Upload AI-made sound to Freesound or Pixabay: tags and labels

Both sites accept AI-made audio, with conditions. Freesound's FAQ says AI generated audio "is permitted as long as it is marked with the GenAI tag and the model is indicated". Pixabay's terms require an upload to be marked with its "AI-generated" checkbox. Neither rule says anything about whether your tool's output may be uploaded; that is a separate question for the tool's own terms.

How do the two sites compare?

Their rules are short, so the table covers them all.

Where do you find the model name from a Sume job?

A Music Router job echoes the id you requested in job.model, so sume/music-auto stays sume/music-auto. The engine that actually ran is in job.request.routed_model, for example lyria-3.5, on both the router and the Music 1.0 routes. That is the value to type into a model field. Keep the whole job result with the file so you can show what produced it.

Is there a watermark to know about?

Google's Lyria 3.5 page says all generated audio includes a SynthID audio watermark, imperceptible to the human ear. Sume's Music Router runs Lyria 3.5 today, so a track from it may carry that mark; you do not add anything on top. It is a provenance signal, not a license.

What should you check before you upload?

Uploading publishes the file for others to reuse, so check these first:

• Whether your plan and the engine's terms let you publish the output under a public license.
• That the title, tags and description do not name a real artist or brand, which Pixabay prohibits as tags.
• That the prompt did not ask for a named artist's voice; Google says Lyria's safety filters block that request.
• That the model name you enter matches routed_model on the job.

Is it a good idea to upload generated sound at all?

It can be useful: a labelled, model-named upload gives others something to reuse and gives you a public record. It also commits you to the library's rules and means anyone can reuse the file under the library's license. If the track is a client deliverable, ask the client first. If it is a generic effect, a loop or an ambience, uploading is low risk once the checks above pass.

Keep the original job result and your prompt. If a moderator asks how the sound was made, you can answer with facts.

What does a Freesound uploader choose about AI training?

The FAQ describes a preferences panel with four choices: no additional preferences, Open Source Models only, Non-Commercial Open Source Models only, or No generative AI. If you upload a generated sound, think about which you want; you are publishing it into a library that others can use as training or reference material.