MOSS Card Layout and Wardrobe
Three/four-panel MOSS character-card layout with a searchable 1,407-outfit wardrobe, live previews and automatic SHA-256-verified image caching.
Nodes (14)
MOSS Character Studio
Build a complete four-panel character card from a selected face and outfit: portrait, adult anatomy reference, clothed front and clothed back. The reference workflow uses one full-canvas sampler and one VAE decode, with JSON/bbox layout, portrait-scoped reference conditioning, automatic framing and shared skin controls.
Release: nodes 1.4.1 · reference LoRA 1.0.0 · ComfyUI 0.43.0 plus the bundled reference-support patch. The existing 16-step text workflow remains available.
Install
Search MOSS Card Layout and Wardrobe in ComfyUI Manager. The legacy Manager catalog contains this Git repository. Registry version approval is separate; if a version is unavailable there, use Install via Git URL:
https://github.com/Coffmeteor/ComfyUI-MOSS-Card-Layout
Install the package requirements with ComfyUI's Python, restart and refresh. For reference generation, complete the one-time reference setup. The setup verifies the exact supported core version and backs up both modified files. It is never run automatically when the package is imported.
Make and revise a card
- Open example_workflows/Character Studio - Reference.json.
- Click Choose face. Search features/hair tags and select a thumbnail.
- Click Choose outfit. Search garment, color or material and select an outfit.
- Set character controls and Privacy coverage. Off requests an unobstructed adult anatomy panel; On requests two opaque rectangles. The whole card is regenerated when this changes.
- Run. Preview as Text shows the exact JSON used by the model; Save Image shows the complete result. Change a component and run again to revise it.
The selected recipe is 1536×864, 25 steps, CFG 3, Euler/sgm_uniform, AuraFlow shift 5, SLG layer 22 and reference LoRA strength 1.0. No trigger word is used. The LoRA loader includes a Hugging Face model-download link in workflow metadata.
Character Studio - Text.json uses the original independent 16-step accelerator and descriptors instead of reference-image identity conditioning. It has its own settings; do not substitute that accelerator into the reference workflow.
Add your own assets
Add my own accepts PNG/JPEG/WebP images with a name, tags and editable fields. Face geometry, hairstyle/color and expression have separate fields. Outfits use shared garment identity, front/back details, portrait clothing and wearer-side accessory descriptions. Uploaded descriptions are entered manually.
My library, Edit, Remove and Undo remove manage personal items.
They live in ComfyUI/user/moss-card-layout/library and survive package updates.
A workflow JSON does not include personal images; share those separately when needed.
Downloads and dependencies
The curated library contains 60 faces and 100 outfits. Catalog descriptions ship
with the package. Image previews download from the original
portrait gallery and
costume gallery, with bounded downloads,
checksum verification and persistent caching under ComfyUI/user/moss-card-layout.
Source images are not redistributed or relicensed by this repository.
Automatic reference framing downloads Google's pinned 3.76 MB MediaPipe landmark
model on first use into ComfyUI/models/mediapipe. It verifies the exact size and
SHA-256 before installation. Existing verified models are reused; an optional
MOSS_FACE_LANDMARKER_MODEL environment variable supports a shared local copy.
Observed behavior
The release examples retain the four roles and apply selected outfits in one sampling pass. Short neck remnants, facial/expression variation, accessory changes and differences in skin illumination remain visible in some outputs. Automatic framing can level a tilted face, suppress headwear and locally adjust unequal eye geometry; it can also alter outer hair. Direct reference access is scoped to the portrait, while target image-to-image and text self-attention remain shared.
The node pack is MIT licensed. The bundled ComfyUI reference patch is GPL-3.0;
see licenses/ComfyUI-GPL-3.0.txt. Model weights have their separate Apache-2.0
license and attribution on Hugging Face. Gallery assets retain their source terms.