Kling 3.0 vs 4.0 on identical stills: the gap is whether the action finishes in 10 seconds
Same eight stills and prompts, one generation each: image quality is close, but 4.0 lands the action within the clip (train stops, girl turns, fox sneezes, condensation runs) while 3.0 runs out of time; neither nails the can's quarter turn.
This test runs Kling 3.0 and Kling 4.0 on the same eight stills with the same prompts, one generation each, nothing re-rolled. Image quality is close. Both hold the look the still sets, the flat cel style in the 2D test and the soft render on the 3D fox. The gap is in timing, whether the action written in the prompt actually finishes inside the 10 seconds. In the 2D test, 3.0's train is still coming in when the clip ends. 4.0's train pulls in and stops, and the girl turns to follow it. The fox sneezes late and small on 3.0, while 4.0 gives the puff and the shake of the ears. On the drink can, only 4.0 runs the bead of condensation down the side. For an editor that's the difference that counts. When the action lands around second 7 you have room to cut. When it's still happening at second 10 you either hold the shot or generate it again. Neither model gave the can its quarter turn, so on product work I'd still keep the move simple and get the turn from a second shot. Kling 3 is what I used on French Bee and Cocktail Vision. This makes me want to rerun one of those shots on 4.0.
This post quotes another post: “Kling 4.0 vs 3.0: eight image-to-video tests, same image and prompt”