Cinematic Character Animator · সর্বজনীন Open Skill
Cinematic Character Animator
Create professional cinematic dance videos from a simple idea, reference image, or music concept. The skill automatically develops the concept into a complete short dance performance with choreography, realistic body movement, beat synchronization, dynamic camera angles, cinematic lighting, music, sound effects, smooth transitions, and polished video assembly. Support multiple dance styles including K-Pop, Hip-Hop, Pop, Jazz, Contemporary, Street Dance, and other modern styles. When the user provides a reference person, preserve their face, identity, body proportions, hairstyle, clothing, and overall appearance as consistently as possible throughout the video. Generate visually engaging short-form dance videos designed for TikTok, YouTube Shorts, Instagram Reels, and other social platforms. Make the opening immediately attention-grabbing, keep the choreography energetic and readable, synchronize important movements with the beat, and finish with a memorable pose or visual moment. The user only needs to provide a simple idea, reference image, or dance concept. Handle the choreography, visual direction, music, camera movement, scene generation, video creation, synchronization, and final assembly automatically.
Film
5 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4055872/2026-10-07/2m5s3ge6.png
# Cinematic Character Animator
Bring 2D or 3D characters and scenes to life with natural movement, expressive facial animation, believable body language, and cinematic camera direction. The skill analyzes the scene, characters, dialogue, music, and intended emotion before animating, keeps character appearance consistent, and uses dynamic camera moves to make scenes engaging.
## When to use
- The user wants to animate a character or scene (from an uploaded character/scene image, a text idea, a script/dialogue, or music/audio).
- The user wants a single animated clip, or a full multi-shot cinematic scene stitched together.
- The user wants to refine one stage: generate the next shot, restyle a character, re-time a clip to audio, or redo a clip.
## Reference files
- `references/animation-direction.md`: Detailed guidance on writing movement/emotion/camera prompts and keeping character consistency. Read it before generating any animation.
## Steps
1. **Clarify the request.** Determine: the input (character/scene reference image, text idea, script/dialogue, or music/audio), the intended emotion/action, the output scope (single clip vs multi-shot scene), and the orientation (`portrait` 9:16 for vertical short-form, `landscape` 16:9 for cinematic widescreen, `square` 1:1). If any of these is unclear, ask before generating. For multi-step work, first call `write_todos` to list the steps and update it as you progress.
2. **Analyze the input.**
- If a character/scene image is provided, call `recognize_image(image_url=<asset url>, prompt="Describe the character's appearance, clothing, colors, proportions, facial features, and personality in detail")` to lock the identity details for consistency.
- If audio (dialogue/music) is provided, call `transcribe_audio(audio_url=<asset url>)` to extract dialogue/lyrics for timing and lip-sync.
- If an existing video is provided, call `recognize_video(video_url=<asset url>, prompt="Describe the visuals, plot, shots, and style")` to understand it.
3. **Create or confirm the character/scene reference.**
- If no character reference exists, generate one with `generate_image` (`model='t2i'`, `orientation` matching the target, `name=<character name>`, `entity_type='characters'`). Describe appearance, clothing, colors, proportions, and facial features in detail.
- If a scene backdrop is needed separately, generate it with `generate_image` (`name=<scene name>`, `entity_type='locations'`).
4. **Generate the animation.** Read `references/animation-direction.md` first, then call `generate_video` with the channel that fits the input:
- Character/scene reference image(s) → `model='ref2v'`, passing the reference image(s) in `input_urls` (preserves character consistency). This is the default and preferred channel.
- Single scene image → `model='i2v'` with the image as the start frame.
- Pure text idea with no image → `model='t2v'` explicitly.
- Audio-driven (dialogue/music provided) → `model='ref2v'` (or `ai2v` with image+audio) and pass `audio_url` (optionally `audio_start_time`).
- Set `orientation` to match the target, `resolution` per quality need (default `720P`), and `duration` per shot (typically 4–8s). Write a detailed prompt describing the action, facial expression, body language, emotion, and camera move (see reference file). Use `name=<shot title>`, `entity_type='storyboard_videos'`.
5. **Multi-shot scenes.** For a multi-shot scene, generate each shot as in step 4 (each with `entity_type='storyboard_videos'`), keeping the same character reference image(s) across shots for consistency. Then stitch them with `mix_clips` (`clips=[{video:{url:<shot1>}}, {video:{url:<shot2>}}, ...]`), optionally adding a music/voiceover track and burned-in subtitles. Use `name=<final cut title>`, `entity_type='final_videos'`, and set `orientation` to match the clips.
6. **Optional audio.** If the user wants voiceover, call `generate_tts` (`model='tts.design'` with `voice_instruction`, o