comfyui-mcp 0.21.0 → 0.22.0

package/packs/artokun-flow/manifest.yaml

@@ -1,22 +1,25 @@
-# Artokun Flow v1 — WAN Animate (replacement + non-replacement modes). PERSONAL pack.
-# Supersedes the old cozy-flow pack. Two modes (per in-graph notes): non-replacement
-# (default animate/drive) and REPLACEMENT MODE (unmute the "REPLACEMENT MODE" group
-# to replace the person in the video), with a ReActor first-frame face-lock
-# (bypassable) and a muted "Upscale4x-RIFE-1080p" subgraph (RIFE + SeedVR2, opt-in).
+# Artokun Flow — WAN Animate. THE go-to pack for "take this person and animate them":
+# give it a character image + a driving video and it makes that character dance/move
+# with the clip's motion. Two modes via the "REPLACEMENT MODE" subgraph node, which
+# ships DISABLED = OFF (default — animate the whole character + background to the driving
+# motion). Set that node's Mode to Active = ON to replace the person in the real video;
+# Mute or Bypass = OFF. ReActor first-frame face-lock and the "Upscale4x-RIFE-1080p"
+# subgraph are bypass toggles (OFF = pass frames through unchanged); the upscale ships
+# off (opt-in).
 #
 # SAM3, frame-interpolation, math, mask, and post-processing nodes are CORE in a
 # recent ComfyUI (comfy_extras) — no custom node needed for those; just keep ComfyUI
 # up to date so SAM3_VideoTrack/SAM3_TrackToMask exist.
 #
-# PRIVATE NODES (NOT auto-installable — user-supplied): TSColorMatch,
-# TSPoseDataSmoother, TSVideoCombineNoMetadata from teskor-hub (private repos). The
-# graph won't fully load without them; swap to KJNodes ColorMatch / VHS_VideoCombine
-# if you don't have them.
+# TS UTILITY NODES (now auto-installed): TSColorMatch (reference-free temporal color
+# stabilizer), TSPoseDataSmoother (pose-jitter smoother), TSVideoCombineNoMetadata
+# (metadata-free video save) — all three from github.com/artokun/comfyui-teskors-utils,
+# a public fork of teskor-hub/comfyui-teskors-utils. The color/pose nodes have no public
+# drop-in equivalents, so the pack ships its own maintained fork.
 #
-# MODEL NAMING: model URLs are taken from the OFMHUB gist (hosts the base models
-# under short placeholder names at wdsfdsdf/OFMHUB). The workflow references the REAL
-# Kijai filenames, so base models are downloaded from the gist and SAVED under the
-# workflow's real names (rename-on-download). Same content, matching filenames.
+# MODEL NAMING: the base-model host serves them under short placeholder names. The
+# workflow references the REAL Kijai filenames, so base models are downloaded and SAVED
+# under the workflow's real names (rename-on-download). Same content, matching filenames.
 
 pip: []
 
@@ -27,11 +30,9 @@ custom_nodes:
   - https://github.com/kijai/ComfyUI-KJNodes               # ImageResizeKJv2, GetImageSizeAndCount, DrawMaskOnImage, ImageConcatMulti, PreviewAnimation
   - https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite # VHS_LoadVideo, VHS_VideoInfoLoaded
   - https://github.com/Gourieff/ComfyUI-ReActor            # ReActorFaceSwap, ReActorBuildFaceModel (first-frame face-lock; auto-fetches inswapper_128 + GPEN-BFR-512)
+  - https://github.com/artokun/comfyui-teskors-utils       # TSColorMatch, TSPoseDataSmoother, TSVideoCombineNoMetadata (public fork of teskor-hub)
   # --- OPTIONAL (only for the muted Upscale4x-RIFE-1080p subgraph) ---
   # - https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler  # SeedVR2* — un-mute the upscale subgraph to use
-  # --- PRIVATE / user-supplied (NOT cloned) — see header ---
-  # - teskor-hub/NEW-UTILS                 # TSColorMatch
-  # - teskor-hub/comfyui-teskors-utils     # TSPoseDataSmoother, TSVideoCombineNoMetadata
 
 models:
   # --- WAN base — gist hosts these under placeholder names; saved as the real names

