GEMA v Suno ruling: what to check before AI music goes in an ad
A Munich court ruled against Suno on 31 July 2026, not final. What it says, what it leaves open, and a record to keep for any AI track you put in an ad.

Does the GEMA ruling mean you cannot use AI music in an ad? Not as a blanket rule, and nothing in the reports says so. On 31 July 2026 Munich Regional Court I ruled largely for GEMA against Suno, and the judgment is not yet final. It concerned a generator trained on six protected works and the songs it handed back to users. It does not say that every AI track is infringing, and it does not rule on any other vendor.
What it does change is the question a buyer should ask. Before an AI track goes under a paid ad, you want to know what the engine was trained on, what the vendor's terms say about your use, and which engine made your file. This post lays out what each source states as of 2026-10-03, what Sume's docs do and do not say, and the small record worth keeping.
What did the Munich court decide?
The case tracker reports that GEMA claimed copyright infringement over an AI music generator trained, it says, on six protected musical works obtained by stream-ripping from YouTube. The court found that storing protected songs in the model infringes the reproduction right, and that giving those songs to users through outputs infringes the right of making them available to the public. Suno must stop using the works without consent, give information, and pay damages in an amount still to be set.
Two limits matter. First, the tracker is a law firm's summary, not the judgment, so read this as reported. Second, it concerns Suno's model and six works. An ad team cannot read from it how any other engine would fare.
What do the vendors themselves say?
Vendors say different things about training and rights, and some say nothing. Only vendor pages are quoted below; the Suno row is press reporting.
| Source | Stated about training | Stated about your use |
|---|---|---|
| Stability AI, Stable Audio 3.0 (vendor page) | Licensed and Creative Commons material, attribution covering 1,278,902 recordings | You own the audio you generate and can commercialise it under the Community License; organisations above $1 million annual revenue need an Enterprise license that adds legal indemnification |
| Google DeepMind, Lyria 3.5 (vendor page) | Not stated on the page | Page does not specify licensing or commercial terms; it states tracks carry a SynthID watermark |
| Suno Speech (reported by The Decoder) | Reported as not disclosed | Not covered in the report |
| Sume Music Router docs | Names the engine it routes to: Lyria 3.5 today | Docs describe request fields and a fixed price per generation; they make no licence statement |
Where does Sume fit?
Sume's Music Router sends a request with model omitted, or set to sume/music-auto, to Lyria 3.5 today. You can also pin lyria-3.5 or lyria-3-pro. The docs say the job echoes the model you asked for and job.request.routed_model names the engine that ran. That gives you the engine, which is the first thing a rights review asks for.
It does not give you a licence opinion. Neither the Sume docs nor Google's Lyria page states commercial terms for the output, so do not treat a generated file as cleared because it came from a paid API. Read the terms that apply to your account and, for a paid campaign, ask your counsel.
What record should you keep per ad track?
Keep one record per track that ships. None of it needs special tooling:
- The exact prompt and the date you ran it.
- The engine:
job.request.routed_modelfrom the finished job. - The artifact URL on media.sume.com and a copy of the file you actually shipped.
- A saved copy of the vendor terms page you relied on, with the date you read it.
- Your own tag: the
metadatafield is stored on the job and is not sent to the provider, so put the campaign name there.
curl -X POST https://api.sume.com/v1/music-router/generate \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: autumn-ad-bed-001" \
-d '{
"model": "lyria-3.5",
"prompt": "Warm acoustic bed, 96 BPM, D major. Fingerpicked guitar, soft kick, light shaker. A 30-second track. Instrumental, no vocals.",
"metadata": {"campaign": "autumn-ad", "rights_review": "pending"}
}'Should you pin the engine?
Pinning lyria-3.5 rather than leaving the router on auto means a later change to what auto picks cannot move your track to a different engine. The price stays the fixed Music price per generation either way, per the router docs.
If you want the wider checklist for paid placements, the stored guides on Suno, ElevenLabs and Lyria for ads and pending Suno lawsuits cover other angles. This one only adds the Munich ruling and the per-track record.
A last point for agencies: the ruling is first-instance and not final, so the honest position with a client is that the law is moving. Say which engine made the track, what the vendor says about training, and that no Sume document grants a licence opinion. That is a shorter and safer conversation than claiming a track is cleared.
Sources
- CMS case tracker: GEMA v. Suno, 31 July 2026 (reported, read 2026-10-03)
- Stability AI: Stable Audio vs competitors, licensing and export rights (read 2026-10-03)
- Google DeepMind: Lyria (read 2026-10-03)
- The Decoder: Suno can now create spoken audio with matching background music (reported, read 2026-10-03)
- Music Router (Sume docs)
- Music 1.0 (Sume docs)
Related posts
More in Comparisons
- Pocket TTS voice cloning: a wav in, and what Sume does instead
Pocket TTS clones from a wav file you pass to --voice, with consent rules in its model card. Sume's API takes voice ids, not audio. Here is the difference.
- Schedule a weekly AI video: Sume Scheduled, Hermes cron or API
Three ways to run an AI video on a weekly clock with Sume: a dashboard schedule, a Hermes cron job, or plain cron calling a Format. What each can and cannot do.
- Which image and video models on Sume have open weights?
Sume's catalog is hosted. Where vendors publish weights for families Sume lists (Qwen-Image, MiniMax H3, FLUX.2) and why a null hugging_face_id proves nothing.
- Wix Stores promo video maker vs a custom AI product clip
Wix Stores can auto-make a promo video per product, free. When is a custom Sume image-to-video clip from your own photo worth the extra step? A side-by-side.
Written by Sume