Considering Generative AI in Art and Content Creation

2 September 2026

Considering Generative AI in Art and Content Creation

A person works at a digital station in a vibrant, futuristic landscape with a large abstract human face, geometric shapes, and an AI symbol in the background.

Introduction

Sometimes, a technology appears poised to upend an industry. Generative AI is proving to be the next, and possibly the most significant, in a series of disruptions experienced by creatives from the online influencer to the Hollywood director.

A similar disruption occurred in the late 1920s, when recording technologies began replacing live bands and orchestras in bars and theaters and on radio throughout the world. Musicians were feeling a loss of control over their art, and the recording industry was not seen as adequately remunerating them for their labor. Then on August 1, 1942, the musicians of America decided to stop recording permanently, leading to one of the most significant labor actions of the 20th century that pioneered the principle that recorded music should generate ongoing payments beyond a one-time session fee. Before the 20th century, our understanding of music was that it was a live performance by human beings. Recording technologies fundamentally changed that understanding. Now, generative AI — trained in part on the past works of musicians, artists, filmmakers and more — marks another change in how we define and reward performing art.

It is impossible to predict how the impact of generative AI will compare to that of the recording industry, digital streaming, VFX, and countless other technologies that have simultaneously disrupted jobs and created new ones. This article explores some of the discourse between creatives on the impact of generative AI on their industry.

AI “Slop”?

Britannica defines “AI slop” as a “colloquial term for low-quality media content, including images, video, audio, and text, generated by artificial intelligence with seemingly little effort from a human interlocutor.” But is there a point at which content created using AI becomes art? Some point to the concept of craft or artistic intention: the act of meaningful human decision-making. This could mean a singer’s voice, the settings of an audio effect, the movement of a camera, or the choice of lighting. It does not necessarily preclude the use of AI in the creative process, as AI tools such as denoising or source separation are an extension of the creative palette already offered by modern production tools.

However, this nuance is not universally accepted, leading to the concept of contamination theory: that any amount of AI, be it for wind noise reduction to a movie created from a text prompt, irredeemably taints the work. The argument questions the way in which training data was obtained and that the original authors are not adequately paid for their contribution (more on this later). The pragmatist might agree that the application of AI tools lies on a gradient, and there are no binary absolutes. Like the disputes in the music recording industry of 1942, the question lies in how we measure the provenance of human effort and the ethics of using a machine to replace the job of a human being.

Misunderstandings About Generative AI?

One of the broad concerns about generative AI is the training of models on scraped, unlicensed data, consuming large amounts of energy in the process. While this has demonstrably been true, there is a growing industry that prides itself on creating models using only carefully curated licensed or original data. Artists are also able to commission custom models using their own training data as a component of traditional production pipelines. This may not redeem the industry for the creation and use of unethically sourced models, but it does represent an important step toward the ethical creation and deployment of AI models. Transparency becomes key: knowing where AI was used may help to build trust, and the legislation is starting to catch up. For example, the European Union has created a series of icons that help to identify images, audio, and video content that resemble existing persons, objects, places, entities, or events (deepfakes). The question of energy usage is a vast area in its own right, so we will limit this discussion to the perspectives of content creators.

Another area where the public may have been misled is the rebranding of traditional machine learning (ML) as AI for marketing purposes. While this has the advantage of checking the AI box for those who seek the label as a hallmark of cutting-edge technology, much of ML is built upon a relatively small amount of training data, with lots of hand-crafted features and hyperparameters. Consequently, the ethical discussion around generative AI does not largely apply to traditional ML, although it is understandably difficult for the general public to make the distinction.

Changing Creative Roles?

The industry is seeing a trend toward generalizing previously specialized roles due to AI tools that make adjacent parts of the production process more accessible. Directors, VFX artists, editors, and animators, for example, are now less well defined. Filmmakers, producers, assistants and more are now building their own tools using AI-assisted coding, closing the gap between developer and end user. Enabling the creative to build their own tools leads to tighter feedback between the tool and the trained eyes and ears of the artist than can be typically achieved with software developers and artists as specialists in their own domain. This may, however, come with significant downsides: generalists are typically paid less than specialists, they carry more cognitive load, and the consolidation of roles often means fewer jobs overall, not richer ones.

Even as AI tools for creative use cases mature, there remains a need for a human in the loop with an understanding of the craft. As anyone who has used an LLM to describe a concept with which they are familiar knows, spontaneous hallucinations can yield a very eloquent and convincing — yet ultimately incorrect — response. The same is true with AI tools for content creation, where image details can be spontaneously inserted or omitted, or different components from a single audio source can appear in multiple separated tracks. Coaxing a black box AI tool to do the task expected of it can be a source of frustration in a community where control over tunable and automatable parameters is a fundamental part of the workflow.

