DoodleBook / docs /demo-video-script.md
Codex
Add detailed project paper and demo video script
bc7f386
|
Raw
History Blame Contribute Delete
2.93 kB

A newer version of the Gradio SDK is available: 6.20.0

Upgrade

DoodleBook Demo Video Script

Target length: 60–90 seconds
Recommended format: 1080p MP4, H.264 video, AAC audio

0:00–0:08 β€” The idea

Screen: Show the original child's doodle beside the DoodleBook title.

Narration:

What if a child's drawing could become the hero of its own story? DoodleBook turns one doodle into a narrated six-page picture book and a matching printable coloring book.

0:08–0:20 β€” Inputs

Screen: Upload the doodle, enter the hero name, select a theme, and select a narrator.

Narration:

The child uploads a drawing, names the hero, chooses one of ten meaningful themes, and selects a narrator. They can even upload a family voice for zero-shot voice cloning.

0:20–0:34 β€” Small model stack

Screen: Start generation, then briefly overlay the three model names.

Narration:

MiniCPM5-1B writes the story and plans every illustration. FLUX.2-klein-4B turns the doodle into one canonical hero and keeps that character consistent across all six pages. VoxCPM2 performs the narration. The complete deployed stack is only seven billion parameters.

0:34–0:48 β€” Audio and illustrated pages

Screen: Play a short narration excerpt, then scroll through three pages showing the same hero.

Narration:

Audio appears first, followed by the printable PDF and the illustrated pages. Theme-specific prompting gives each story a real emotional arc, natural dialogue, playful sound effects, and a lesson that follows from the hero's choices.

0:48–1:00 β€” Coloring book and downloads

Screen: Show the PDF download and matching coloring-book page.

Narration:

The coloring book is not a noisy edge trace. FLUX semantically redraws each finished page as clean line art, and a local cleanup pass makes it ready to print and color.

1:00–1:12 β€” Engineering and transparency

Screen: Open the trace panel and show seed, model information, and timings.

Narration:

DoodleBook runs publicly on one Hugging Face ZeroGPU Space. Its Open Trace exposes the seed, model pipeline, timings, and fallbacks, making every generation understandable and reproducible.

1:12–1:18 β€” Close

Screen: Show the finished book cover, live Space URL, and sponsor model logos or names.

Narration:

One drawing becomes a story a child can see, hear, print, keep, and make their own.

Recording Checklist

  • Use a simple, colorful doodle with distinctive features.
  • Pre-generate one successful book as a backup.
  • Keep waiting time out of the final edit; use a short time-lapse if needed.
  • Include at least one custom-voice example if consent is available.
  • Keep the original doodle visible when showing generated pages.
  • Show MiniCPM5-1B, VoxCPM2, FLUX.2-klein-4B, Hugging Face, and ZeroGPU by name.
  • End with the exact live URL: huggingface.co/spaces/build-small-hackathon/DoodleBook.