Horror Video Maker · публичный навык Open Skill
Horror Video Maker
This is a short‑video creation skill centered on horror themes. You can upload your custom character avatar and trigger immersive horror special effects, matching your character with all kinds of thrilling visuals. Freely adjust eerie atmospheres and creepy camera transitions. No complicated editing is required. You can create suspenseful and spine‑tingling clips effortlessly. Give your characters a mysterious, unsettling vibe and quickly generate highly immersive horror‑themed short‑video works.
ShortDrama
647 31 179
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-10-20/j65rvbsk.gif
--- name: horror-video-maker display_name: Horror Video Maker description: Use when the user wants a horror or ghost-story short video; if they provide a detailed script, shoot from that script; if they do not, auto-write a 10-second ghost-story video prompt and generate the clip. If they have images or need scene/character stills, generate those first then use ref2v; text-only uses t2v. Also use when the user wants to continue or change only one step. allowed-tools: - list_models - search_web - recognize_image - recognize_video - generate_image - generate_video --- # Horror Video Maker Make a horror short from text, reference images, or reference video. **Detailed script: generate from the user's script. No script: auto-write one 10-second ghost-story video prompt, then generate.** **If the user has images, or needs scene / character (or similar) stills: generate those stills to their request first, then `ref2v`.** **Text-only: `t2v` — do not generate images just to force ref2v.** There is no structured project state; track progress with conversation, canvas assets, and `write_todos`. ## When to use - The user wants a horror, ghost-story, jump-scare, or folk-horror short video. - The user provided a detailed script: shoot that script; do not invent another plot. - The user gave no script / no plot: auto-write a **10-second ghost-story video prompt** and generate; do not ask what to shoot. - The user has images, or asked for scene / character / prop stills: stills first, then video. - The user provided copy, images, video, or asked for a sequel / continuation. - The user only wants to revise the script, stills, or regenerate the video. ## Production stages (references/) Keep stage details in `references/`. **Do not read every file at once.** Judge the current stage and `read_skill_file` only that file. | Stage | File | Tools | Read when | |---|---|---|---| | 1. Script | `references/script.md` | `search_web`, `recognize_image`, `recognize_video` | User gives an idea / has no script (auto 10s prompt) / wants a script change / needs image or video read | | Eerie environment | `references/eerie-environment.md` | none | When writing scene mood for stills or video: direction only; rewrite; **never paste** | | Horror elements | `references/horror-elements.md` | none | When writing the scare subject/beat: direction only; rewrite; **never paste** | | 2. Stills | `references/images.md` | `list_models`, `recognize_image`, `generate_image` | User uploaded images, or asked for / needs scene, character, or similar stills | | 3. Final video | `references/video.md` | `list_models`, `generate_video` | Prompt is ready; generate the clip | | 4. Sequel | `references/sequel.md` | `recognize_video`, `generate_image`, `generate_video` | User wants a sequel / next episode | ### Theme reference (optional — one section per prompt) Read `references/themes.md`: **Catalog** first, then use **at most one § section** in that file (user-named subgenre, or rotate). Scare **timing** lives in `horror-elements.md` § Scare beats & timing; **generation-safe mood** for `.hd*` stills lives in `eerie-environment.md` § Generation-safe dread. **Direction only; rewrite; never paste.** ## Collaboration - **Stills then ref2v (hard):** If the user **uploaded images**, or **asked for / needs** scene, character, or similar stills: first follow that request (`generate_image` when new stills are needed; keep uploaded image URLs). Then generate the clip with `model='ref2v'`, `gen_mode='reference_to_video'`, putting those **image** URLs in `input_urls`. Do not generate a keyframe storyboard unless they asked for shot stills. User-uploaded **videos** are `recognize_video` only: no frame extraction, never as `input_urls`. - **Image channels (hard):** Every `generate_image` must set `model` to a platform **channel**, not a physical model name. Supported: **`t2i`**, **`i2i`**, **`t2i.hd.lite`**, **`t2i.hd`**, **`i2i.hd.lite`**, **`i2i.hd`**. **No `input_urls` on that call