Sureel AI

Excited to share that our research paper “Detecting and Mitigating Incoherent Input of Latent Diffusion Models” has been accepted to ICIP 2025! At Sureel, our mission is to make generative AI more responsible, transparent, and explainable. Understanding how diffusion models learn from the work of human creators is not only critical for fair attribution and recognition — it also helps us identify and mitigate the limitations of these models. In this paper, we explore how diffusion models respond to suboptimal input and propose a novel method for improving their robustness across modalities. Our framework boosts aesthetic scores by over 10% and is consistently preferred in human evaluations. Grateful to my co-authors Boğaç Ü., Diego Ponce De León Vera, Christoph Burgmair, and Dr. Tamay Aykut. Stay tuned and see you in Anchorage! #ICIP2025 #GenAI #LatentDiffusion #ResponsibleAI #ExplainableAI

✈️ Heading to LA for Digital Entertainment World on 22 May! I’ll be speaking on the panel “Opting In, Opting Out of AI Training.” Spoiler: the choice isn’t just yes or no. It’s a spectrum: from “never train,” through “yes with clear rules,” all the way to “full license with attribution and tracking.” My team and I have built tools that let creators pick any point on that spectrum. Got thoughts or challenges? DM me! Always happy to exchange ideas on explainable, particularly responsible AI and where it’s taking the media & entertainment world. See you in LA! 👋 (Full agenda link in the comments.) #DEW2025 #AI #MediaTech #SureelAI #Sureel #Sureel.ai #XAI #ExplainableAI #ResponsbileAI #EthicalAI #Rightsholders #Copyright cc Sureel AI

Jake Beaumont-Nesbitt (International Music Managers Forum Forum), Tamay Aykut (Sureel AI), and Joey Seiler (Granderson Des Rochers, LLP) answer questions on codifying, investment and values in relation to art, artists, IPs, and AI 💵🎸 #MusicBiz2025

Creators who embrace AI won’t just keep up, they’ll lead. In this clip, Benji Rogers explains why creators are leaning into AI, not to replace their work, but to scale it, protect it, and take ownership of their output. The message is clear for labels and AI companies: the future of music needs infrastructure that empowers creative control and AI-native workflows. Watch the full video and see where it’s headed: https://lnkd.in/gQR2Zcaa

We are happy to share Sureel.ai is partnering with BeatStars to do what regulation hasn’t: Give music creators control over AI training. All tracks on BeatStars are now shielded by default, and opt-outs are documented via Sureel’s API, public ledger, and robots.txt — making artists’ rights visible and enforceable. We’re building infrastructure for ethical AI at scale. Read more: https://lnkd.in/gRpuQS64

🔎 Just released on Water & Music: A brand-new, 4,500-word report on how music AI attribution actually works. -- The state of music AI today can be measured in the millions — millions of dollars raised, millions of users engaged, millions of AI-generated tracks uploaded to streaming services. As the market continues to grow at breathtaking speed, an urgent business question has emerged: Who deserves credit and compensation when AI generates music? The answer could determine how billions in potential revenue flow through the music economy in the coming years. The core concept underpinning the ideal of a more granular rights holder compensation framework for AI is ATTRIBUTION, or the linking of AI outputs back to the specific training inputs that influenced them. Unlike copyright or deepfake detection systems that simply flag similarity or infringement, attribution aims to establish causal relationships between inputs and outputs. For music, attribution can help answer questions like: ❓ Which songs in the training data most influenced this output? ❓ How strong was each influence, and how should we quantify it? ❓ What specific musical elements were borrowed or transformed? -- Yung Spielburg, Alexander Flores, and I spent the last several months on intensive research and interviews with teams forging new ground in music AI licensing and attribution deals, including Sureel AI, Musical AI, Soundverse AI, Lemonaide Music, LANDR, SourceAudio, Human Native, and more. A preview of some of the insights we covered: 💡 No "perfect" attribution system currently exists, but this hasn't stopped deals from happening. Perfection isn't a prerequisite for business. 💡 AI music generation is not sampling. Models don't copy fragments outright from training data, but rather absorb concepts on a more abstract level across multilayered architectures, making attribution exponentially more complex. 💡 Every attribution system makes inherent value judgments about which musical elements deserve (or don't deserve) compensation, potentially shaping revenue distribution for years to come. 💡 Tech alone won't solve the market coordination problem. Trust between rights holders and AI developers matters as much as technical innovation; a technically perfect solution is practically useless without industry adoption. -- Understanding the realities of attribution isn't merely a matter of technical curiosity — it's business critical for anyone working with music rights today, and will determine how the music industry adapts and thrives in an AI-centric future. Read our full analysis here (members only): https://lnkd.in/eBnHAEGn #MusicTech #AIMusic #MusicBusiness #MusicAI #MusicIndustry #Interpretability #Explainability #XAI

Your work is your digital gold—but AI companies treat it like free fuel. Protect, control, and monetize on your terms with APIs. See how Sureel.ai makes it possible. Watch now: https://lnkd.in/gQR2Zcaa