Creativity Support Lab
The Creativity Support Lab designs, develops, and evaluates creativity support tools: computational systems that draw out, sustain, and expand human creative practice. Our work takes a broad and ecosystemic view of creativity: artists, writers, designers, scientists, curators, readers, players, and everyday people all engage in creative processes, and these processes are shaped by a wide variety of infrastructures, tools and networks, both inside and outside the visible “creative moment”. Wherever possible, we seek generalizable knowledge about creative practice and how to support it – while also striving to ground our technical work in personal, sustained engagement with particular creative communities and domains.
We’re based at the Cornell Tech campus in NYC. Our recent work has been published at a range of highly selective HCI and AI conferences, including CHI, UIST, DIS, C&C, COLM, and EMNLP.
Research themes
Creative activity tracing involves the capture, representation, analysis, and sharing of traces that emerge from the creative process. Trace representations can be used to assess how tools shape creative practice; presented back to creative practitioners to facilitate reflection; transformed into a substrate for rationale capture; shared amongst practitioners to support social learning; and reasoned about by trace-aware co-creative tools to improve human-AI interaction dynamics. Trace comparisons can aid understanding of different strategies for design space exploration and even reveal structural similarities between diverse creative domains. Recently, we organized the Herding CATs workshop at CHI 2026, which brought together a large number of researchers to sketch out a roadmap for future creative activity tracing research.
New media authoring tools allow users to craft interactive narrative and play experiences, in particular radically open-ended or malleable experiences in which the full range of player behavior can be difficult for authors to anticipate. Past work on Dramamancer, Bonsai, and Elsewise has introduced new ways to extend human authorial control over LLM-based interactive storytelling; generative game design tools like Germinate and GBS have enabled the rapid and intentful exploration of interactive media design spaces; and computational narrative technologies like StoryAssembler, Felt, Winnow, and StU have probed the new forms of narrative authorship that can arise around the definition of flexible, playable interactive simulations.
Expressive interaction techniques make it easier to get ideas into the computer more naturally, across a wider range of modalities. Our work has found that evocative gestural inputs can complement prose-writing in the drafting of fiction; that malleable interactive visualizations can help worldbuilders comprehend, refine, and grow expansive fictional settings; and that making LLM output distributions and textual constraints directly manipulable can help poets discover simultaneously surprising and compelling bits of language. We’ve also experimented with multi-user multimodal canvases as spaces for collaborative AI-supported creation, recombinatory “mutant shopping” approaches to iterative text revision, and embodied “weird social” interactions in virtual reality creativity support environments.
Generating creative material for later human evaluation, curation, revision, and expansion can support new creative workflows – but it can also homogenize creative output, dilute human expressive intent, and generally reshape creative interaction dynamics. To generate better creative material, we’ve worked to finetune text generation models for improved diversity and stylistic personalization; to steer story generation processes toward wider-ranging expressions of the same story premise; and to construct generative pipelines that can effectively distribute intent across multiple different facets of generated output. We’ve also sought to characterize interaction patterns that might lead to stronger assertion of human control over creative processes that involve generated material.
Joining the lab
Potential PhD students can apply to Cornell Tech via the Information Science PhD program, listing Max Kreminski as a prospective advisor. Current Cornell Tech master’s students can also get involved; drop by Max’s office hours for more information. We don’t currently have the bandwidth to take on new undergraduate or pre-college research assistants.