Generating a Bird-to-Dragon Video Locally Using scail2
EasternAd8821 · reddit · 2026-07-13
The author ran a local "bird→dragon" video generation experiment using scail2 to test the model's performance on the following challenges:
- Almost no real-world training data (dragons are mostly CGI)
- Complex motion (wings, waves)
- Long-duration generation and character consistency (about 30 seconds)
He modified a ComfyUI workflow to achieve this. Generating the 26-second video on an 5090 took 19 minutes, averaging roughly 45 seconds of generation time per minute of final footage. He also noted the use of 3 reference images + 1 pure background image, applying separate masks for the dragon, the landing rock, and both simultaneously; the driving motion came from a public video of a seagull.
The author concludes that while it hasn't reached the level of seedance2, it's quite impressive for "local generation," especially since the motion transfer significantly reduces the CGI feel. The biggest shortcoming currently is the water surface effect, and he plans to try wan2.2 low for further refinement or video super-resolution.
More from Embodied
- Nothing phone mockup turns a film joke into a modular design meme — ZeYanjie · 2026-07-22
- Lightwheel AI Launches SimReadyGen: Text-to-Physics-Accurate Robot Sim Assets — ZeYanjie · 2026-07-22
- Humanoid robot sorting packages in a warehouse sparks debate over job loss — MonaJalal_ · 2026-07-22
- NVIDIA pushes OpenUSD as the common layer for simulation and physical AI — MonaJalal_ · 2026-07-22
- A quadruped robot gets a custom glow-up with a new shell and screen — DynamicWebPaige · 2026-07-22
- A VR teleop demo for an SO-101 arm gets absurdly low latency by using one Python script — MoonL88537 · 2026-07-22