ComfyUI Extension: sweet-tea-nodes

Authored by tea-time-labs

Created

Updated

0 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Advanced ComfyUI node pack: seam-locked pose tiled upscaling and cap-based temporal stitch inpainting for WAN VACE.

README

Sweet Tea Nodes

ComfyUI node pack focused on high-control video generation and high-fidelity tiled upscaling.

Node Highlights

  • Ultimate SD Upscale (Pose Tiled ControlNet) (UltimateSDUpscalePoseTiled)
    • Major tiled-upscaling upgrade with lockable padding, ControlNet tile alignment fixes, normalized overlap blending, and seam-fix passes designed to remove bleed and ghost offsets.
  • Wan VACE -> Video (caps inpaint) (WanVaceToVideoCapsInpaint)
    • Temporal cap-stitching node that treats start/end caps as known regions and forces middle-region generation with explicit inpaint mask semantics.
    • Built for precise clip stitching and transition reconstruction workflows.
  • Wan VACE -> Video (hybrid subject swap) (WanVaceToVideoHybridSubjectSwap)
    • Depth-first hybrid subject-swap node for Wan VACE workflows.
    • Combines structural control with a character reference and optional source-RGB luma/detail/background contexts for controllable illustrated/live-action blending.
  • Sweet Tea Preview Video (SweetTeaPreviewVideo)
    • No-save output sink for ComfyUI's native VIDEO type.
    • Exposes an existing temp-backed video in place, or byte-copies an otherwise unservable source into Comfy temp, without decoding or re-encoding its video or audio streams.
  • Sweet Tea Execution Receipt (SweetTeaExecutionReceipt)
    • No-save metadata sink for allowlisted external API execution facts.
    • Publishes the provider, request, endpoint, operation, and optional estimated cost for Sweet Tea Studio provenance without exposing API credentials.

Source Provenance

This repository packages node code synced from the sweettea R2 bucket paths:

  • custom_nodes/usdu_pose_tiled/*
  • custom_nodes/wan_vace_caps_inpaint/*
  • custom_nodes/wan_vace_hybrid_subject/*

Install (ComfyUI)

  1. Open ComfyUI Manager.
  2. Search for Sweet Tea Nodes in the registry (after publish), then install.
  3. Restart ComfyUI.

Manual install from Git:

cd ComfyUI/custom_nodes
git clone https://github.com/tea-time-labs/sweet-tea-nodes.git

Local Validation Commands

Run from repo root:

python3 -m unittest tests.test_wan_vace_hybrid_subject
python3 -m py_compile __init__.py wan_vace_caps_inpaint/*.py wan_vace_hybrid_subject/*.py usdu_pose_tiled/*.py
comfy --skip-prompt --here node validate
comfy --skip-prompt --here node pack

Release Automation

This repo includes two GitHub workflows:

  • Validate Node Pack: four public hybrid-subject contracts, compile checks, and comfy node validate on push and PR.
  • Release and Publish: automatic semantic version bump from commit history, git tag creation, GitHub release creation, and Comfy Registry publish.

Autoversion rules in main pushes:

  • BREAKING CHANGE or type!: -> major bump
  • feat: -> minor bump
  • everything else -> patch bump

Required secret for registry publish:

  • REGISTRY_ACCESS_TOKEN

Notes

  • WanVaceToVideoCapsInpaint supports variable lengths and cap sizes.

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Learn more