Giga-World-1 · Controllable Image-to-Video (nano / 1.3B)
Giga-World-1 is an open Wan2.1-FunControl based interactive world model. Give it a first frame, a control video (a camera / motion trajectory), and a text prompt, and it rolls out a future video that follows the control signal.
The reference data is a robotic-manipulation 3-view layout (front + left/right wrist cams), so the bundled example produces a wide, three-panel robot rollout. Use the example below to reproduce the authors' sample, or bring your own first frame + control video pair.
⚠️ This is a heavy video model running on ZeroGPU. Generation of a single chunk takes a while; increase frames only if you have patience for longer runs.
33 99
10 40
1 8
256 480
832 1920
0 2147483647
Examples
| First frame (input image) | Control video (camera / motion trajectory) | Prompt | Number of frames | Inference steps | Guidance scale | Height | Width | Seed |
|---|