Veo 4 vs Veo 3.1: A Decision Tree

Veo 4 is not on fal.ai yet. Veo 3.1 is, and it is the reference point you should use when planning work this quarter. Pricing sits at $0.40 per second at 1080p, with a fast tier near $0.25 per 1080p 8 second clip, and a hard 8 second ceiling at up to 4K. Assume Veo 4 keeps the same parameter names and roughly the same shape, and expect it to land in the same premium band. The real question is not which model is better. It is which one fits the shot you are about to render.

Start with the length. If you need more than 8 seconds of continuous action, neither model solves it in a single call. You are stitching. That pushes you into multi shot territory and the decision becomes about continuity budget, not per second cost. If you can fit inside 8 seconds, keep reading.

Next, ask what the clip is for. A social cut that runs once, gets a few thousand views, and is replaced next week does not earn a premium render. Use Veo 3.1 Fast. At roughly $0.25 per 1080p 8 second clip you can afford four variations and pick the best take. A hero piece on a landing page that will run for months is the opposite case. Every extra dollar of quality pays back in retention. That is where Veo 3.1 at full quality, or Veo 4 when it ships, earns its keep.

Now the tricky one. Camera motion. Veo 3.1 handles a single clear move well: a dolly in, a pan, a rise. It starts to struggle when you ask for a compound move, like a dolly combined with a tilt and a focus pull. If your storyboard calls for that kind of choreography, you have two options. Either simplify the move and rerender cheaply on 3.1, or wait for 4 and expect to pay the premium for the added stability. Do not try to force 3.1 into a move it is not built for. You will burn ten takes getting a mediocre result.

Here is the baseline call you will actually run. It works today and, assuming parameter parity, should work against the 4 endpoint when it lands.

1import { fal } from "@fal-ai/client";
2
3// or fal-ai/veo4/text-to-video once available
4const result = await fal.subscribe("fal-ai/veo3.1/text-to-video", {
5  input: {
6    prompt: "slow dolly in on a weathered lighthouse at dusk, fog rolling off the rocks, warm key light from the lamp cutting through cool ambient",
7    aspect_ratio: "16:9",
8    duration: "8s",
9    resolution: "1080p",
10    generate_audio: true
11  },
12  logs: true
13});
14
15console.log(result.data.video.url);

Notice the prompt names the move first, the subject second, then the time of day, the atmosphere, and the lighting. Veo responds to that order. It responds worse when you front load abstract mood words.

A few more branches worth writing down.

If the shot includes legible text in frame, pick the highest quality tier you can afford and expect to rerender twice. Text is still the weakest point across the Veo line. Veo 4 is expected to improve this, so for long horizon projects it may be worth delaying until the endpoint ships.

If the shot is a talking head with lip sync, use the full quality tier and lock your audio prompt carefully. The fast tier saves money but drops detail where you notice it most, which is the mouth.

If you are doing exploratory work, shot finding, storyboard validation, or style tests, stay on Fast. You are buying breadth, not polish. Ten fast renders beat two full quality ones when you do not yet know what you want.

If you are rendering the final version of a piece that will be graded and cut in a timeline, go full quality. The extra $1.20 per 8 second clip is rounding error against edit time.

So the decision tree reduces to four questions. Is the clip longer than 8 seconds. Is it disposable or hero. Is the camera move compound. Does it need legible text or lip sync. Answer those four and the model picks itself. Veo 4 will not change the tree. It will just move the quality ceiling up and the price along with it.