Adobe Research’s new, intuitive paradigm for generative video honored at UIST 2026
Newly published work from Adobe researchers has been honored with a Best Paper Honorable Mention Award at this year’s ACM Symposium on User Interface Software and Technology (UIST). The paper, A Text-Native Interface for Generative Video Authoring, describes an experimental tool, and a new paradigm, that allows users to create and refine generative videos using text, all within a single document.
“Video is one of the most powerful ways to tell a story, and for a long time that power belonged mostly to big studios and production companies. With AI, that’s starting to change,” says Research Scientist Bruce Liu, who collaborated with Senior Principal Scientist Mira Dontcheva and Senior Research Scientist Ding Li on the project. “With this research, we wanted to find a really simple, minimal, and delightful way to make high-quality generative videos. I can’t wait to see the stories people will tell.”
The ideas behind Project Doki
The experimental new tool, which Adobe researchers codenamed Project Doki, was inspired by something we all learned to do in school: write a story in freeform text. Then the tool turns simple text into a video.
With this new framework, users define their story’s characters, setting, and visual style in a text document. Then, they write the story and add audio all in the same place. The tool provides in-line visual previews that users can adjust and a conversational AI agent if users want a collaborator. Once the story is ready, the tool compiles the text into video shots. From there, users can edit the video using text or the AI agent, refining until they’ve captured their vision.
One of the researchers’ main goals with this project was to explore intuitive ways for users to interact with, and control, generative AI.
“What excites me most is how natural it can feel. You write your story the way you’d write a document, and it becomes a film,” says Liu. “I hope the interaction ideas in this research help shape how creative tools work with generative video, so AI feels like a creative partner.”
So far, users are excited about the possibilities. As part of the research process, the team recruited ten people with a wide range of video experience to test the project. Over the course of a week, they created videos and rated the experience as excellent on the System Usability Scale that compiled quantitative feedback on their user experience. They reported that the tool’s text-native structure helped them move more quickly from idea to output and enhanced their understanding of narrative flow. And both novices and experts were enthusiastic about the new things they could do: novices liked the ability to create visual stories in a way they never could before, while experts felt that the tool improved their ideation and storyboarding processes
As Li explains, “The way people interact with this tool is really novel. I’m excited because we are already seeing how the same backend technology can power user experiences that delight novices and experts alike. From here, we’re looking forward to helping users generate better videos to support better storytelling.”
Taking Project Doki to UIST
The annual UIST conference brings together innovators from around the world to share research on human-computer interfaces. This year, Liu is excited to see the latest developments in AI image and video generation, especially advances in interactions and new ways of using the models.
Liu is also looking forward to sharing what he and the team have learned from their research. “I was thrilled when I learned that our work had received the best paper honorable mention,” he says.
“I felt the paper was solid and I really loved the work, but you never know whether reviewers will connect with your ideas. So it meant a lot to see them recognized. I’m also very grateful to my collaborators because this was truly a team effort.”
Wondering what else is happening in Adobe Research? Check out our latest news here.