Photo to felt and knitted looks: three Grok Imagine calls, 7.5 cents

To see one photo as felt, knitted wool and needle-felted fiber, send three requests to x-ai/grok-image, each with the photo as an input_references entry and a different craft in the prompt. Sume's catalog lists this row at $0.025 per image and caps it at one image per call, so the three looks are three calls and $0.075 in total, with no way to batch them into n: 3.

Why three calls and not n=3?

The n range for Grok Imagine on Sume is 1 to 1. Most other rows take 1 to 4. A request for more than the row's ceiling comes back as a 400, so the loop is the supported route. Sume also rejects any field the row does not list, so keep the body to model, prompt, input_references and aspect_ratio.

Grok's ratio list is the longest on the platform: 2:1, 20:9, 19.5:9, 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16, 9:19.5, 9:20 and 1:2. It has no 4:5 and no auto, so crop a 3:4 result if you need an Instagram portrait.

What does the script look like?

Each craft gets its own prompt. The loop stops on the first non-200 response, because a 202 means a job envelope and needs the job result instead.

import os
import requests

PHOTO = "https://example.com/photo.jpg"
CRAFTS = {
    "felt": "flat wool felt appliqué, stitched edges, visible fibers",
    "knit": "hand-knitted wool, chunky stitches, soft yarn texture",
    "needle": "needle-felted miniature, fuzzy fibers, studio light",
}
head = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}

for name, look in CRAFTS.items():
    body = {
        "model": "x-ai/grok-image",
        "prompt": f"This photo remade as {look}",
        "input_references": [{"type": "image_url",
                              "image_url": {"url": PHOTO}}],
        "aspect_ratio": "3:4",
    }
    r = requests.post("https://api.sume.com/v1/images",
                      headers=head, json=body, timeout=60)
    if r.status_code != 200:
        raise SystemExit(f"{name}: {r.status_code} {r.text[:150]}")
    print(name, r.json()["data"][0]["url"])

How do I pick the ratio for a landscape photo?

The row has no auto, so choose the closest value from its list yourself: 3:4 for a portrait photo, 4:3 or 3:2 for landscape, 1:1 for a square crop. Divide width by height, compare against the list, and take the nearest. Sume returns Sume-hosted signed URLs rather than inline image data, so download each file in the same script instead of storing the link for later.

What does the set cost?

Sume bills a completed generation in full and does not bill a failed or cancelled one, according to its docs. Endpoint pricing lines already include the Sume margin, so the cost is the row's cost_usd times the image count.

Run the three looks as three separate calls with the same photo and a different one-line material prompt, felt, knitted wool and embroidered patch. At $0.025 each, the set is $0.075, so you can keep all three and decide by eye. Because n is capped at 1 on this row, there is no cheaper batch option; the loop is the batch.

Which look should I trust first?

Run the flat felt appliqué first. It has the simplest geometry, so a good result there tells you the portrait ratio and the reference are fine before you spend on the fuzzier looks. If a call times out at 30 seconds, the 202 path is described in Jobs and results.