“A muscular barbarian destroying a CRT television set with a weapon, cinematic, 8K, studio lighting”
“A frightened woman in Victorian outfit running through a forest, dolly shot”
Low angle static shot: a teddy bear sits on a picnic blanket in a park and eats a slice of pizza. The teddy bear is brown and fluffy, with a red bow tie, and the pizza slice is gooey with cheese and pepperoni. The sun sets and casts a golden glow over the scene”
“Aerial view of a small American town flooded with liquid cheese after a massive cheese rainstorm, during which liquid cheese rained down and dripped over the buildings”
We also added a new one: “A young woman doing a complex floor gymnastics exercise at the Olympics, involving running and somersaults.”
Weighing the results
Overall, the results shown above seem quite comparable to Gen-3 Alpha and Minimax video-01, and that's notable because HunyuanVideo is free to download, fine-tuned, and run locally in an uncensored manner (with the right hardware).
Of course there are some shortcomings. The vaudeville robots aren't animals, the cat drinks from a weird transparent beer can, and the guy eating spaghetti is clearly not Will Smith. There appears to be some celebrity censorship in the metadata/labeling of the training data, which is different from Kling and Minimax's AI video offerings. And yes, the gymnast has some anatomical problems.
