2.2 KiB
FunPack Shot Prompt Planner
FunPack Shot Prompt Planner rewrites one transition-heavy prompt into ordered shot clauses before text encoding. It is meant to sit after a prompt enhancer and before the text encoder / Video Refiner path.
Why This Exists
Context windows can limit temporal memory, but they do not invent new semantic targets. If the conditioning remains one smooth paragraph, the model may still treat all concepts as one continuous scene.
This node detects transition language such as then, after this, cut to, scene change, transition to, next, and finally, then rewrites the text into explicit shot structure:
Shot 1: ...
cinematic cut, new shot, changed camera context.
Shot 2: ...
The structured_prompt output is the simplest path: connect it to your text encoder, then pass the encoded conditioning through FunPack Video Refiner.
The individual shot1 through shot10 outputs are for stronger context-window workflows where each shot is encoded separately and combined before a manual context window node with split_conds_to_windows enabled.
Recommended Workflow
Simple automatic path:
Prompt Enhancer -> Shot Prompt Planner structured_prompt -> text encoder -> Video Refiner -> sampler
Stronger context-window path:
Prompt Enhancer -> Shot Prompt Planner shot outputs -> text encoders / conditioning combine
MODEL -> Context Windows (Manual) with split_conds_to_windows enabled
combined conditioning + wrapped model -> sampler
Parameters
transition_enforcement
off: pass prompt through.auto: only structure prompts when transition triggers are detected.soft: always add shot structure and soft new-shot wording.hard: add hard new-shot wording that asks the model to drop previous scene context.
shot_count: auto uses detected clauses. A number forces that many shot slots.
transition_style: Default transition wording between shots.
camera_intensity: Adds subtle, medium, or strong camera/viewpoint changes per shot.
global_style: Persistent style or identity text appended to every shot.
negative_hints: Positive-side avoidance language for stiff outputs, such as static pose or locked camera.