Remove Image Backgrounds in Google Colab or Jupyter Without a GPU (API Notebook)

In Colab, open the Secrets panel (the key icon in the left sidebar), add BGCLEAR_API_KEY with your key from the dashboard, and enable notebook access. Secrets are not saved into the .ipynb, so the notebook can be shared safely. Local Jupyter: use an environment variable instead.

import os, requests
try:
    from google.colab import userdata           # Colab
    KEY = userdata.get("BGCLEAR_API_KEY")
except ImportError:
    KEY = os.environ["BGCLEAR_API_KEY"]         # local Jupyter

API = "https://www.bgclear.ai/api/v1/remove"
HEADERS = {"Authorization": f"Bearer {KEY}"}

def remove_background(data: bytes, size="preview", fmt="png", bg_color=None) -> bytes:
    form = {"size": size, "format": fmt}
    if bg_color: form["bg_color"] = bg_color
    r = requests.post(API, headers=HEADERS, files={"image_file": ("image.jpg", data)}, data=form, timeout=90)
    if r.status_code != 200:
        raise RuntimeError(r.json()["error"]["message"])
    print("credits remaining:", r.headers.get("X-Credits-Remaining"))
    return r.content

size="preview" (up to 800 px) is free and unlimited — ideal while you iterate; switch to "full" (one credit) or "auto" for final outputs.

from google.colab import files          # local Jupyter: use ipywidgets FileUpload or a path
from IPython.display import Image, display

uploaded = files.upload()               # opens the file picker
for name, data in uploaded.items():
    png = remove_background(data, size="preview")
    display(Image(data=png, width=400))
    with open(name.rsplit(".", 1)[0] + "-cutout.png", "wb") as f:
        f.write(png)

The checkerboard you would see in an editor is not shown by display — transparent areas render on white in the notebook, which is fine for checking edges. For a white-background product image, pass bg_color="ffffff", fmt="jpg".

Mount Drive, walk a folder, skip files already done, pace under the 60-requests-a-minute limit, and use each file's hash as an Idempotency-Key so a rerun after a runtime reset never charges twice.

import hashlib, pathlib, time
from google.colab import drive
drive.mount("/content/drive")

src = pathlib.Path("/content/drive/MyDrive/products/raw")
dst = pathlib.Path("/content/drive/MyDrive/products/cutouts"); dst.mkdir(parents=True, exist_ok=True)

for path in sorted(src.glob("*.jp*g")) + sorted(src.glob("*.png")):
    out = dst / (path.stem + ".png")
    if out.exists():
        continue                                        # resumable
    data = path.read_bytes()
    r = requests.post(API, headers={**HEADERS, "Idempotency-Key": hashlib.sha256(data).hexdigest()},
                      files={"image_file": (path.name, data)}, data={"size": "full", "format": "png"}, timeout=90)
    if r.status_code == 429:
        time.sleep(int(r.headers.get("X-RateLimit-Reset", "5"))); continue
    if r.status_code == 402:
        raise SystemExit("Out of credits — https://www.bgclear.ai/api-pricing/")
    if r.status_code != 200:
        print(path.name, "->", r.json()["error"]["message"]); continue
    out.write_bytes(r.content)
    print(out.name, "| credits left:", r.headers.get("X-Credits-Remaining"))
    time.sleep(1.1)                                     # ≤60/min

Results land in Drive as they finish, so a Colab timeout mid-run loses nothing — rerun the cell and it continues.

You can — rembg installs in Colab — but the free tier's CPU runtime takes several seconds per image, the model download repeats on every fresh runtime, and GPU availability on the free tier is not guaranteed. The rembg comparison goes through the trade-offs. For a notebook that has to work every time for a class, a client or a colleague, an API call is the boring choice: previews free, full resolution one credit — $9 for 100, $39 for 500, $129 for 2,000, no subscription (pricing). Images above 4 megapixels go through the async jobs endpoint; the Python tutorial has that cell too.