AI Video Prompting for Beginners: How to Avoid the Slop Look

AI video crossed a line recently: generating a clip is now trivial. Generating a watchable clip is not — which is why feeds are drowning in "AI slop": dreamy, drifting footage where cameras float aimlessly, faces morph mid-shot, and nothing ever quite happens.

Slop isn't a model problem. Like the generic image look, it's what you get when the prompt makes no decisions. Video just punishes indecision harder, because it has more dimensions to be vague in. Here's how beginners get out of the slop zone fast.

Start From the Image Ingredients

Everything from image prompting applies — subject, setting, style, light, mood, shot, detail (the full system: The Seven Ingredients). A video prompt without them produces slop with motion. But video adds three decisions images never asked of you:

The Three Video-Only Decisions

1. Camera movement — pick one, or lock it down

The signature of slop is the aimless drift: the camera floats because you didn't tell it not to. Decide explicitly: "static locked-off shot" (underrated — most real cinema is static), "slow push-in", "handheld follow", "drone pull-back reveal". One movement per clip. "Static shot" alone removes half the AI look instantly.

2. The action — one subject does one thing

Images depict; video happens. Slop happens when nothing is happening — or when five things half-happen. Give each clip a single, completable action with a beginning and end: "she looks up from the book and smiles", "the wave breaks over the rock", "he sets the cup down and leaves frame". If you can't say what happens in one sentence, the model certainly can't show it.

3. Duration thinking — write for seconds, not scenes

Current models generate seconds at a time. Prompting "a man's whole morning routine" forces the model to smear a story across a clip built for a moment — morphing guaranteed. Think in shots, like an editor: one beat per generation, then sequence your clips in an editor afterwards. The professionals treating AI video as footage to cut, not films to summon, are the ones producing watchable work.

The Slop Checklist

Before generating, your prompt should answer five questions:

• What is the one action? (beginning and end)
• What is the camera doing? (or explicitly not doing)
• What is the light? (video without specified light looks even more rendered than images do)
• What's the style reference? ("35mm film", "phone footage", "nature documentary" — video models respond powerfully to medium cues)
• Would this beat fit in under ten seconds?

Five yeses and you're out of the slop zone before you've spent a single generation.

Practise Where Feedback Is Instant

Video prompting is the most feedback-hungry skill in the AI creative stack — every ingredient interacts, and you only develop instincts by watching changes land. Cocoon's free Creative Lab teaches image and video prompting hands-on: build prompts from the ingredients, level by level, and see results as you learn. Free, browser-based, no signup.

Start there, think in shots, decide your camera. The tools are already good enough for watchable — the prompts mostly aren't, and yours are about to be. Enter the lab →