Skip to content
AI Primer
workflow

A creator shows how Seedance 2.5 can keep five cats in frame

Creators are using explicit stillness constraints and named positions in Seedance 2.5 prompts to control cat behavior and head counts. In one 15-second test, a five-cat scene repeatedly returned four until each cat had a distinct position.

3 min read
A creator shows how Seedance 2.5 can keep five cats in frame
A creator shows how Seedance 2.5 can keep five cats in frame

TL;DR

Water lands by a black cat's head in GlennHasABeard's 15-second test, which treats its refusal to flinch as the whole shot. A five-cat feather clip reserves the only paw lift for the closing beat; a bedtime scene exposes the stranger failure, two similar tuxedos collapsing into one. Dreamina's official guide lists reference media, duration, audio, and export settings as controls that vary by account and plan.

Stillness contracts

GlennHasABeard used one cutout reference for a 15-second Seedance 2.5 shot and wrote the behavioral instruction, “He blinks once, slowly, and does not move.”

The falling drop supplies the disturbance, but GlennHasABeard says the reference cannot teach the cat not to react, so that constraint belongs in the prompt. The same notes identify a 480p render followed by a Topaz upscale to 4K.

A five-cat motion budget

Adding four cats turned inactivity into the main constraint. GlennHasABeard bound five cat references by name, put one feather above them, and left the black cat with the sole action on the last beat.

The shot's allocation was explicit:

  • Five named cat references
  • One feather
  • One black cat lifting a paw at the end
  • Four cats watching without moving

“Four cats not moving is harder to get than four cats moving,” GlennHasABeard wrote. The clip makes the useful shift plain: motion gets assigned to a single actor, while stillness is assigned to everyone else.

Five heads and five positions

A separate 15-second bedtime scene repeatedly returned four cats before the creator changed the instruction and staging.

The repair was a compact set of count and layout constraints:

  • “Five separate bodies, five heads, never four, never six”
  • Five cat references, bound by name
  • A distinct position for each cat
  • A head count checked at every end state

Two tuxedos in the same frame were the specific trouble case: GlennHasABeard says they merged. The reader was also placed last, giving the camera a final subject after the count had held.

Gray-blob previs

For a different animation built with seven image references and four Blender previs clips, GlennHasABeard described a way to keep the blocking from becoming an unwanted reference.

His recipe is deliberately crude:

  • One smooth gray blob per character
  • No limbs
  • No textures
  • A prompt mapping each blob to its character reference, such as “the gray egg is the wizard from Image 1”

A detailed blockout gets preserved because Seedance copies the geometry it sees, GlennHasABeard explained; a simple blob is more readily replaced by the referenced character.

Share on X