Download social media videos in Python

httpx over the standard library's http.client is worth it here for the timeout argument alone: a source that hangs should fail your request, not hold it. Thirty seconds is a sane ceiling — the gateway's own upstream budget is lower than that.

import httpx

r = httpx.get(
    "https://api.saveapi.org/v1/download",
    params={"url": "https://www.instagram.com/reel/Cx1a2b3c/"},
    headers={"Authorization": "Bearer sk_live_xxxx"},
    timeout=30,
)
data = r.json()

Every response carries the same array whether the post held one video or ten images. Mixed posts keep their types in post order, so a carousel with two photos and a clip gives you two image entries and one video entry. Branch on type and the same code handles every platform.

for media in data["medias"]:
    if media["type"] == "video":
        print("video:", media["url"])
    elif media["type"] == "image":
        print("image:", media["url"])
    elif media["type"] == "audio":
        print("track:", media["url"])

Some sources publish several renditions, and those responses carry formats[] — each entry with a ready URL and its exact size in megabytes. There is no second call and no format parameter: filter the list client-side for the largest file that fits your destination's ceiling.

best = None
for fmt in data.get("formats", []):
    if fmt["size_mb"] <= 45 and (best is None or fmt["size_mb"] > best["size_mb"]):
        best = fmt

url = best["url"] if best else data["medias"][0]["url"]

A timeout or an upstream error is worth one more attempt. A private post, a deleted post or an unsupported domain is not — the second call returns the same answer and, for the errors that are billed, spends a second credit. Split the codes into two sets and act on the set, never on the message text.

RETRY = {"UPSTREAM_TIMEOUT", "UPSTREAM_ERROR"}

def download(link: str, attempts: int = 2) -> dict:
    for attempt in range(attempts):
        data = httpx.get(API, params={"url": link}, headers=HEAD, timeout=30).json()
        if data["success"]:
            return data
        if data["error"]["code"] not in RETRY:
            raise RuntimeError(data["error"]["message"])   # qayta urinish yordam bermaydi
    raise RuntimeError("source unavailable")

Every response carries the key's per-minute limit and what is left of it, and paid keys also carry the credits remaining in the pack. Read them from the response you already have instead of polling a balance endpoint on a timer.

print(r.headers["X-RateLimit-Limit"])       # kalitning daqiqadagi chegarasi
print(r.headers["X-RateLimit-Remaining"])   # shu daqiqada qolgani
print(r.headers.get("X-Credits-Remaining")) # paketdagi qolgan kredit

The resolved URL points straight at the source CDN, so you fetch it directly and nothing passes through the API. Stream it to disk: a 4K reel in memory is the fastest way to get a container killed.

with httpx.stream("GET", data["medias"][0]["url"]) as source:
    with open("reel.mp4", "wb") as out:
        for chunk in source.iter_bytes():
            out.write(chunk)