Dialogue Video Builder · skill pubblica Open Skill
Dialogue Video Builder
Dialogue Video Builder è una Open Skill pubblica su Digen. Aprila per vedere cosa crea questo workflow.
Film
4 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4156513/2026-09-28/r3v6yaz3.jpg
# Dialogue Video Builder Turns a dialogue script (like "THE EMPIRE BUILDER") into a ~30-second vertical (9:16) video with voiceover, using the two preset character reference photos as the two speakers. ## Preset assets - `messi_ref`: reference photo of speaker 1 (Messi). Use as `input_urls` for character/scene stills. - `suarez_ref`: reference photo of speaker 2 (Suárez). Use as `input_urls` for character/scene stills. ## When to use - The user pastes a dialogue script with speaker lines and camera/door directions and wants a short vertical video with voiceover. - The user wants to redo/advance one step of such a video (regenerate a clip, redo the voiceover, restyle a still, re-mix). ## Steps 1. First call `write_todos` to list the stages, then update it as you progress. 2. Parse the script into **scenes** (opening, each door, closing). For each scene, extract: the dialogue lines (speaker + text) and the camera/door direction (e.g. "closed door, text 'UE CORNELLA'"). 3. Generate a **scene still** for each scene with `generate_image` (`model='i2i'`, `orientation='portrait'`, `resolution='720P'`), passing the two preset refs in `input_urls` (`messi_ref`, `suarez_ref`) so both characters stay consistent. Prompt describes the scene: the two characters in the setting, plus the closed door with the club name text when the scene calls for it. Use `name='scene_<n>'`, `entity_type='storyboard_images'`. 4. Generate a **video clip** per scene with `generate_video` (`model='i2v'`, `orientation='portrait'`, `duration=5`), using that scene's still as `input_urls`. Prompt describes the subtle motion (camera zoom on a face, door staying closed, both characters talking). Use `name='clip_<n>'`, `entity_type='storyboard_videos'`. 5. Generate **voiceover** for each dialogue line with `generate_tts` (`model='tts.design'`). Use two consistent voices: Messi = calm, confident male, Argentine-accented; Suárez = deeper, warm male, Uruguayan-accented. Pass each line's text with the matching `voice_instruction`. Use `name='vo_<speaker>_<n>'`, `entity_type='voiceovers'`. 6. **Mix** all clips in script order with `mix_clips`, attaching each dialogue line's voiceover audio to the clip where it is spoken (use `audios` with `delay`/`start` so lines land on the right clip). Use `name='final_<title>'`, `entity_type='final_videos'`, `orientation='portrait'`. 7. Reply in natural language summarizing the final video; call `set_guidance` to suggest next steps (redo a clip, restyle a still, redo voiceover, re-mix). ## Notes - Always use `orientation='portrait'` (9:16) for stills, clips, and the final mix. - Keep the two character identities consistent by always passing both preset refs into `i2i` still generation. - If the user only wants to change one step (e.g. redo the voiceover or re-mix), do only that step and reuse the existing assets. - Report failures honestly; never fabricate URLs. Video generation does not support NSFW/adult content.