> ## Documentation Index > Fetch the complete documentation index at: https://comfytv.org/llms.txt > Use this file to discover all available pages before exploring further. # 3D Director > A previz blocking stage: place low-poly actors on a ground plane, run them along paths, cut between camera shots, then capture stills or record the whole sequence as camera/blocking reference for video models. 3D Director ## What this node does **3D Director** is a previsualization console. You populate a simple 3D stage with **actors** — characters, horses, cars, dogs, plus set dressing like trees, houses, rocks, walls, pillars and props — give the moving ones **paths** to run along, and define **shots**: camera moves with their own tracks, each holding for its duration before cutting to the next. Press play to preview the whole sequence, then **capture** a frame per shot or **record** the full timeline to a video. The point of the output is *reference, not render*: the recorded low-poly video pins down camera trajectory, timing and character blocking for a reference-to-video model (feed it as `@video_0` alongside styled character images), and captured frames pin per-shot composition for image models. ## When to use it * Multi-shot scenes where camera moves and actor blocking must stay consistent across takes * Producing a motion reference for reference-to-video workflows (the model copies your camera, you keep full control) * Quick spatial thinking: staging chases, dialogues, reveals before spending GPU time ## How ComfyTV designed this * **Actors on tracks**: each mobile actor follows a spline path across the whole timeline with an editable speed curve; set dressing stays put. Mounts let a character ride a horse or car. * **Shots cut, actors persist**: shots are camera setups with their own path and duration; the world keeps moving through cuts, like a real multi-camera shoot. A shot can **lock** onto an actor to keep it framed while the camera travels. * **Sun, ground, aspect**: a sun direction for readable shadows, ground styles, and an aspect preset that letterboxes the viewport to your delivery format. * **Everything in one widget**: the full scene (actors, paths, shots, sun, ground) serializes into `previz_state`, so it saves and copies with the workflow. * **Agent access**: the [MCP tools](https://github.com/jtydhr88/ComfyTV/blob/main/docs/mcp.md) `previz_get` / `previz_edit` / `previz_capture` / `previz_record` drive the same stage, including laying exact paths from coordinate lists and warning about overlapping actors. ## Types (COMFYTV\_\* vs native ComfyUI) | ComfyTV type | What it is | vs ComfyUI | | ---------------- | -------------------- | ----------------------- | | `COMFYTV_IMAGE` | Image URL snapshot | Bridge to/from `IMAGE` | | `COMFYTV_IMAGES` | Image batch snapshot | Bridge to `IMAGE` batch | | `COMFYTV_VIDEO` | Video URL snapshot | Bridge to/from `VIDEO` | ## Parameters ### width / height Capture and recording resolution (default 1280×720). ### The editor Add actors from the toolbar, drag them into place, draw paths for the movers; add shots, shape each camera track, set per-shot duration and an optional actor lock; scrub or play the timeline to check the cut. ## Outputs | Output | Type | Meaning | Downstream | | ---------- | ---------------- | --------------------- | ----------------------------- | | **image** | `COMFYTV_IMAGE` | Latest captured frame | Image workflows, references | | **video** | `COMFYTV_VIDEO` | Recorded sequence | Reference-to-video workflows | | **images** | `COMFYTV_IMAGES` | All captured frames | Shot-by-shot image generation | ## Step by step 1. Add **3D Director**; clear the demo actors if you don't need them. 2. Place your cast and set dressing; draw paths for anything that moves. 3. Add one shot per camera setup; shape its track, set duration, lock it to the subject if needed. 4. Play to verify blocking and cut rhythm. 5. **Record** the sequence → wire **video** into a reference-to-video stage as `@video_0`, together with styled character images (`@image_0`, `@image_1`). 6. In the video prompt, describe which mannequin maps to which character and state the output is photorealistic from the first frame — the previz look must never leak through. ## Full guides (recommended reading) | Guide | Contents | | ------------------------------------------------------------------------------- | ------------------------------------- | | [Generate](https://github.com/jtydhr88/ComfyTV/blob/main/docs/generate.md) | Video stages, references, @mentions | | [Agent access (MCP)](https://github.com/jtydhr88/ComfyTV/blob/main/docs/mcp.md) | Driving the 3D Director from an agent | ## Repository and workflows | Resource | Link | | ---------------------- | ------------------------------------------------------------------------------------------------------------------ | | **GitHub repository** | [https://github.com/jtydhr88/ComfyTV](https://github.com/jtydhr88/ComfyTV) | | **User guides index** | [https://github.com/jtydhr88/ComfyTV/tree/main/docs](https://github.com/jtydhr88/ComfyTV/tree/main/docs) | | **Built-in workflows** | [https://github.com/jtydhr88/ComfyTV/tree/main/workflows](https://github.com/jtydhr88/ComfyTV/tree/main/workflows) | ## FAQ **Q: The video model's first second shows the checkerboard ground and grey mannequins.**\ A: Tell the model in the prompt that the reference is low-poly previz for camera and blocking only, and that output must be fully photorealistic from the very first frame — name which mannequin becomes which character. **Q: What's the difference from Scene 3D?**\ A: Scene 3D is a single-camera look-development stage (lighting, materials, animation clips, capture channels). 3D Director is multi-shot narrative blocking — actors on paths, shots that cut. **Q: How long can the sequence be?**\ A: The timeline spans your shots' total duration; record produces a video matching it. Keep reference clips near your video model's segment length for best adherence. ## Related nodes * **Scene 3D** — single-shot look development and capture channels * **Video Stage** — consumes the recorded reference * **Director** — assembles the generated shots into a film