Self·E
Early DALL-E inpainting generated object and outfit variations from a single selfie; Runway ML frame interpolation blended them into one fluid motion portrait.
Demo
Process
The Selfie
The entire project starts from one photograph - a standard selfie. This single image serves as the foundation for every variation that follows. The name "Self-E" is a play on "selfie," reflecting both the source material and the AI-driven transformation process.
DALL-E Inpainting
Using the early DALL-E inpainting model, the original selfie was selectively masked and regenerated to produce distinct variations. Each generation introduced different objects, outfits, textures, and visual styles while maintaining the core portrait composition. The early model's raw, imperfect outputs became part of the project's aesthetic.
Runway ML Frame Interpolation
With all variations generated, the images were brought into Runway ML for frame interpolation. The AI model generated intermediate frames between each variation, creating smooth, fluid transitions that morph one version of the portrait into the next. The result is a continuous motion piece where identity shifts and reforms.
Variations
Context
Self·E was created during the early days of DALL-E, when AI image generation was still a raw and unpredictable tool. The project was an exploration of what these models could do with a human portrait as the starting point - not to create photorealistic fakes, but to use AI as a creative collaborator in reimagining the self. The imperfections and artifacts of the early model are part of the work's character, capturing a specific moment in AI's evolution as a creative tool.
Tech Stack
DALL-E
Early inpainting model used to generate object and outfit variations from the original selfie
Runway ML
Frame interpolation to generate smooth transitions between each AI-generated variation
Adobe Photoshop
Image preparation, masking, and compositing for the inpainting workflow
