Kling 4.0 hands-on: physics holds, cut timing slips
Three 15s 1080p text-to-video tests on Higgsfield: heavy water and mechanical physics hold with no morphing, one take runs 15s uncut, but requested cuts land early, macro never settles, and dolly-zoom is refused.
Kling 4.0 shipped at 1080p for roughly 1 credit per second on @higgsfield for promotion time i ran three tests on it Identical settings on all three, so the prompt is the only variable: 15s each, 1080p, text-to-video, no reference images. Results: Test 1 — environmental physics Storm surge detonating on a concrete sea wall. 15s, 3 shots. Question: does heavy water carry real mass, or dissolve into particles? It carried it. Gravity-bound drainage, no morphing, and debris that landed stayed landed across the cut. Test 2 — camera discipline A gantry crane lowering a container. 15s, one unbroken take. Question: does it hold a single take, or hide a cut when the move gets hard? Zero cuts in 15 seconds. The load swung with real momentum. It did ignore the dolly-zoom and push in physically instead. Test 3 — macro physics, vertical One droplet into a still pool. 15s, 2 shots, 9:16. Question: how far down the scale does the physics hold? Rebound column, secondary droplet, real surface tension. Excellent in vertical. Two misses: never fully settled; a stray droplet spawned at 0:03. Three findings that held across all three: Cuts landed early every time. i asked 5s / 10.5s / 8s, got roughly 4s / 10s / 6s. It compresses your timing. Audio generated by the model Dolly-zoom is the one move it refused. Where it's strong: unbroken takes, heavy water and mechanical physics, object continuity across a cut. Where it slips: precise cut timing, macro settling, compound camera moves. Next: reference-to-video with elements, and editing an existing clip. Prompts for all three tests below.