REMAKE OF VIDEO STYLE · skill Open Skill công khai
REMAKE OF VIDEO STYLE
REMAKE OF VIDEO STYLE là một Open Skill công khai trên Digen. Mở ra để xem workflow này tạo gì.
Other
26 3 11
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1463189/2026-09-29/rxxjxjze.jpg
# Video Remake & Character Replacement Recreates a target video — AI-generated or real footage — using the AI tools actually available in the current environment. The user picks one of two paths up front: - **Path A — Faithful / Tweak remake**: reproduce the original as closely as the tools allow (same shots, motion, lighting, style, audio intent), optionally applying the user's explicit change instructions (restyle, change setting/time of day, etc.). - **Path B — Character swap / replacement**: deconstruct the footage, map a custom character identity from the user's reference photo(s), and regenerate the sequence with that identity locked across all shots. **Use when:** the user supplies or describes a reference video/clip and wants it remade, remixed, restyled, matched shot-for-shot, restyled into any trending/current style (cinematic, anime, claymation, VHS, TikTok/FYP, etc.), or has its actors/subjects replaced with custom characters. **Do NOT use for:** restyling a single still image (use image editing directly); subtitle/translation-only jobs; recreating a real person's likeness to deceive (deepfakes) — decline those. # Required inputs 1. **Reference video** — file path, URL, or keyframes/screenshots. If none is provided, ask for one, or accept a detailed written description of every shot. 2. **Path** — A (faithful/tweak remake) or B (character swap). Ask if unclear. 3. **Tweak instructions** — the user's changes, quoted verbatim (Path A tweak only). This includes restyling into a named trending/current style. 4. **Character reference(s)** — reference photo(s) or detailed description of the replacement subject (Path B only). 5. **Constraints** — target duration, aspect ratio, resolution, language of any on-screen text or dialogue, and any tool/cost limits. # Workflow ## 0. Choose the path Ask the user to pick **Path A** or **Path B** before doing any work. If they only said "remake it," default to Path A. If they said "swap/replace the character/actor," use Path B. First use `write_todos` to list the workflow steps and update the whole list as you progress; call `set_guidance` to suggest next actions when a stage completes. ## 0b. Set your model preferences (user-defined) Before generating anything, ask the user which models/channels they want to use for this job, per modality: **video, image, audio/voiceover, music**. They can name specific models, name channels, or say "you pick" (use the built-in defaults). Verify any named model with `list_models`; if it isn't available, say so and offer the closest default. Record the choices and use them throughout — see `references/tool-mapping.md` for how "your pick" overrides the default. If the user names a different model mid-run, switch immediately (see the override rule in the reference file). This keeps the core principle intact: **remake any video into any style**, using whatever models the user decides on. ## 1. Intake and disambiguation - Confirm path and collect tweak/character instructions before doing any work. - If the reference is a file, extract keyframes (first frame, mid-frames, last frame, and every scene change). If you cannot process video directly, ask the user for 3–8 screenshots covering each distinct shot, plus duration and aspect ratio. - If the user provides a single grid/sheet image of keyframes, use `split_image` to separate it into individual frames first. - For a social link (YouTube/TikTok/IG): `download_video(url=...)` → pass the returned `downloadUrl` to `recognize_video` / `transcribe_audio`. For FB/X/LinkedIn: `fetch_video_info(url=...)` for metadata only. - State upfront what you can and cannot do with currently available tools — never promise video output if only image generation exists. ## 2. Analyze the reference (shot breakdown) Use `recognize_video(video_url=..., prompt=...)` to understand visuals/plot/shots/style; add `transcribe_audio(audio_url=...)` only when dialogue/lyrics/subtitles matter. For each shot, record: - **S