









A ComfyUI workflow designed for rapid multi-angle scene generation greatly facilitated the creation of keyframes for AI video production.
Concept made in 2024. Generated base assets with AI, photobashed the compositions, and repeatedly fed them back into the AI pipeline for fine-tuning.

Achieved control over forms, materials, and styles through an iterative, manually-guided AI workflow.
A sequence from the early 2026 1-minute, continuous one-take AI animation, produced using a multi-tool pipeline combining Seedance 2.0, ComfyUI (Qwen), Nano Banana, and Adobe.

Made 3d models with Tripo for consistency.
For the final 4-minute continuous one-take sequence, I created 3D previs in Blender to guide the camera motion and framing, then used it as a structural reference for Seedance 2.5 video generation.