Claire Yland Li / Selected practice
AI &
Creative Tech
Narrative meets emerging technology. Experiments in AI-assisted filmmaking, immersive worlds, and visual development.
Project Cradle
Created for the FutureVision XPRIZE.
Project Cradle
Feature film in development
Trailer · 1 min 42 sec
As Earth’s last fertile soils fail, a soil ecologist engineers dung beetles to restore agriculture, only to discover nature cannot be rebuilt one species at a time.
Read the synopsis
By the late twenty-second century, Earth’s fertile soils have collapsed. Humanity survives on synthetic foods, insect protein, and industrial ocean farming. But when industrial ocean farming begins to fail, civilization is forced back to the land in search of a future.
Soil ecologist Dr. Mira Lee joins Project Cradle, an ambitious global initiative dedicated to restoring Earth’s ecological foundations. Early restoration efforts repeatedly succeed in the laboratory but fail to sustain themselves in the wild. Searching for the missing ecological process, Mira revives an abandoned proposal to restore dung beetles.
The team successfully engineers dung beetles capable of surviving the hotter, drier climate of the twenty-second century. The first field deployments transform exhausted soils and ignite global optimism. Five years later, however, the populations begin disappearing, with no signs of a new generation emerging. As Mira investigates, the team discovers that while the engineered adults survive in the wild, their larvae cannot develop in modern livestock dung. Mira traces the missing conditions to an unexpected source: generations of dung beetles had depended on the nutrient-rich dung of elephants.
Project Cradle responds with its most ambitious undertaking yet: bringing elephants back from extinction. Although the program succeeds scientifically, sustaining even a small population demands immense amounts of water, land, and human resources, forcing Mira to confront the limits of rebuilding ecosystems through technology alone.
Then a routine survey of a long-abandoned marine reserve reveals vast blooms of salps drifting through recovering waters. The discovery forces Mira to reconsider everything Project Cradle has been trying to achieve. “Perhaps we overestimated the role we were meant to play,” she concludes. “Restoration is not about replacing every missing piece of nature, but about helping ecosystems become self-sustaining again.”
Solo creator
End-to-end trailer creation, from concept and visual development through shot generation and editing.
Nano Banana · Seedance · Kling
AI-assisted image and video creation, developed into a complete trailer.
Story & visual development → AI image / video generation → shot selection & iteration → edit & trailer assembly.
English Channel
An Indian boy moves to the US and learns English through TV channels. A 15-minute short film blending 2D animation, 3D animation, and live action, with AI-assisted workflows developed in ComfyUI and MiniMax.
Producer & AI Artist · August 2025–present
View two short animation tests · work in progress
Preview-resolution tests of character performance and shot continuity. Style and background consistency remain in development.
MiniMax / ComfyUI
From reference to refined shot
Reference-conditioned video generation, with separate draft, refinement, and finishing stages.
- 01
Reference inputs
Images, prompt, duration, and resolution; video and audio reference branches.
- 02
Draft generation
Reference conditioning and a first sampling pass establish the shot.
- 03
Latent upscale
A separate sampling pass increases visual detail from the draft latent.
- 04
Face refinement
Track, crop, refine, mask, and stitch faces; retain the audio branch.
- 05
RTX upscale
A final super-resolution pass before the video export.
Controls in the current graph
The graph localizes refinement to tracked face crops, adjusts denoise by face size, uses a feathered face mask, and stitches the refined crops back into the base frames. A dedicated audio-lock node is included in the refinement branch.
Separate the decisions
Draft generation, latent upscaling, face refinement, and final super-resolution are separate stages. This makes motion and composition review distinct from detail and finishing work.
Style & background consistency
Style and background drift remain unresolved. The next approach under exploration is a 360° scene layout to provide a shared spatial reference and better control the visual style and background across shots.
Status: research and testing.
Inspect the graph stages
- Reference conditioning
- MiniMaxH3ReferenceToVideo, image loaders, video loader, and audio loader.
- Draft sampling
- First Pass Sampling subgraph, sampler, scheduler, and seeded noise.
- Latent upscale
- MiniMax latent upscaler, AV latent separation / recombination, and a second sampler.
- Face refinement
- H3FaceTrackCrop, H3InjectVideoLatent, H3PerFrameDenoise, MiniMaxH3NativeAudioLock, H3FaceMaskSAM, and H3FaceStitch.
- Finishing
- RTXVideoSuperResolution and video export nodes.
Remnants
Narrative Lead
Best Use of AI & Immersive Media · Unreal Jam LA
Build with real LEGO bricks; an overhead camera detects their colors and connects the physical build to Unreal Engine.
As the bricks are placed, the matching virtual item—a gas mask—assembles on screen, turning hands-on play into a real-time interaction.
Pitch decks &
series bibles
AI-assisted concept visualization for pitch decks and series bibles, developed while working as a teaching assistant at USC School of Cinematic Arts.
Connecting story structure and visual language to communicate a project’s world.
View stories & pitch decks ↗