compositionAI-generated, human-edited
from the creatorBoth characters are invented, designed as reference sheets with GPT Image 2.5 and Seedream 5.0 Pro; no real person appears. The picture comes from three generations with Seedance 2.5 on BytePlus ModelArk at native 1080p (two of 30 s and one of 13 s), each one a single character cut between a few camera set-ups, with the timing of every line written into the prompt; the shot/reverse-shot ending intercuts two of them. The dialogue on screen is the model's own audio, in sync with the lips, and his voice from the first generation was the audio reference for the second. The interviewer's off-screen questions were voiced with ElevenLabs from a temporary clone of her own on-screen voice, deleted afterwards. The lullaby was composed separately with Stable Audio 2.5 from a text prompt. Edited, colour-matched and mixed by hand (-16 LUFS); English and Spanish subtitles written by hand, timed to the spoken lines and burned in.
toolkitAIMovies Labffmpeg