IAMCCS V2V Bus To SCAIL

SCAIL is a video-replacement model family: it takes a pose video and a reference image and generates a new video where the reference subject performs the pose. V2V Bus To Scail is the adapter that unpacks the IAMCCS media/control buses into exactly the inputs SCAIL's pipeline wants - pose video, reference image, optional masks, prompts, and geometry.

Same shape as the other BusTo nodes, with a SCAIL-specific twist: it surfaces optional masks. pose_video_mask_image and reference_image_mask_image let you restrict what counts as "pose" and what region of the reference gets carried over - which is how you keep a multi-person or cluttered scene from confusing the model. There's also a replacement_mode output (a boolean) that tells the downstream backend whether this is a full replacement or a partial one, driven by what the control bus asked for.

Inputs

• media_bus, control_bus - required.
• pose_video_mask_image, reference_image_mask_image (optional) - masks for pose and reference.

Outputs

• pose_video - the motion source for SCAIL.
• reference_image - the subject to re-enact.
• pose_video_mask_image, reference_image_mask_image - the masks, echoed if provided.
• replacement_mode (boolean) - full vs partial replacement signal.
• positive_prompt, negative_prompt, fps, generation_width, generation_height, output_prefix, source_audio, source_video_info, report.

Install

Ships in IAMCCS-nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

or via ComfyUI Manager (search "IAMCCS"), restart. The SCAIL pipeline itself needs its own model assets (SAM3.x for tracking, CLIP vision, VAE) - this node only does the bus conversion, it doesn't supply the model.

Practical notes

SCAIL's backends in this pack come in single-person and multi-person identity flavors (scail vs scail_multi on the Active Backend Switch), and the masks are where you steer between them. If you're doing single-person replacement, a tight reference mask is usually enough; for multi-person you'll want the identity tracking pipeline and to feed mask outputs into it. Because the mask inputs are optional, an unconnected mask just means "use everything" - which is often fine for clean single-subject clips. If the reference is busy, though, mask it or the model will try to carry over background, too.