Regardless of the evolution of post-production roles, growing AI adoption has not been met with uniform enthusiasm across all creative communities. Writers, actors, and musicians, for example, have shown much greater resistance and skepticism, reflecting where creative control sits in the production chain. This again echoes the concerns that led to the recording embargo of 1942.

Just Another Tool?

The word “just” does a lot of work when describing one of the most disruptive technologies humanity has ever seen. A possible framing is that a tool ceases to be “just another tool” when it fundamentally changes the workflows that surround it, such as the introduction of digital filmmaking in a world of photographic film. Many also believe that generative AI belongs in a category of its own due to the power it has to impact people that didn’t think they would be impacted, the generalization by former specialists mentioned above being one such example. Automation of what was previously a specialist task can lead to what creatives are calling over-automation: automating for its own sake, not because it provides genuine utility. To quote Stanford professor Ge Wang, “What if the point of art is that we actually make it?”

Nevertheless, AI does have a place in automating some more repetitive tasks to enable creatives to spend more time engaging in creative work. As a case in point, there is growing demand for spatial music mixes derived from stereo masters, much of which lack the original stems (isolated, usually dry audio tracks). The expanded soundfield of spatial mixes makes the listener more unforgiving of bad edits and tuning issues that might hide in stereo mixes, creating a need for reliable source separation, denoising, and dereverberation tools. Additional artifacts introduced in the AI source separation process, such as the missing decay of a cymbal or the transient pluck of a bass guitar, as well as phase and “muddiness” issues, have become familiar to mixers experimenting with these tools. Again, there remains a strong need for the human in the loop who is familiar with the sound of high-quality stems.

The ultimate balance is found when the tools complement human artistic intention instead of replacing it.

Authorship and Attribution?

A debate exists around whether generative AI, trained on the creativity of human beings, should be granted or considered capable of “authorship”, in the sense that we mean it today, at all. Transparent research is built upon the practice of citing the work of others, whereas the dependence of copyrighted training data on today’s generative AI models is far more opaque.

The question of copyright in AI-assisted works is generally unresolved in law. Parallels are often drawn with existing practices, but in most cases the ownership and compensation of material have established legal frameworks. For example, the process of producing a cover song is fundamentally different, as legal mechanisms exist for crediting and compensating the authors of the source material. Consider the singer-songwriter who cannot afford a session musician and uses a generative AI bassist in the same way as a virtual drummer in an audio workstation; the original author retains copyright to the final work, but how do we compensate the musicians who provided the training data used to train the AI musician? The creation of the mechanisms necessary to ensure fair and equitable attribution to the artists whose work was used as training data is one of the most challenging legal questions of our time.

Advice for the Next Generation

Be it performance, post-production, directing, photography, or numerous other creative roles, a philosophy to live by is to strive to understand the medium you are passionate about and to learn the fundamentals from those who have gone before you. This process involves a great deal of trial and error as you learn to work within the limitations of the medium, training your eyes or your ears to know what is wanted. It is through this process that great artists have produced their most influential and enduring works. Like the invention of the camera not making oil paintings redundant, it is the ways in which the medium deviates from reality that can make it endearing. But we must be careful of gatekeeping: access to training and the freedom to go through the pain of learning a skill are themselves not equally distributed, so we should not automatically discount works produced with the assistance of AI as illegitimate as a matter of principle.

Artists, like engineers, can derive a lot of identity from approaching their craft with specific tools and processes. Generative AI should be approached with the same curiosity and skepticism as any other tool, used where it is found to be effective, and not when the human in the loop finds that it distracts from the creative vision. AI tooling is likely to change very quickly, and creatives should try to avoid becoming attached to one tool.

Be it art or technology, people sometimes don’t know what they want until it’s presented to them. Generative AI, by its nature, is a machine for generating a statistical average. Exceptional art and technology are not average: Nobody in 1990 knew they needed a smartphone, and nobody in 1930 knew of the power of rock and roll. Ask a person in 1880 how to make their lives easier, and they might respond by asking, as put by Henry Ford, for faster horses. If the ultimate conclusion is a hyper-personalized world in which generative AI provides exclusively what is asked of it, we risk losing serendipity, exposure to new work, and the incentive for innovation. As we navigate increasing AI-adoption, human creativity will remain as important as ever.

Conclusion

This article may have raised more questions than it has answered, and that was the point. Like any optimization, the challenge lies more in formulating the problem than the solution itself. The difficulty is that we lack precedent; we can look to examples of disruptive technologies affecting the creative industry in the past, but their long-term effects are all, by their nature, difficult to predict. The legal and ethical frameworks may take decades to catch up with the technologies they are supposed to regulate, and in hindsight they may seem obvious. It’s tempting to think of the present as the conclusion to everything that went before it, and not another moment in human history. What is clear is that generative AI has the potential to be an extraordinarily disruptive technology, and no technology has yet quelled the human desire to create art.

Publications & Resources

Conferences & Events

Education & Training

Community & Involvement

Career & Industry

About IEEE SPS

For Volunteers