Sign In

SceneRef Workflow

Download

1 variant available

Config Other

SceneRef - Showcase.json

56.28 KB

Verified:

Type
Workflows
Stats

143

Reviews
Published

Oct 30, 2026

Base Model

MiniMax H3

Hash
AutoV20E5739C5E4

The MiniMax H3 Community License Agreement’s grant excludes the European Union, the United Kingdom, the Republic of Korea and the United States of America.

MiniMax H3

workflow_cafe.png
# SceneRef Showcase: stage a shot in 3D (or from photos), cast your own characters, get pictures and movies

**Nodes (required):** https://github.com/AliC-2001/ComfyUI-SceneRef-Nodes
Install: `cd ComfyUI/custom_nodes` → `git clone https://github.com/AliC-2001/ComfyUI-SceneRef-Nodes.git` → restart ComfyUI.

## What's in the zip

* `SceneRef - Showcase.json`: the workflow (drag it onto the canvas)

## What this workflow does

* **A 3D scene as the reference.** Open a Blender `.blend` or Daz Studio `.duf` right on the node, walk
  through it, place posable mannequins where people should be, pick the camera. The demo room comes
  with the nodes (`examples/living_room.blend`) and opens by itself.
* **A scene from photos.** No 3D? A few photos of a place (a `.scene` file) work too.
* **Your own cast.** `.char` files, DaSiWa RefMods or plain photos; each person can be tied to a
  mannequin colour.
* **One control panel for everything.** SceneRef Studio writes the prompt with a vision model (✎), builds
  the whole generation graph inside, and outputs a **FLUX.2 Klein 9B picture** or a **MiniMax H3 movie
  with sound and speech**.
* **Make your own** `.char` and `.scene` files (the muted group at the bottom left).

## Models

* Pictures: FLUX.2 Klein 9B (`flux-2-klein-9b.safetensors`, `qwen_3_8b_fp8mixed.safetensors`, `flux2-vae.safetensors`)
* Movies: MiniMax H3 ref2va, its Qwen3-VL 32B text encoder, the H3 video and audio VAEs, the ref2v turbo LoRA
* Prompt writer: a Qwen3-VL 8B file with its `lm_head` (e.g. `qwen3vl_8b_int8_convrot.safetensors`)

The Studio finds the files by name (`auto`); every model selector's tooltip says what it looks for. The
repository README lists every file and folder.

## How to use

1. Install the nodes, restart, load the workflow.
2. Group 1: the cast (Emma and Jack, or your own `.char` files / photos).
3. Group 2: the demo room opens by itself (`examples/living_room.blend`). Pick *Camera* in the camera
   list, move the mannequins or add more with **+ Figure**. **Browse…** opens your own scene.
4. Group 3: the café (or your own `.scene` / photos of a place).
5. The **switch** picks the scene: ON = the 3D room, OFF = the photos (only that one runs).
6. In the Studio: `create` = image (FLUX.2 Klein 9B) or movie (MiniMax H3). Write the action, press
   **✎ Write the prompt**, check the prompt box, **Run**. Results are saved to `output/SceneRef/`.

Queue from the ComfyUI page: the 3D view is rendered by the browser.

## Notes

* Tested on Windows 11 with a 16 GB NVIDIA card. Blender 4.2+ for `.blend`, Daz Studio 4.x for `.duf`.
* The people and the café are fictional, made with FLUX.2 Klein for this showcase.
  Only use photos of real people who agreed to it.
* Each model has its own license; check them before using the results.