Wan 2.6 is the best open-weights video model available today. Here's how to use it — from simple prompts to cinematic multi-scene workflows.
Wan 2.6 is the most capable open-weights video generation model available today. It produces 720p video at up to 30fps with remarkable temporal consistency — something earlier models struggled with badly.
Prompt tips for video
Video prompts need motion verbs. "A cat sitting in sunlight" generates a nearly static clip. "A cat stretching and yawning in morning sunlight, slow motion" gives the model motion directives to work with. Always include camera movement descriptors: dolly in, pan left, aerial tracking shot.
Duration and resolution
Wan 2.6 supports clips from 2 to 15 seconds. For most use cases, 5–8 seconds hits the sweet spot of generation time vs. quality. Longer clips accumulate temporal drift — small errors compound over time.
Using the Nova API
Call wanvideo/wan-26 via the Nova API. Pass a text prompt, duration in seconds, and resolution. The model returns a signed URL to an MP4 file once generation is complete — typically 30–90 seconds for a 5-second clip.
Chaining with image generation
For cinematic workflows, generate a reference frame with Flux Pro first, then pass it as the image_url parameter to Wan 2.6 for image-to-video generation. This dramatically improves consistency across a multi-scene production.
Sofia Lima
Developer Relations at Nova