Seedream 5.0 Lite makes 1-6 images per run: how n works on Sume

fal says Seedream 5.0 Lite supports batch generation of 1 to 6 images per generation. On Sume, n is a request field with a 1 to 10 cap for the route and a lower ceiling per model, published as the n range descriptor in the catalog, so the safe move is to read supported_parameters.n.max for bytedance-seed/seedream-5-lite instead of copying fal's 6.

The 1 to 6 figure is from fal's page, read 2026-10-03. The n rules are from the Image API docs.

Why the ceiling differs

Sume's docs say it requests up to 10 images per call with n, and that per-model ceilings are lower, to be read from the catalog. Validation is catalog-driven: a value the model does not list is rejected with 400 unsupported_parameter instead of being clamped, so a request that is fine on fal can fail on Sume if the catalog ceiling is smaller.

Read the real number

This script prints the n ceiling for every listed model; look for the Seedream rows.

import os
import requests

key = os.environ["SUME_API_KEY"]
url = "https://api.sume.com/v1/images/models"
resp = requests.get(url, headers={"Authorization": f"Bearer {key}"}, timeout=30)
resp.raise_for_status()
for model in resp.json()["data"]:
    params = model["supported_parameters"]
    refs = params.get("input_references", {}).get("max", 0)
    n_max = params.get("n", {}).get("max", 1)
    print(model["id"], "refs:", refs, "n:", n_max)

Limits

Billing is per generation and, per the docs, cost_usd times n is what you pay, so a larger n is a larger charge. Large batches are also the ones likely to run past the 30-second wait and come back as a 202 job envelope, so read the result from the job endpoints; the Jobs and results page covers that.