JSON or not, there is a clear limitation on what Veo3 can do. A scene with too many moving parts is simply too complex for it to consistently generate. Surprisingly though, Veo3 (Fast) seems to be following instructions more consistently than Veo3 (Quality). Here's a video of cat jockey racing on a dog. Prompt in the comment.

{ "shot": { "composition": "Epic wide-angle tracking shot with 28mm lens, shot on RED Komodo in 6K for dramatic sweep", "camera_motion": "high-speed follow cam with whip pans and low-angle shots to emphasize speed and intensity", "frame_rate": "48fps with occasional slow-mo inserts during intense overtakes", "film_grain": "cinematic grain with action-movie polish" }, "subject": { "description": "Two cats wearing tiny racing goggles and harnesses, clinging tightly as they ride two medium-sized dogs (a golden retriever and a French bulldog) tearing through the house like racehorses; each cat leans forward intensely, tails streaming like capes", "wardrobe": "tiny leather racing harnesses with color-coded scarves (red and blue), goggles, and faux walkie-talkies on their collars" }, "scene": { "location": "modern house turned into a makeshift racetrack with pillows, ramps, and cones marking tight corners", "time_of_day": "late afternoon with sunbeams cutting through blinds", "environment": "hallway straightaways, tight kitchen corner drift, dramatic stair ramp, and couch jump finale with scattered pet toys flying" }, "visual_details": { "action": "cats lean left and right as their canine mounts skid around corners; the French bulldog drifts under a table, while the golden retriever takes a risky shortcut over a couch; dramatic slow-mo as one cat nearly loses their grip, but regains it midair", "props": "makeshift racetrack with pet obstacles, cardboard boxes as fake audience stands, finish line made of string and catnip sprinkles" }, "cinematography": { "lighting": "strong directional sunlight with dramatic shadows, lens flares for extra action flair", "tone": "hyper-serious sports drama tone with a completely ridiculous subject" }, "audio": { "ambient": "wind rushing through house, jingling of dog tags", "sound_effects": "intense revving engine sound overlaid comedically, action drums, meows dubbed like anime battle cries" }, "color_palette": "Bold and contrasty—sleek blacks, racing reds and blues, glowing highlights for turbo vibe", "dialogue": { "character": "Cat Jockey #1", "line": "No one outlaps Whiskerblade on home turf.", "subtitles": true } }