Using lyria-3.5 for E-Commerce Ad Music (Not Product Images)

Lyria 3.5 Makes Music, Not Product Photos — Here's the Right Way to Use It for E-Commerce

If you searched for "lyria-3.5 product images," the short answer is: lyria-3.5 is hiapi's Text-to-Music model — it cannot generate product photos. There's no image output, no image_urls field, nothing visual in its schema at all. What it's actually built for is background music and soundtracks, and that turns out to be a real, underserved need for e-commerce sellers: every product video, ad, and social clip needs a soundtrack, and licensing stock music is slower and pricier than most teams expect.

This guide shows the scenario that actually works: using lyria-3.5 to generate royalty-free background music for e-commerce product videos and ads, with a real request, a real response, and a real track we generated for this piece.

Why This Scenario Fits lyria-3.5

hiapi's own model description for lyria-3.5 calls out ads by name as an intended use case — alongside podcasts and indie pop demos. That's not a stretch we're making; it's the platform's own framing. E-commerce ad music is squarely inside that lane:

• Product demo videos (Shopify/TikTok Shop style) need a 15-40 second instrumental bed that doesn't compete with voiceover.
• Paid social ads need an "upbeat, energetic" mood that signals a sale or launch without licensing headaches.
• Store background playlists (livestream selling, in-app video) need variety without paying per-track sync fees.

Lyria 3.5 generates a complete instrumental (or vocal, if you ask for lyrics) track from a text prompt in about a minute, at $0.09 per song — cheap enough to generate a handful of variations per campaign and pick the best one, which is exactly how you should use a generative music model: as a first draft, not a one-shot final asset.

A Real Generation: Prompt, Request, Response

Here's the exact prompt we used and the exact track it produced — nothing edited or cherry-picked out of a batch.

Prompt:

Upbeat modern electronic-pop instrumental for an e-commerce product ad, bright plucked synths, punchy four-on-the-floor beat, warm bass groove, optimistic and energetic mood, radio-ready polish, no vocals

Request (hiapi's unified async task API):

curl -s -X POST https://api.hiapi.ai/v1/tasks \
  -H "Authorization: Bearer $HIAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "lyria-3.5",
    "input": {
      "prompt": "Upbeat modern electronic-pop instrumental for an e-commerce product ad, bright plucked synths, punchy four-on-the-floor beat, warm bass groove, optimistic and energetic mood, radio-ready polish, no vocals"
    }
  }'

This returns a task ID immediately; music generation runs asynchronously, same as hiapi's image and video models.

Poll until terminal state:

curl -s https://api.hiapi.ai/v1/tasks/tk-hiapi-01M2HM4T0RG0XQNS6TFE9QF14V \
  -H "Authorization: Bearer $HIAPI_API_KEY"

Finished response (trimmed):

{
  "id": "tk-hiapi-01M2HM4T0RG0XQNS6TFE9QF14V",
  "status": "success",
  "output": [
    {
      "title": "Bright Shelf",
      "url": "https://temp.hiapi.ai/7c6ttvrbpt/01M2HM4T0RG0XQNS6TFE9QF14V-0.m4a"
    }
  ],
  "expireAt": 1790050272
}

Listen to the actual result here: Bright Shelf — lyria-3.5 output (.m4a). This link is a temporary output URL and expires roughly 7 days after generation — download and re-host the file yourself if you want to keep it. This is standard for every hiapi task output, audio included: fetch the bytes and store them in your own bucket immediately after the task completes.

The model titled the track "Bright Shelf" on its own — lyria-3.5 auto-generates a short title alongside the audio, which is a nice touch if you're organizing a music library for a campaign.

Python Equivalent

import requests
import time

API_KEY = "your-hiapi-key"
BASE = "https://api.hiapi.ai/v1"

resp = requests.post(f"{BASE}/tasks", headers={"Authorization": f"Bearer {API_KEY}"}, json={
    "model": "lyria-3.5",
    "input": {
        "prompt": "Upbeat modern electronic-pop instrumental for an e-commerce product ad, "
                  "bright plucked synths, punchy four-on-the-floor beat, warm bass groove, "
                  "optimistic and energetic mood, radio-ready polish, no vocals"
    }
})
task_id = resp.json()["id"]

while True:
    task = requests.get(f"{BASE}/tasks/{task_id}", headers={"Authorization": f"Bearer {API_KEY}"}).json()
    if task["status"] in ("success", "fail"):
        break
    time.sleep(5)

if task["status"] == "success":
    audio_url = task["output"][0]["url"]
    audio_bytes = requests.get(audio_url).content
    with open("bright_shelf.m4a", "wb") as f:
        f.write(audio_bytes)

Schema: What lyria-3.5 Actually Accepts

The schema is intentionally narrow — this is a text-to-music model, not a general audio toolkit:

• prompt (string) — required if you're not supplying lyrics directly; describes genre, instrumentation, tempo, mood.
• lyrics (string, optional) — pass your own lyrics if you want a vocal track instead of an instrumental.

There's no duration, image_urls, or reference_audio field. If you send an unsupported field or an empty payload, you'll get a validation error back rather than a silently-ignored parameter — worth checking the response the first time you integrate, since it tells you immediately if a field name is wrong.

On the "product images" mix-up specifically: if your team's brief said "generate product images with lyria-3.5," that instruction is based on a mismatched model. For actual product photography and ad visuals, use an image model — gpt-image-2 is hiapi's most-used text-to-image model and handles product shots and on-image text well. Pair it with lyria-3.5 for the video/ad soundtrack and you've covered both halves of a product video with two API calls instead of a stock-footage subscription.

Production Notes

• Treat every generation as a draft. At $0.09/song, generate 2-3 variants per prompt and pick the one that fits your edit, rather than trying to perfect a single prompt on the first try.
• Re-host before you publish. The temp.hiapi.ai output URL expires — don't link to it from a live storefront or ad; download and serve it from your own CDN.
• Keep prompts mood-and-instrumentation focused. "Upbeat," "warm bass," "no vocals," and a genre label (electronic-pop, lo-fi, acoustic) steer the output more reliably than trying to describe a specific melody.
• If you need lyrics or a longer runtime, hiapi also offers lyria-3-pro ($0.14/song) for full songs up to about three minutes with lyric and multilingual support — worth it once you're past quick ad beds and into branded theme music.

FAQ

Can lyria-3.5 generate product images for e-commerce listings? No. Lyria 3.5 is a text-to-music model with no image output. For product photos, use an image model like gpt-image-2 on hiapi.

How much does one lyria-3.5 track cost? $0.09 per generated song, billed per task regardless of how the track is ultimately used.

How long does generation take? Around a minute from task submission to a finished, downloadable track in our test run.

Can I get vocals instead of an instrumental? Yes — pass your own lyrics in the request instead of (or alongside) a mood/genre prompt.

Is the output royalty-free for commercial ads? The track is generated specifically for your prompt and isn't a licensed sample — treat it the same way you'd treat any AI-generated commercial asset from hiapi: it's yours to use in your ads once generated, but re-host it yourself since the delivery URL expires.

Related

• lyria-3.5 model page — live pricing, schema, and try-it playground.
• hiapi Pricing — compare lyria-3.5 against every other audio, image, and video model.
• hiapi API Docs — full /v1/tasks reference for async generation across all model types.
• minimax-music-3 API: curl & Python Guide — a second music model if you want a longer or differently-styled track.

Ready to try it? Grab an API key from your dashboard and generate your first track with the request above — it's a one-call, one-minute test.