Localize one 30-second ad into six languages: Wan 3.0 cost on Sume

Six localized versions of one 30 second ad cost $22.50 at 720p on Sume's wan-3.0, or $45.00 at 1080p, because each version is its own paid render at $0.125 or $0.25 a second. A cheaper path is to draft all six at 480p for $11.25 and render finals only for the versions whose on-screen text came out right.

The setup

Alibaba's Wan 3.0 repository says the model renders long-form text across 12 languages (read 2026-10-05). That makes it a candidate for ad versions where words are part of the picture: a shop sign, a price card, an end slate. Sume has no localization switch, so each language is a separate request with the translated string written into the prompt.

Cost for six languages

Each row is clip seconds times the Sume per-second rate (provider list x 1.25): $0.0625 at 480p, $0.125 at 720p, $0.25 at 1080p. A 30 second clip is 30 of those.

When the draft pass saves money

Drafting only pays when you drop versions. If all six pass, the drafts are pure overhead: six drafts plus six 720p finals is $11.25 + $22.50 = $33.75, against $22.50 for going straight to 720p. Draft when you expect the text to fail in some scripts, not as a habit.

Keep the prompts identical except the string

Write the translated string once per language and keep the rest of the prompt identical, so differences in the result come from the text. Use a different Idempotency-Key per language so a retry cannot double bill one version and so each version is a distinct job.

import asyncio, json, os, urllib.request

LINES = {"es": "Abierto hasta las 9", "fr": "Ouvert jusqu'a 21 h", "de": "Geoffnet bis 21 Uhr"}

def submit(lang, text):
    body = {"model": "wan-3.0", "resolution": "480p", "duration": 30,
            "aspect_ratio": "16:9",
            "prompt": f"Cafe storefront at dusk. The sign reads exactly: {text}"}
    req = urllib.request.Request("https://api.sume.com/v1/videos",
        data=json.dumps(body).encode(), method="POST",
        headers={"Authorization": "Bearer " + os.environ["SUME_API_KEY"],
                 "Content-Type": "application/json",
                 "Idempotency-Key": f"ad-draft-{lang}-001"})
    return json.load(urllib.request.urlopen(req))["id"]

async def main():
    ids = await asyncio.gather(*(asyncio.to_thread(submit, k, v) for k, v in LINES.items()))
    print(ids)

asyncio.run(main())

Concurrency

Concurrency matters for a six-language batch. Sume accepts valid jobs as queued when your workspace is at its processing limit; the plan limits are 1 (Free), 4 (Pro), 8 (Startup) and 20 (Scale) jobs processing at once per the generation admission docs. Six jobs on Pro run four at a time and two wait in the queue.

Budget for retries and review

The cost of a language version has three parts: the render, the retries, and the review. The render is fixed by the table. Retries are where budgets slip: a garbled sign means a second render of the same length. Plan for one retry in every three versions on a first campaign, and your six-language 720p plan becomes $22.50 + 2 x $3.75 = $30.00. That is a planning habit, not a measured failure rate, so replace it with your own numbers after the first batch.

Review is people time. Ask a native reader to approve each version from a 480p draft, because spelling can be judged at low resolution when the sign is large in frame. Spend the final-resolution money only after approval.

Length and sound

Wan 3.0 accepts 2 to 30 seconds on Sume, so a 30 second ad is the longest single clip. If your ad is 15 seconds, the same six languages cost half as much: $11.25 at 720p. Seedance 2.5 also reaches 30 seconds on Sume at 4 to 30 seconds, but it is billed by tokens rather than a flat per-second rate, so use its catalog entry for a quote rather than this table.

If the voice has to change per language too, use generate_audio carefully: Wan 3.0 on Sume produces audio by default, but the documented fields offer no way to pick a spoken language. For narration in a specific language, many teams generate the video silent and lay narration over it in an editor.

Sources: Wan 3.0 repository (read 2026-10-05) for the vendor feature, and the Video Router docs for the 2 to 30 second range.