Soul 2 lookbook batch of 4: Higgsfield Soul on Sume

Can Soul 2 make a lookbook set in one request?

Yes: ask for a batch of 4. Higgsfield's Soul 2 page says "batch_size is 1 or 4 only", and Sume's catalog entry higgsfield/soul follows that, so one request returns either one image or four, never two or three.

Soul 2 is text-to-image only on Sume, so it cannot restyle a photo you already have. It suits mood and styling boards where the garment is described, not product-accurate catalog shots.

What can I set?

Higgsfield's page lists a 720p default and a 1080p option, a set of aspect ratios, an optional seed, an enhance_prompt flag that defaults to false, and an optional custom character reference with a strength above 0 and up to 1. Sume's catalog exposes a narrower set, shown below.

How should I plan a lookbook run?

Fix the ratio first. Portrait 3:4 or 2:3 suits a lookbook page and 16:9 suits a hero banner; Sume's list has no 4:5, so use 3:4 for tall shots. Set aspect_ratio explicitly instead of leaning on a default, because Higgsfield's page names 1:1 as its default.

Run at 720p to pick a direction and 1080p for the set you keep. Each request with n: 4 returns four takes of the same prompt, so change the prompt between requests to vary outfits or settings.

import asyncio, os
import requests

async def main():
    r = requests.post(
        "https://api.sume.com/v1/images",
        headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
        json={
            "model": "higgsfield/soul",
            "prompt": "Editorial lookbook, linen jacket, soft morning light, street corner",
            "aspect_ratio": "3:4",
            "resolution": "1080p",
            "n": 4,
        },
        timeout=60,
    )
    print(r.status_code)
    print(r.json())

asyncio.run(main())

What about price and timing?

Sume's docs do not publish a fixed Soul price on the Image API page, so read the pricing line from the model's endpoint record at /v1/images/models/higgsfield/soul/endpoints before you budget a batch. A request that outlives the 30-second wait returns a 202 job envelope, which is likely for four images at 1080p, so poll the job result rather than retrying.