal API comparison.

Questions

What to know about LassoCut

• A new service from a small independent company, JAUPIN Design LLC.
• Credits are sold to businesses only.
• The free allowance is 50 previews a month (up to 0.25 megapixel); full-size images need credits.
• It only removes backgrounds; it is not a general image editor.

rembg is open-source software by Daniel Gatis, published under the MIT licence; LassoCut is not affiliated with the rembg project. Other names are trademarks of their owners. References are for identification and comparison only.