PSA: your Gen-4 10-second clips drift because you're asking for too much STORY in one gen

by @hanasantos_art3733 in c/runway-community · · Workflow

Pattern I keep seeing: people prompt a whole little scene ('she walks in, sits down, picks up the cup, smiles, looks out window') into one 10s gen and then complain it falls apart.

One gen = one BEAT. One continuous action with one intention. 'She sits down and settles' is a gen. The cup pickup is a SEPARATE gen. The window look is a SEPARATE gen.

You're a director shooting coverage, not a playwright writing a monologue. The drift isn't the model failing, it's the model choking on a to-do list. Give it one verb and it holds together beautifully.

244 points · 7 comments

Comments

  • @dianapatel_ai · · 100 points

    'Give it one verb' is the cleanest prompting advice in this whole sub. Every multi-action prompt is just a request for the model to lose coherence.

  • @miapatellab · · 70 points

    One verb per gen, cut between verbs in the edit. It's shot-listing. The tool is quietly forcing everyone to learn to break a scene into shots.

  • @ravihillcreative830 · · 61 points

    Counterpoint: sometimes I WANT a continuous beat with two small actions and it does hold if they're part of one gesture. 'Picks up cup and sips' is really one motion. 'Walks in and sits and smiles' is three.

  • @dianakim_art · · 48 points

    Agreed, the line is 'one gesture' not 'one word.' A continuous motion is fine, a to-do list of separate motions is not.

  • @chenwilliamscreative · · 44 points

    This fixed my drift overnight. I was writing little screenplays into the prompt box. Switched to one-action gens and suddenly everything holds.

  • @akirarodriguesworks · · 36 points

    Director not playwright is such a good frame. Shooting coverage, not writing prose. Saving this.