DANAV Suspense Series · öffentlicher Open Skill
DANAV Suspense Series
DANAV Suspense Series ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.
Series
4 2 6
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3821453/2026-09-25/b4hwsnh4.png
# DANAV Suspense Series
## When to use
- The user wants to write, continue, or produce an episode of the DANAV suspense mini-series (Vasu investigating his father's death).
- The user provides a one-liner, a draft, a script, or a character still for a DANAV episode.
- The user wants to re-render a clip, restyle a still, or mix an episode of this series.
## Series bible
Read `references/series-bible.md` before any episode work. It holds the locked characters, world, visual style, dialogue language, and the ongoing mystery. Do not invent facts that contradict it; if the user's request conflicts with the bible, flag it rather than silently changing canon.
## Locked look (preset assets)
The series' established character and location design is bound as preset assets. Reuse these URIs in `input_urls` for any clip that features them; do not regenerate or restyle them unless the user asks.
- `vasu` — Vasu (protagonist), Indian man mid-20s, dark t-shirt and jeans.
- `kabir` — ACP Kabir, Indian police officer in his 50s, khaki uniform.
- `vasu_room` — Vasu's bedroom at night, old wooden almirah, rain on the window.
- `police_station` — Indian police station office, metal desk, case-file wall.
New on-camera faces or a changed main space get a new `t2i` still (landscape, 720P); otherwise reuse these.
## Memory (cross-session)
- Before starting, call `memory_query` to recall the series state (episodes done, character stills, locked style).
- After finishing an episode, call `memory_store` (kind=`episode`) with the episode number, its clips, and any new facts revealed.
- Store character fine cards and the locked visual style once (kind=`project`).
- To correct/forget a stored memory, `memory_query` to find the id, then `memory_forget`.
## Production workflow (per episode)
1. Write the episode Spec (literary script): time-place, people, action, dialogue (Hindi, kept in source language), sound.
2. Slice the Spec into 7 ten-second clips; each clip needs a scene goal, dialogue, action, and sound.
3. Assemble `input_urls` from the preset assets above (characters + location for that clip). Generate a new character still (`generate_image`, model='t2i', resolution='720P', orientation='landscape', entity_type='characters', name=<character>) only for a new on-camera face; a new location still (entity_type='locations') only if the main space changed.
4. Render each clip with `generate_video(model='ref2v', input_urls=[character/location URIs], duration=10, orientation='portrait', resolution='480P', name=Ep{N}-Seg{K}, entity_type='storyboard_videos')`. One clip per step.
5. When all clips of the episode exist, `mix_clips(clips=[{videos:[seg1, seg2, ...]}], name=Ep{N}-Full, entity_type='final_videos', orientation='portrait')`.
6. Update memory and report the episode.
## Notes
- Keep dialogue and on-screen text in Hindi (source language); keep image/video prompts in English.
- This is a suspense thriller, not a romance: use standard cinematic MCU framing (chest-up) for characters; do not apply cleavage/exposure-tier defaults.
- Never invent facts that contradict the series bible.
- Report failures honestly; never fabricate URLs.
- Use `write_todos` to track the episode's clips and mix.