package/packs/artokun-flow/pack.yaml

@@ -1,32 +1,37 @@
 name: artokun-flow
-display_name: Artokun Flow v1 — WAN Animate (replace / animate)
+display_name: Artokun Flow — Character Dance / Motion Transfer (WAN Animate 2.2)
 family: wan
 kind: video
 description: >-
-  WAN Animate 2.2 video pipeline (supersedes the old cozy-flow pack). Drives a
-  character with a video's pose/face (ViTPose + YOLOv10), Uni3C controlnet for
-  temporal stability, SAM3 tracking/masking, a ReActor first-frame face-lock, color
-  match, and video output. TWO MODES (per in-graph notes): non-replacement (default
-  animate/drive) and REPLACEMENT MODE — unmute the "REPLACEMENT MODE" group to
-  replace the person in the video. An optional muted "Upscale4x-RIFE-1080p" subgraph
-  (RIFE frame-interp + SeedVR2) can be unmuted for 1080p. Kijai WanVideo stack +
-  lightx2v/relight/Pusa/Fun LoRAs.
+  THE go-to / default pack for "take this person and animate them" — give it a
+  character image (a photo or a generated still) plus a driving video and it makes
+  that character dance, move, and mimic the clip's pose, face and motion, then outputs
+  the animated video. Reach for THIS for any character-animation / motion-transfer /
+  "have/make <character> dance this video" / video-to-video character drive or replace
+  request. WAN Animate 2.2 + Uni3C controlnet (temporal stability) + ViTPose/YOLOv10
+  pose+face detection + SAM3 tracking/masking + a ReActor first-frame face-lock + color
+  match. TWO MODES via the "REPLACEMENT MODE" subgraph node: it ships disabled = OFF
+  (default — animate the whole character + background to the driving motion); set its
+  Mode to Active = ON to replace the person in the real video (Mute or Bypass = OFF).
+  Optional off-by-default "Upscale4x-RIFE-1080p" subgraph (RIFE + SeedVR2) for 1080p.
+  Kijai WanVideo stack + lightx2v/relight/Pusa/Fun LoRAs.
 vram: 24GB+
 workflow: workflow.json
-skill: video-extend
+skill: wan-t2v-video
 launch_args: []
 sources:
-  gist: https://gist.github.com/T1M00R/318e7955fa73d82e200307179fd8e316
-  ofmhub_models: https://huggingface.co/wdsfdsdf/OFMHUB
+  base_models: https://huggingface.co/wdsfdsdf/OFMHUB
   kijai: https://huggingface.co/Kijai/WanVideo_comfy
 notes:
-  - PERSONAL PACK — requires the private teskor-hub nodes (TSColorMatch,
-    TSPoseDataSmoother, TSVideoCombineNoMetadata), NOT auto-installable. Install
-    manually or swap to KJNodes ColorMatch / VHS_VideoCombine.
+  - "AUTHOR'S MAINTAINED PACK — this is the recommended/default WAN Animate
+    character-animation pipeline. The TS utility nodes (TSColorMatch temporal color
+    stabilizer, TSPoseDataSmoother pose-jitter smoother, TSVideoCombineNoMetadata) now
+    auto-install from github.com/artokun/comfyui-teskors-utils (a maintained public fork);
+    no public drop-in equivalents exist for the color/pose nodes."
   - SAM3 + frame-interpolation + math/mask/post-processing are CORE in recent
     ComfyUI (comfy_extras) — keep ComfyUI updated so SAM3_VideoTrack exists.
-  - Base models download from the OFMHUB gist (placeholder names) saved under the
-    workflow's real Kijai filenames (rename-on-download) — same content.
+  - Base models download under placeholder names and are saved under the workflow's
+    real Kijai filenames (rename-on-download) — same content.
   - GAPS the gist does NOT cover, source separately. (a) SAM3 checkpoint
     sam3.1_multiplex_fp16.safetensors in models/checkpoints — ACTIVE, required by
     the SAM3 subgraph. (b) ReActor inswapper_128.onnx + GPEN-BFR-512.onnx — ReActor
@@ -35,10 +40,9 @@ notes:
     seedvr2_ema_3b_fp8_e4m3fn.safetensors, ema_vae_fp16.safetensors,
     rife_v4.26.safetensors.
   - Bypass the ReActor face-lock node to disable first-frame identity locking.
-  - NOT render-verified here (WAN Animate models are tens of GB). Static-validated
-    (clean lint + UI→API convert; only the muted upscale subgraph is skipped).
+  - Static-validated (clean lint + UI→API convert; the muted upscale subgraph is
+    skipped). Run locally to confirm on your hardware.
 post_install:
-  - Install the private teskor-hub nodes (or swap to standard equivalents).
   - Place the SAM3 checkpoint in models/checkpoints/ (see gaps above).
   - Restart ComfyUI (or ComfyUI-Manager -> Install missing custom nodes).
   - Load workflow.json.