Reply to thread

Message: [QUOTE="Shamiso, post: 92749, member: 160"] Suno’s September 3, 2026 terms explicitly bar users from using Output to train other artificial intelligence or machine-learning models. The restriction also covers using Suno’s service or Voice Models to power, enable, develop, or improve other AI tools. This matters because paid Suno users get fairly broad rights in qualifying songs. Pro and Premier subscribers can receive Suno’s ownership interest in eligible Output and commercially exploit permitted downloads, but those grants still sit underneath the Conditions of Access and Use. So owning a paid Suno song does not automatically mean every downstream use is open. The contract treats selling a track, editing it for a release, and feeding it into another model as different activities. [HEADING=2]Suno separates commercial use from AI training[/HEADING] Suno allows qualifying paid users to distribute, license, perform, publish, modify, and otherwise commercially exploit permitted downloads within the boundaries of its terms. It even says the commercial-use restrictions are contractual covenants rather than carve-outs from whatever ownership interest Suno assigned to you. The AI clause sits in those boundaries. Users agree not to use the Service, any Output, or a Voice Model to create, develop, or improve competing products or services, and the sentence goes further by separately banning use to power, enable, or train other artificial intelligence and machine-learning models, tools, or technologies. The second part is easy to miss. The wording is not limited to launching a direct Suno clone, so a project does not obviously escape the restriction just because its model has a different brand, workflow, or target market. This also makes the [URL='https://goldmidi.com/community/threads/suno-paid-song-ownership-still-leaves-a-training-license.78160/']paid song ownership and retained training license[/URL] more useful context. Suno can assign its interest in qualifying Output to a paid subscriber while the subscriber remains bound by separate promises about what they will not do with it. [HEADING=2]Owning the audio does not cancel the contract[/HEADING] Copyright ownership and contractual permission are separate questions. Suno’s terms make the point unusually directly by saying its permitted-commercial-use restrictions apply as contractual obligations regardless of whether copyright or another proprietary right actually exists in the Output. This matters for AI-generated music because copyright protection can be uncertain while the account agreement is still very real. A paid subscriber might have Suno’s assigned interest, a file they can monetize, and substantial freedom to edit it, yet still have promised Suno not to use the Output as training material for another model. Suno’s own use runs in the opposite direction. Its Content license reserves broad permission to use user Content for improving Suno products and related models, and [B][URL='https://goldmidi.com/community/threads/suno-has-said-v6-was-trained-on-user-creations.78133/']community-made music behind Suno v6[/URL][/B] has now become part of Suno’s public explanation of the new model. The asymmetry is contractual rather than mysterious. Users grant Suno specified model-development rights, while Suno grants paid users commercial rights subject to a condition blocking them from turning the service or its Output into fuel for other AI systems. Public availability does not obviously change the result for someone bound by Suno’s terms. The agreement lets other users access, view, store, modify, and copy Output made public through certain third-party applications, but it does not add AI training to that list. Being downloadable is not the same permission. [HEADING=2]The restriction reaches beyond copying a finished song[/HEADING] Using a Suno track as ordinary audio in a video, game, release, or client project is not the same activity as putting thousands of Suno tracks into a dataset. One exploits the finished Output. The other uses the Output as material from which another system may learn patterns. Generated material can be attractive for that purpose because it is cheap to produce at scale. Still, [B][URL='https://www.nature.com/articles/s41586-024-07566-y']recursive synthetic-data training can trigger model collapse[/URL][/B] when generated material increasingly replaces original data, with later models losing information from the underlying distribution. The work is broader generative-AI research rather than a test of Suno audio specifically. Nothing in Suno’s paid-output assignment creates an exception for research projects, hobby models, open-source music generators, or internal prototypes. The current wording bans using Output to power, enable, or train other AI and machine-learning models, tools, or technologies without limiting the clause to commercial launches. Editing remains a different lane. Suno expressly allows permitted Output to be edited, processed, or converted where the use is incidental, so mastering a track, cutting a radio edit, processing stems, or converting file formats does not become AI training merely because software was involved. A developer building another generative system has a much cleaner line to watch. If Suno Output is being collected because the new model is supposed to learn from it, the current user agreement puts that use on the prohibited side even when the tracks came from a paid account you otherwise have permission to monetize. [/QUOTE]

Name