Most of the time you spend on AI video is spent waiting to find out that an idea was wrong.

This is the turbo workflow with two settings changed. Steps down to four, and the seed box set to increment instead of randomise. You get a look at your idea in about a third of the time, and you can go back to any one you liked.

What the four-step pass actually tells you

I had this wrong at first, so it is worth being clear about. Four steps does not preview your final frame. I ran the same seed at four steps and at eight and compared them:

• Stays the same: the subject, the clothes, the weather, the light, the whole mood.

• Changes: the camera position, the buildings behind, where the traffic was.

So a fast pass tells you whether your prompt is going somewhere good. It does not tell you whether this take is the one. Judge the idea at four, then commit at eight and judge the shot.

The seed box is the part people skip

On randomise you cannot go back. You see something you liked, you run again, and it is gone for good. On increment the seed goes up by one each run, so every take has a number you can type back in.

The timings

Measured at 960 x 544, 124 frames, model already loaded, nothing else on the card:

• twenty steps, no turbo file: 113 s

• eight steps with it: 53 s

• four steps with it: 34 s

Your card will give you different numbers. The gap between them is the point, not the seconds.

What is in the file

Stock ComfyUI H3 nodes only. Nothing custom, nothing of mine you would have to install. Drop it in and it runs.

My other free workflows and the written guides are over here if you want them.