It is no secret that Veo 3 is quite an expensive video model. If you are planning to use it to generate AI videos, you need to do your homework and optimize your prompts. A lot of folks are using JSON to create their prompts but as Google Labs explains on X, you don’t need to. You will just have to be specific with your setting, subject, action, lighting, shot type, camera angle, and audio. Here is an example:
Static 35 mm full-frame 16:9 shot reveals a vast pale-grey warehouse: plain back wall, concrete floor, grey rafters above. A cardboard box sits center frame. In one seamless motion the box bursts open; furniture erupts upward, sweeping out on fast, elastic, physics-true arcs. Graphite sofas, patterned rugs, tall shelves, and bronze floor lamps land neatly behind and around it, while a solid table thumps at the exact center of the layout. A cushion rebounds onto the sofa, a framed picture snaps to the wall, and a bronze pendant fixture swings from the rafters as dust sparkles over the newfound living room.
We’ve seen amazing Flow videos created using JSON prompts! It’s not the only or even “best” way, just one way to help structure hyper-specific visual directions. No JSON? No problem! Regardless of how you prompt, it helps to be specific with your setting, subject, action,… pic.twitter.com/ULD2rvvYiG
— Google Labs (@GoogleLabs) August 1, 2025
[HT]
