Repeatable AI images: seed returns 400 on Sume, so do this

Do not send seed. The field is in the request schema for OpenRouter parity, but the catalog is the allowlist and no model advertises a seed descriptor, so Sume returns 400 unsupported_parameter rather than ignoring it. output_compression is in the same state. For a repeatable result, keep the generated file instead of trying to regenerate it.

Fields that are in the schema but rejected

From the residuals table in the repo's docs/api/images.md and the public docs.

Getting consistency without a seed

For a family of images that must match, use a reference. Generate one approved frame, then pass it in input_references with an edit-capable row and describe the change. For a set that must share a style, reuse the same model id, prompt prefix and ratio, and keep the files you like.

Keep a log of the requested model, the stored model and the job id, since retired ids run as their successors.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={"model": "openai/gpt-image-2.5", "prompt": "test", "seed": 7},
    timeout=60,
)
print(r.status_code, r.json().get("error", {}).get("code"))

Note

Check the live catalog; a future row may list seed, and the descriptor will say so.