Cinematic Shorts Studio · skill Open Skill công khai
Cinematic Shorts Studio
Cinematic Shorts Studio là một Open Skill công khai trên Digen. Mở ra để xem workflow này tạo gì.
Film
18 3 5
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4247767/2026-10-10/jbzpjd7r.jpg
# Cinematic Shorts Studio Turn a single idea into a short, filmic, atmospheric video clip or micro-short with a **signature cinematic language**. Every short is built around ONE strong visual motif and a consistent moody grade, so the result feels like a frame from a real film rather than a generic clip. ## Signature look (apply to every short unless the user says otherwise) - **Grade**: moody, desaturated with a teal-orange or cold-blue tint; soft shadows, gentle film grain, subtle vignette. - **Lighting**: motivated light (window light, neon, headlights, candlelight) — never flat. - **Camera**: slow push-in or slow drift; shallow depth of field; handheld micro-shake for tension. - **Motif rule**: pick ONE recurring visual motif (rain on glass, a red object in a cold scene, a swinging light, drifting smoke, a single bird) and thread it through the short. - **Pacing**: one idea per shot; hold the frame; let the mood breathe. ## When to use - User wants a cinematic short clip / micro-short from an idea, one-liner, reference image, or character description. - User wants to redo/refine the last short, change one detail (subject, setting, mood, grade, duration), or mix clips together. - User wants a signature "Neo-Noir" / moody atmospheric short. ## Steps 1. **Clarify (one short round if unclear)**: the core idea or subject, the mood (e.g. lonely, tense, dreamy, noir), orientation (`portrait` 9:16 default for shorts / `landscape` 16:9 if requested), and whether they want a single clip or a 2-4 shot mini-sequence. If they only say "make it cinematic," apply the signature look above. 2. **Recall preferences**: call `memory_query` for saved style/mood/character preferences (e.g. "preferred cinematic mood") and apply them; if nothing is found, proceed with the current request. 3. **Plan the shot(s)** (write_todos for multi-shot work): define the motif, the grade, and each shot's subject/action/camera. For a mini-sequence, list 2-4 shots. 4. **Build the visual** with `generate_image`: - **No source image:** `model='t2i'`, prompt in English describing the subject, setting, lighting, grade, and the motif. Set `orientation` and `resolution` (default `720P`). Pass `name` = shot title, `entity_type` = `cinematic_frames`. - **From a reference image / character:** `model='i2i'`, `input_urls=[<asset_id>]`, prompt describing the scene and grade. Pass `name` and `entity_type` = `cinematic_frames`. 5. **Animate** with `generate_video`: - **Single image → video:** `model='i2v'`, `input_urls=[<frame asset_id>]`, prompt describing the camera move and micro-action (e.g. "slow push-in, rain streaks on glass, a single red umbrella drifts past"). Set `orientation` to match, `resolution` (default `720P`), `duration` (default 5s). Pass `name` = shot title, `entity_type` = `cinematic_clips`. - **Consistent character across shots:** `model='ref2v'`, `input_urls=[<character/ref asset_id(s)>]`, prompt describing the shot. Pass `name` and `entity_type` = `cinematic_clips`. 6. **Optional sound**: add a score with `generate_music` (prompt describing the mood, e.g. "slow ambient piano, sparse, melancholic, cinematic") and/or a voiceover with `generate_tts` (`model='tts.design'`, `voice_instruction` in English describing the voice). Pass `name` and `entity_type` = `cinematic_audio`. 7. **Mix (for mini-sequences)**: stitch clips with `mix_clips` (`clip-mixer`), ordering the shots, adding the music/voiceover tracks, and optionally burning subtitles. Pass `name` = final cut title, `entity_type` = `final_videos`, `orientation` matching the clips. 8. **Reply** in natural language summarizing the short (motif, grade, shots, runtime); don't paste raw URLs. If it isn't what they wanted, offer to redo or tweak one detail. 9. **Remember preferences** (only when the user states a lasting preference): call `memory_store` with `kind='preference'` for things like "prefers cold-blue noir grade", "recurring character: [description]", so future sessions reuse them.