Four music variants in parallel: Python asyncio, one key per take

To get four Music Router variants at once, run four submit-and-poll coroutines with asyncio, each with its own Idempotency-Key. The cost is 4 x $0.125 = $0.50, and a retry with the same key does not create a second job. The script below uses only the standard library.

The script

It posts to /v1/music-router/generate with sume/music-auto, then polls /v1/jobs/{id}/status until terminal is true, then reads /v1/jobs/{id}/result. Calls to urllib run in threads so the four jobs overlap. Set SUME_API_KEY first.

import asyncio, json, os, urllib.request

H = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"],
     "Content-Type": "application/json"}

def call(method, path, body=None, key=None):
    h = dict(H, **({"Idempotency-Key": key} if key else {}))
    data = json.dumps(body).encode() if body else None
    req = urllib.request.Request("https://api.sume.com" + path, data, h, method=method)
    with urllib.request.urlopen(req) as r:
        return json.load(r)

async def take(n, bpm):
    prompt = f"Warm lo-fi, {bpm} BPM, C minor. A 30-second track. Instrumental, no vocals."
    job = await asyncio.to_thread(call, "POST", "/v1/music-router/generate",
        {"model": "sume/music-auto", "prompt": prompt}, f"jingle-take-{n}")
    jid = job["request_id"]
    while True:
        st = await asyncio.to_thread(call, "GET", f"/v1/jobs/{jid}/status")
        if st.get("terminal"):
            break
        await asyncio.sleep(st.get("next_poll_after_seconds") or 5)
    return await asyncio.to_thread(call, "GET", f"/v1/jobs/{jid}/result")

async def main():
    results = await asyncio.gather(*(take(i, 72 + 12 * i) for i in range(4)))
    print(json.dumps(results, indent=2)[:2000])

asyncio.run(main())

Why each take has its own key

Writes need an Idempotency-Key. If you reuse one key across four different prompts, the later requests can be treated as repeats of the first. A key per take (here jingle-take-0 to jingle-take-3) means a network retry returns the same job, and a deliberate new variant uses a new key.

The tempos step 12 BPM apart (72, 84, 96, 108), which the Music docs suggest when you want clearly contrasting takes.

Reading the result

Read the audio artifact from result.artifacts where type is audio. When present, result.lyrics carries the model-reported lyrics or section map. job.request.routed_model tells you which engine ran. The prompt asks for a 30-second track, but there is no duration parameter, so check the length of what comes back and trim with Timeline if needed.

Extending it

To run twelve takes, change range(4) to range(12), but respect your account's concurrency limits: if you hit a rate limit, run in batches of four. A twelve-take run costs 12 x $0.125 = $1.50.

The script prints the first 2,000 characters of the combined JSON. In production, pull the audio URL from the artifacts list and download it right away, since durable media URLs are what you want to keep.