Hi everyone,
Back in June 2024 we published the Freesound in the era of generative Artificial Intelligence blog post in which we were discussing some thoughts about how Freesound and the community fits in the global context of Generative Artificial Intelligence (Gen AI). Since the publication of this blog post, a conversation has been happening in the forums in which many of you expressed your thoughts and in which we’ve been further explaining our position. From such discussion, it became clear that sound uploaders should have a way to indicate when sounds are created using Gen AI (to that end we made some modifications in the sound description form), and that sound uploaders should have a way to state under which circumstances they would allow their sounds to be used for the purpose of training Gen AI models (complementary to the CC license terms) .
Earlier this year, we introduced a prototype Gen AI preference option with the goal of giving sound uploaders a new tool for expressing their preferences regarding the use of their sounds for the specific case of training Gen AI models. After hearing your feedback and discussing about the available options (in that forum thread linked above), today we are officially releasing the “Generative AI preferences” panel that you’ll find in your Account Settings page.
Using that panel, Freesound sound uploaders can express their preferences regarding the training of Gen AI models using one of these options:
- No additional preferences: No additional statement, your sounds remain subject to the Creative Commons license terms selected for them.
- Open Source Models only: In addition to the Creative Commons license terms, you express a preference for your sounds to be used for the purpose of training Gen AI models only when the trained models are released as Open Source Models and made freely available to the public.
- Non-Commercial Open Source Models only: In addition to the Creative Commons license terms, you express a preference for your sounds to be used for the purpose of training Gen AI models only when the trained models are released as Open Source Models, made freely available to the public, and not trained in a commercial setting nor used for commercial purposes.
- No generative AI: You express a preference that your sounds not be used for the purpose of training Gen AI models.
In addition to that, an extra preference setting is provided to opt-out from Gen AI model training for the particular case of voice recordings. This option is designed to express a preference against the use of such sounds for training models that may generate speech that could be misused to impersonate real people.
When 3rd party model developers access Freesound data for training models, they can now filter according to uploader preferences so these are taken into account. To further facilitate the correct usage of sounds, we can also provide regularly updated Data Packs to 3rd parties which contain all that information.
You can find more detailed explanations about the Generative AI preferences and how these are applied in the Usage of my sounds for training generative AI models help section.
That’s it for today, thanks for reading!!!
frederic
on behalf of the Freesound team
Friendly greetings frederic and team, I admire your thoughtful and nuanced approach to ALL OF THIS. Instead of just shutting Gen AI out — letting individual creators and uploaders of these sounds choose how their work could be shaped and used… or at least strongly stating their own preferences. I reckon that is a deeply individual thing, and for those who are curious how they might get remixed, I hope there are more purpose-designed plugins that could be used on consensual training data sets and let a creator discover patterns and combinations within their own work and others who’ve opted-in. This’ll surely look brilliant in hindsight. Thank you, thank you for for this approach. It is so much better than just going nanana-ignorant, ‘cuz that’s not realistic at all.
Thank you all for being so thoughtful about this new world that we are in!
Or you know, just not have AI at all? It’s useless garbage for lazy people
Good stuff!
Does this mean slop sounds are now allowed to be uploaded here? This destroys the site as a resource. Hugely disappointing. I come here for real sounds, recorded by real people. Not glop. Any moron can generate glop for cheap.
RIP FreeSound, 20 years & dead.
I’m not sure if I missed it in the article, but does this apply to all old sounds uploaded prior to the panel update? Many of my uploaded sounds are voice samples, and I would want to opt out of them being used for AI training. Thanks!
If uploaders are indicating when sounds are made with gen AI, will downloaders have the option to filter out AI-generated sounds in search? Not knowing if a sound is AI generated on freesound turns this website from an extremely useful resource to something totally worthless.
Hi everyone,
Thanks for all the comments!
In resopnse to @Amœnus: “slop” sounds were always allowed, as long as they are labeled as being AI generated. Even though there are gen AI sounds in Freesound, we have not observed these flooding the site in anyway so we had no need to take counter measures. Remember that all uploaded sounds are manually moderated, so we would know if this becomes a problem and could act accordingly. For example by providing a filter to remove AI sounds from search results. Fortunately Freesound is not like other platforms in which users might have incentives to flood it with slop. Preserving the human spirit of Freesound is in the interest of everyone.
Answering to @Iceofdoom: yes, this applies to previously uploaded sounds.
I say screw ai! Freesound should be a haven free from such lazy and banal garbage. If you want ai sounds, get an Elevenlabs account or something. F*ck off – don’t upload it here!