Alibaba’s latest video model can generate up to 30 seconds of 1080p video in various aspect ratios.
It can do 480p at $0.05 per second. I tested this out using a Jean Phil video which is all over TikTok and Ideogram right now.
I used GPT Image 2.5 Flare to prepare a single image to set the scene. This is currently my go to image editing model (image to image). Well that and it Sunburst which is a slower, more expensive but higher quality model from the same family.
I found a photo of the entrance of JFK Park in Lima and took a screen grab from one of Jean Phil’s videos. Prompt was “Use the backgroundf of #Image1 but add the guy in the #Image2”.
Then over to Wan 3.0. Load the image from GPT and an original Jean Phil video I want to use. Only catch here was the reference video can only be up to 15 seconds. The prompt was: “Create a video using the #Image1 image as the setting with the unusually dressed man. Use the motion from the video to drive the action. Follow the movement in video 1 as close as possible. The people in the background of #image1 move naturally, some glance at the man but otherwise ignore him as they carry on their day.”
Cost was $0.50, generation time 4min 30s.
