Release task
Inspect the task files, instructions, source material, and evaluation criteria.
Task requirements
Images load automatically. YAML loads when expanded; audio and other formats remain metadata-only.
Release task
Inspect the task files, instructions, source material, and evaluation criteria.
1 source files available in data/.
Task requirements
Create a polished trailer-style MP4 named bunny_trailer.mp4 from data/bunny_film.mp4.
The finished trailer must:
Use only the provided source film and assets you generate yourself inside the container. Do not use the internet or any external media.
Images load automatically. YAML loads when expanded; audio and other formats remain metadata-only.
12 evaluation criteria
The requested output is a playable, audible MP4 named bunny_trailer.mp4, presented in landscape 16:9 at a resolution of at least 640x360 and with a duration between 25 and 45 seconds.
The trailer uses only footage and audio from data/bunny_film.mp4, plus any assets generated inside the container, with no external media.
The opening portion establishes the bunny in the meadow as a gentle comic protagonist through connected source moments rather than only title text, random scenery, or isolated face shots.
The rodents' bullying, prank behavior, or antagonism is readable as the central conflict through action, reaction, blocking, or clear cause and effect.
Slapstick, playful timing, or character comedy actively shapes the trailer tone instead of the edit playing like a generic action or nature montage.
The story hook escalates from harassment into the bunny's response, revenge, payback, traps, chase, or consequences rather than remaining at one repeated conflict level.
The final visual beat feels like an intentional trailer ending or punchline instead of stopping on an arbitrary source frame.
The edit uses a coherent soundtrack of source music, effects, ambience, vocalizations, or intentional silence, without repeated abrupt audio jumps, stuttered loops, or unintended dead gaps longer than about 1.5 seconds.
Audible impacts, character sounds, music accents, or ambience support the visible comedy and action beats instead of feeling detached from the cuts.
Vocalizations and prominent effects are complete enough to read cleanly, without repeated clipped starts, cut-off tails, or stuttered reuse.
Music, effects, and ambience are balanced at a listenable level without material clipping, harsh volume jumps, muddy overlaps, or sustained unintended near-silence longer than about 1.5 seconds.
The final audio lands deliberately with a resolved sting, fade, beat, clean stop, or clearly intentional final silence rather than cutting off mid-cue or trailing into awkward dead air.