Nearly 90 years after dazzling audiences with Technicolor, The Wizard of Oz is making history again, this time at Sphere in Las Vegas. The beloved classic has been transformed into a full-immersion sensory experience, filling the Sphere’s 160,000 sq ft, 16K dome with wind in your hair, scents in the air, and soft rumbles under your seat.

My shots covered wide shots of Dorothy and Glinda, Wicked Witch of the West material, Nikko the Flying Monkey material, Munchkin group dance sequences, and a Scarecrow close-up. The pipeline for each: segmenting characters from source footage, then upscaling in ComfyUI using ControlNet and a character-specific LoRA trained per subject to recover degraded detail. Keyframe prep for video generation was heavily Photoshop-based - rather than iterating on a single generation until something worked, we'd generate multiple image variants and composite the best elements together in Photoshop to build a single clean keyframe. That composite then fed into Veo for video generation, with color/lighting parity checks against source frames before handing off to comp.

Pipeline

1. Original footage processing: segment characters from source and isolate them from background for downstream generation.

2. Image upscaling and baseline correction: recover degraded subjects (especially Munchkin material) using ComfyUI with ControlNet and per-character LoRA, then run base color correction.

3. Video generation: generate shots from segmented and upscaled inputs using Veo, preserving original action timing and choreography.

4. Post-composition handoff: deliver generated material to composition for final integration.

5. Image detail optimization: perform large-volume still upscales, compare against source frames, and finalize lighting/color parity checks.

What I built

• Dorothy and Glinda wide-shot assets.

• Wicked Witch of the West character assets.

• Nikko the Flying Monkey assets.

• High-volume Munchkin character assets and dance-related shot material.

• One Scarecrow close-up sequence.

Challenge

The Munchkin dance sequences were the hardest part to generated motion had to stay on beat with the original choreography and music timing, while the upscaling still needed to hold character detail at Sphere resolution. Both constraints at once.