LONDON – Artist and researcher Felicity Hammond has implemented a production methodology termed a “human-AI feedback loop” to examine the intersection of generative technology and visual authorship.
The process moves beyond the standard linear application of text-to-image prompts, instead establishing a recursive cycle where the AI’s output directly informs the human’s subsequent creative decisions. This shift in workflow mirrors the iterative nature of professional art direction and conceptual development used in high-end visual production, and reflects a growing move in the creative industries to treat generative systems as collaborators rather than simple image engines.
Iterative Production Methodology
The feedback loop operates by treating the generative AI not as a tool for immediate execution, but as a collaborative partner within a structured studio workflow. Hammond initiates the process with a prompt, analyzes the resulting image, and uses the AI’s specific interpretations-including its errors and unexpected visual deviations-to refine the next set of instructions.
This cycle creates a dialogue between the human operator and the machine. Rather than seeking a predetermined outcome through precise prompting, the methodology allows the AI’s autonomous patterns to steer the aesthetic direction of the work, while the artist assumes responsibility for selection, sequencing and contextual framing of each iteration.
In practical terms, the loop can extend over dozens of cycles, with Hammond adjusting text prompts, training data constraints or compositional parameters in response to what the system produces. The resulting body of work is less a single “AI image” and more a time-based record of negotiation between human and model.
Impact on Visual Creative Workflows
The application of this loop challenges the industry perception of AI as a means of automation or cost-reduction in the creative pipeline. By prioritizing the “friction” between human intent and machine output, the process repositioned the role of the artist from a prompt-engineer to a curator and director who actively interprets, edits and, where necessary, rejects machine-generated material.
This approach aligns with broader industry discussions regarding co-authorship and the governance of intellectual property in generative media, now shaped in part by emerging case law and policy debates responding to large-scale training datasets and synthetic content. It also speaks directly to the question of what constitutes “substantial human contribution” in works involving AI, a standard that regulators and courts are beginning to test.
For cultural institutions, commissioners and brand clients, Hammond’s methodology offers a concrete model for crediting and compensating human creators in AI-supported projects. It suggests that value-and potential legal authorship-resides not only in initial prompting but in the sustained editorial judgment applied across multiple iterations.
The methodology emphasizes the following structural changes to the creative process:
- Transition from linear “input-output” models to recursive feedback cycles that can be documented and audited.
- Utilization of AI-generated anomalies as primary creative drivers rather than defects to be removed.
- Shift in authorship from the creation of the prompt to the curation and sequencing of the iteration, foregrounding editorial decision-making.
- Integration of explicit review stages where works are assessed for compliance with emerging AI-content guidelines from regulators such as the EU Artificial Intelligence Act.
The project is currently documented as a study in human-AI collaboration, and is being observed by curators, policy advisers and creative directors seeking practical frameworks for working with generative systems inside galleries, production houses and other institutional settings. As copyright offices and standards bodies refine their positions on AI-assisted authorship, methodologies like Hammond’s are likely to serve as early reference points for how responsibility, credit and control are allocated in hybrid human-machine artworks.
