A simplified ComfyUI helper plugin for SCAIL-2. This plugin does not replace ComfyUI's native SCAIL-2 implementation. It reuses native ComfyUI support and wraps the parts that are repetitive or easy to wire incorrectly:

Fit input videos to valid SCAIL-2 dimensions
Hide the wiring differences between animation and replacement modes
Generate SCAIL-2 colored masks internally
Run CLIP vision encoding internally
Generate long videos with automatic chunks and overlap
Support 512p, 704p, and custom resolution
Support multi-reference Reference Packs
Output frames through VideoHelperSuite VHS_VideoCombine

Nodes

This plugin mainly provides these nodes:

SCAIL-2 Fit Video
SCAIL-2 Simple Video
SCAIL-2 Reference Pack
SCAIL-2 Reference SAM Builder

SCAIL-2 Fit Video has one main parameter:

512p: scale the short side to 512, calculate the long side from the original aspect ratio, then align both dimensions to multiples of 32
704p: scale the short side to 704, calculate the long side from the original aspect ratio, then align both dimensions to multiples of 32
custom: expose custom_width and custom_height

SCAIL-2 Simple Video is the main generation node. The main user-facing controls are:

seed
cfg
mode
max_frames

Advanced controls are hidden by default. They include chunk size, overlap frames, and long-video mode.

SCAIL-2 Reference Pack organizes multi-reference inputs:

Up to 6 subjects
One main image for each subject
Up to 5 extra reference images
Optional scene image for animation background guidance

For multi-subject packs, the subject main images are composed into one primary reference collage before being used by SCAIL-2. Extra reference images provide additional views, clothing details, story frames, or scene information.

SCAIL-2 Reference SAM Builder is recommended for multi-reference workflows. For multi-subject or replacement mode, set SAM3_VideoTrack.max_objects high enough to cover all target people in the driving video.

Included Workflow

The included workflows are:

workflow/1. SCAIL2_simple.json
workflow/2. SCAIL2_multi_ref.json

Requirements

Use a recent ComfyUI build that includes native SCAIL-2 support:

feat: Add model support for SCAIL-2

The default workflow also expects these common custom nodes:

ComfyUI-Manager
ComfyUI-VideoHelperSuite
ComfyUI-KJNodes
ComfyUI-SAM3 or an environment that provides SAM3_VideoTrack
Nodes that provide DiffusionModelLoaderKJ, WanChunkFeedForward, and LoraLoaderModelOnly

If the workflow opens with red nodes, install the missing nodes through ComfyUI-Manager.

Example Directory Layout

Recommended layout for the default workflow:

ComfyUI/
├─ custom_nodes/
│  └─ ComfyUI-SCAIL2-Easy/
│     ├─ __init__.py
│     ├─ nodes.py
│     ├─ requirements.txt
│     ├─ README.md
│     ├─ README_EN.md
│     ├─ LICENSE
│     ├─ web/
│     │  └─ scail2_easy.js
│     └─ workflow/
│        ├─ 1. SCAIL2_simple.json
│        └─ 2. SCAIL2_multi_ref.json
└─ models/
   ├─ diffusion_models/
   │  └─ wan2.1_14B_SCAIL_2_fp8_scaled.safetensors
   ├─ text_encoders/
   │  └─ umt5_xxl_fp8_e4m3fn_scaled.safetensors
   ├─ clip_vision/
   │  └─ clip_vision_vit_h.safetensors
   ├─ vae/
   │  └─ Wan2_1_VAE_bf16.safetensors
   ├─ loras/
   │  └─ Lightx2v/
   │     └─ lightx2v_I2V_14B_480p_cfg_step_distill_rank128_bf16.safetensors
   └─ checkpoints/
      └─ sam3.1_multiplex_fp16.safetensors

Usage

Install the plugin into:

ComfyUI/custom_nodes/ComfyUI-SCAIL2-Easy/

Restart ComfyUI.
Import:

ComfyUI-SCAIL2-Easy/workflow/1. SCAIL2_simple.json
ComfyUI-SCAIL2-Easy/workflow/2. SCAIL2_multi_ref.json

Load the reference image in LoadImage.
Load the driving video in VHS_LoadVideo.
Select the resolution in SCAIL-2 Fit Video.
Select the mode in SCAIL-2 Simple Video.
Use VHS_VideoCombine to output the video.

Multi-Reference Usage

Use 2. SCAIL2_multi_ref.json:

Set subject_count in SCAIL-2 Reference Pack.
Connect one main image for each subject.
Increase reference_count if you need extra views or details.
Connect scene_image if you want background guidance.
Check that SAM3_VideoTrack.max_objects covers the target people.
Select animation or replacement in SCAIL-2 Simple Video and run.

Start with a small number of reference images. Clear full-body subject images are usually more stable.

Modes

animation

Motion transfer mode. In simple terms:

Make the person in the reference image perform the motion from the driving video.

This mode tends to preserve the body shape, outfit, and visual style of the reference image.

If you use a Reference Pack, the subject main images are used in the primary reference collage. If scene_image is connected, the result will lean more toward that background.

replacement

Character replacement mode. In simple terms:

Replace the person in the driving video with the person from the reference image.

This mode uses SAM3 track data and generates the colored masks required by SCAIL-2. It follows the position and proportion of the person in the driving video more closely.

Users only need to change SCAIL-2 Simple Video.mode. No rewiring is needed when switching modes.

Long Video Logic

SCAIL-2 Simple Video automatically chunks the driving video:

chunk_frames = 81
overlap_frames = 5

For a short test:

VHS_LoadVideo.frame_load_cap = 81
SCAIL-2 Simple Video.max_frames = 81 or 0

For long videos:

VHS_LoadVideo.frame_load_cap = 0
SCAIL-2 Simple Video.max_frames = 0

0 means no extra frame limit.

License

This project is licensed under the Apache License 2.0. See LICENSE.