Hy Image 3.5 Preview: n=1 and seed, against Sume's n and seed

OpenRouter lists Hy Image 3.5 Preview with n capped at 1 and a seed. Sume's Image API takes n up to 4 on most rows and returns 400 for seed. A field table.

5 min readSume
All posts

If you move a Hy Image 3.5 Preview integration to Sume, two request fields change meaning. The OpenRouter model page (read 2026-10-08) lists n as limited to 1 image per request and a seed for reproducible results. On Sume, n is a per-model range, mostly 1 to 4, and seed is not served at all: a request that sends it returns 400 unsupported_parameter. Sume does not list a Hy Image model, so this is a porting note, not a model swap.

The date matters because Hy Image 3.5 is a preview and its listing may change. The table below is what the OpenRouter page showed on 2026-10-08 against what Sume's docs and catalog code show on origin/main.

The two listings side by side

The comparison covers only the fields the OpenRouter page names. It does not say how Hy Image prices or behaves beyond those fields, and neither does this post.

Request fields, Hy Image 3.5 Preview on OpenRouter vs Sume Image API (read 2026-10-08)
FieldHy Image 3.5 Preview (OpenRouter)Sume Image API
Images per calln limited to 1n is a range read from the catalog; the default ceiling is 4, Grok Imagine is 1, Soul takes 1 or 4
Seedseed acceptedPart of the schema, advertised by no model: 400 unsupported_parameter
Reference imagesUp to 20 input_references10 by default, 16 on ChatGPT Image 2.5, 5 on Ideogram 4.5, 0 on text-only rows
Resolution1K to 4KA resolution enum only where the row lists one (Nano Banana 0.5K to 4K, Imagen 4 Ultra and Ideogram 4.5 1K and 2K, Soul 720p and 1080p)
ResultBase64 in b64_jsonA signed Sume-hosted URL in data[].url plus media_type

What to change for n

A loop that issues one Hy call per variant can become one Sume call per four variants. n is billed per image, so cost_usd × n is the charge; a failed generation is not billed. Read the range from the catalog instead of hard-coding 4, because a row can lower it. The Grok Imagine row, for one, lists a single image.

import os, requests

r = requests.get(
    "https://api.sume.com/v1/images/models",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    timeout=30,
)
r.raise_for_status()
for m in r.json()["data"]:
    n = m["supported_parameters"]["n"]
    print(m["id"], n["min"], n["max"])

What to do instead of a seed

Sume's docs say seed and output_compression are in the schema but no model advertises them, so do not send either. If you need the same picture again, keep the output URL or download the file when the job completes; a rerun of the same prompt is a new generation and a new charge. If you need a close variant, send the first output back as an input_references entry with a short edit prompt, on a row that lists references.

Because the Image API waits up to 30 seconds before it returns a 202 job envelope, a client that decoded b64_json also needs a branch for the job result. The base64 branch goes away: the result is a URL.

Limits of this comparison

The OpenRouter listing is the only vendor-side page used here for Hy Image. Tencent's own pages, and the price and resolution cap they give, are covered in the other posts linked below. Nothing here claims Sume can run Hy Image 3.5 Preview.

Sources

Related posts

More in Models

All Models posts

Written by Sume