Prompt Share: the camera-move syntax that actually works in 2.5 (and the words it quietly ignores)

by @noahdavis_films in c/seedance-community · · Camera Move

Spent way too many credits mapping which camera vocabulary 2.5 actually responds to vs which it silently ignores. Sharing so you don't burn the same gens. These are directional tendencies, not guarantees, but they're consistent enough to rely on.

WORKS RELIABLY:
'dolly in / dolly out' — clean, distinct from zoom now (big 2.5 improvement)
'truck left / truck right' — lateral tracking, solid
'orbit around the subject' — works if the subject is clearly central
'crane up / crane down' — vertical, reliable
'low angle / high angle' as a STARTING framing — respected

FLAKY OR IGNORED:
'zoom' — it conflates with dolly, pick dolly if you want the move
'handheld' — sometimes adds shake, often ignored
'whip pan' — mostly ignored, too fast for its comfort zone
complex combos ('crane up while orbiting') — it picks ONE, usually the first

KEY INSIGHT: ONE camera verb per clip. It cannot do two simultaneous moves reliably, it'll pick one and fumble the other. If you need a compound move, generate the components and edit, or chain clips. Full tested list + example clips in the body.

402 points · 8 comments

Comments

  • @emmamartinez_studio · · 88 points

    'ONE camera verb per clip' is THE rule and it took me a hundred wasted gens to learn it the hard way. Every compound-move prompt is a coin flip on which move it honors and a guarantee it botches the other. Pick one, nail one. This should be pinned.

  • @jamessinghcreative · · 51 points

    The compound-move trap is seductive because real cinematography LOVES compound moves (the crane-and-orbit reveal is gorgeous). But the model can't, so you fake it in the edit or chain it. Decompose your dream shot into single-verb clips. It's the only way right now.

  • @oscarjones · · 36 points

    'Handheld sometimes adds shake, often ignored' — my experience exactly, and when it DOES work the shake is weirdly uniform and fake. If you want real handheld energy, add it in post with a camera-shake plugin. Don't make the model fake organic motion, it's bad at 'imperfect on purpose.'

  • @leomartinez_vfx5450 · · 27 points

    The model is bad at DELIBERATE imperfection in general — handheld shake, whip pans, 'messy' energy. It wants smooth. Add the chaos in post where you control it. Let the model do the clean move, you do the grit.

  • @sashanguyen_films · · 43 points

    Confirming 'zoom conflates with dolly' — and honestly you almost always WANTED dolly anyway. A true zoom (focal length change) is rarely what people mean, they mean 'move toward the subject,' which is a dolly. The model picking dolly is often correcting your vocabulary FOR you.

  • @taylorverma_art · · 33 points

    Real talk, 90% of 'zoom in' requests should've been 'dolly in' and the model quietly doing the right thing is a feature. A zoom flattens perspective, a dolly feels like approaching. People want the approach. Learn the difference and prompt it directly.

  • @hiroshimiller · · 29 points

    This is the most useful post on the sub this month, no notes. The 'quietly ignored' list alone saves everyone credits. Please keep testing the vocabulary as they update.