Mickmumpitz-Nodes
A collection of custom nodes for ComfyUI by Mickmumpitz.
Nodes (84)
Pull clean context frames out of a compressed video
Know whether there's audio without crashing the graph
The one knob that unfreezes audio-driven mouths
One click of color grading across every frame of a video
Pick one prompt batch out of five
Hand your video's broken frames to an interpolator
Sew the interpolated frames back into the clip
Collapse a dozen wires into one cable
From Ideogram's 0-1000 boxes to real pixels
One regex that fixes every axis-order fight
Turn an expensive branch off without re-running it
Edit captions and drag boxes without leaving ComfyUI
Reassemble a batch after the detail pass
Send small faces to the detail branch, big faces straight through
Get Ideogram 4 to caption its own training data
Your training folder, dumped into a graph
The right captioning prompt for every LoRA architecture
Filter Impact's SEGS by how much of the frame they fill
See your captions before you train on them
The caption loader that pairs with your image loader
A YOLO detector provider that skips the Impact Subpack install headache
Tell the graph 'this part is done' before the gate opens
Keep a number where it belongs
Composite a regenerated element so the seam disappears
Pull the context frames out of a pre-generated video
Kill the flash at iteration boundaries
Blend two control sequences without a hard cut
The end frames that anchor each iteration of a long video
Feeds each pass of an iterative video its own chunk of control frames
Hands each iteration its target end frame
Make one node wait until the whole graph is done
A master switch for the 'wait until this subgraph finished' gates
The node that stitches your video together across iterations
Squeeze a 240-frame video into an 80-frame context window
Purge the video model so your image job actually fits
Stop reloading Flux every time you switch models
Batch images like Kijai's, minus the black frames
Tag images here, gather them anywhere
Turn a dozen separate generations into one image batch
The 'is this input connected?' check that won't crash your VAE
Hand a guaranteed-valid image to nodes that demand one
Decides whether you're doing image-to-video or pure reference video
Only process the video after the first pass
A prompt per video iteration, plus where each one starts
Give every video iteration its own seed
The right prompt for the current iteration, automatically
The heartbeat of the iterative video loop
Check whether you actually have a latent — without breaking the graph
LTX-friendly resolutions without doing the math in your head
Grow a mask in one direction without touching the others
The mask version of Image Exists
A/B/C the VFX pipeline without rewiring a thing
Estimate FOV — the node whose author says don't trust it
Kill the visible seam where your photo meets the generated pano
Fix the Krea-2 reference that keeps landing in the corner
Roll a pano left or right without re-entering pixel numbers
A centered seam mask that computes its own position
The one node that fixes the wrap seam on a 360° pano
Toggle a pipeline stage on and off — and actually skip the work
Find the holes your outpaint pass gave up on
Turn one phone photo into a 360° canvas you can outpaint
Stop guessing MiniMax H3 resolutions — let it compute the canvas
Slices control, plate, and mask in lockstep
Resurrect OpenPose keypoints from a plain skeleton image
Detail a hand, a face, or any pose region — without hand-drawing the mask
A rectangular mask around any pose region, no drawing required
Paste your detailed crop back in — and make the seam vanish
Turn DWPose face landmarks into a usable mask
One switch that decides which control image you're feeding
Pick a control condition, no toggle required
Depth, canny, or pose
The lazy router that only runs the preprocessor you picked
One style string, a whole batch of prompts, zero copy-paste
26 pre-calculated sizes so you never do Flux math again
How Mickmumpitz stitches your 81-frame clips into one video
The little tag node that makes multi-shot video editing sane
Pick the story and the look, get the index
Batch a handful of prompts into one clean output
Pull one prompt out of a batch by number
The detector that feeds FaceDetailer
Un-knotting the wire spaghetti in Mickmumpitz workflows
When your workflow has 40 bundles and you can't tell them apart
Stitch your Wan clips end-to-end without re-encoding the planet
Stop guessing what's divisible by 16
ComfyUI Mickmumpitz Nodes
A collection of custom nodes for ComfyUI by Mickmumpitz.
Node Categories
- Iterative Video — Nodes for multi-iteration video generation with frame accumulation, routing, and resume support
- Video Context — Context image and control frame extraction for video workflows
- Video Utilities — Video concatenation and related tools
- String Batch — Story/style selectors and string batch processing
- Utilities — Resolution pickers (including MiniMax H3 Resolution, which takes the aspect ratio from the reference clip because H3 reframes rather than letterboxes, and scales the canvas from there), preprocessing settings, execution gates, and other helper nodes
- Consistent Character Creator — LoRA dataset tools: batch loaders, the model-specific captioning Prompt Studio, the Ideogram 4 tagger, bbox converters, the interactive Dataset Reviewer, face-area batch routing, and a branch gate
Installation
Via ComfyUI Manager (Recommended)
- Open ComfyUI Manager
- Search for "Mickmumpitz"
- Click Install
- Restart ComfyUI
Manual Installation
Clone this repository into your ComfyUI custom nodes directory:
cd ComfyUI/custom_nodes
git clone https://github.com/mickmumpitz/ComfyUI-Mickmumpitz-Nodes.git
Optional dependencies
The Consistent Character Creator face-area nodes work without any extra install if you
wire an UltralyticsDetectorProvider node (ComfyUI Impact Subpack) into the splitter's
bbox_detector input, and SEGS Filter by Relative Area likewise consumes Impact's
SEGS. Only if you leave bbox_detector empty and let Face Area Batch Splitter
load its own model does it need ultralytics (pip install ultralytics) plus a face
model such as face_yolov8m.pt in ComfyUI/models/ultralytics/bbox/. Neither is
required for the pack to load.
License
MIT License