MD Nodes
A wild collection of custom nodes for ComfyUI including noise schedulers, samplers, audio preview, latent visualizers, and more — built for maximal creative chaos.
Nodes (71)
Stop debugging latents blind — see them as plots
Normalize, export, and embed your workflow in one node
One save node for every format, with the workflow baked in
A text box that rolls dice for you
Higher prompt adherence without the CFG burn
A seed bank for the ones you want to keep
The counter that never resets when you blink
Name your files like a project folder, not a timestamp
Richardson extrapolation to skip steps without wrecking the image
Auto-pause your queue before the card hits the throttle wall
Fourteen schedulers in one node, and it can run two at once
The scheduler node you'll actually set in ten seconds
The three-knob scheduler for people who just want it to work
Running Ollama and ComfyUI on the same GPU? This node frees the VRAM first
A whole mastering suite in one node
Repaint ten seconds of a track
The denoise knob for audio-to-audio
The one node that fixes quiet, distorted ACE-Step audio
Loads the 1.5 XL music model with the AuraFlow shift baked in
A seed node that actually remembers what it did
A corrected-Euler sampler that doesn't mush the details
A two-way switch that doesn't care what's on the wires
Shuffle the prompt tokens to break repetitive AI mush
27 target profiles that fix muddy, muffled AI audio in one pass
The node that keeps a NaN from blowing out your speakers
Trim and fade your audio without leaving ComfyUI
See your audio before you master it
Actually see what your generated audio sounds like
One node from raw AI audio to broadcast-ready
Feed your sampler Perlin and Voronoi noise
Stack LoRAs like a preset bank — MD's Style Butler, minus the hype
Aspect-ratio-aware empty latents that won't blow up your model
The tanh 'boomerang' schedule that clusters steps where the model is moving fastest
Update every custom node from inside ComfyUI — with the safety rails on
Catch NaN and blown-out pixels before they reach your save node
When your audio decode crackles or your video flickers, run the latents through this
Direct a latent like a timeline — mask an intro, regen just the chorus
Animate any number over time — the LFO generator that outputs a list *and* a live value
The other half of the .bin cache — read your saved prompts back in
Get your audio to Spotify loudness in one node — and know the number you hit
Multiband compression for AI audio — tame the peaks without killing the punch
Filters, shelves, and four parametric bands
The one-knob gain trim for your audio pipeline
Catch every peak at -0.1 dB before it clips
Get both the int and the float result
B of zero just returns zero
Double it, halve it, or scale any parameter
A − B with both output types, for the times you just need a difference
When your model goes staticky or glitchy, force it back into eval mode
Blend up to five noise sources like image layers
A 5-path router for any data type, with a crash-proof fallback
Never silently generate from a corrupted latent again
Update or roll back your custom nodes from inside the graph — backups included
Stop re-encoding that negative prompt on every run — cache it as .bin
Widen or collapse your stereo image — mid/side in one slider
Route your ComfyUI graph on prompt content with MD String Logic
Keep long ComfyUI queues from dying on you
The MD Universal Context Bus
The settings-as-files layer behind MD's samplers
Edit YAML configs from inside ComfyUI with MD YAML Utils
MD Noise Decay Scheduler
The preset version of MD's sampler for when you don't want 20 dials
The 60-knob sampler MD Nodes is built around
The full-control sampler for hyper-detailed final renders
An ACE-Step prompt farm with an optional LLM brain
Stitch two noise schedules into one with MD Sigma Concatenate
Smooth out a jagged noise schedule before it bites you
Filenames that tell you what the render actually was
One text file in, one line out
MD Universal Wildcard Orchestrator
Structured prompts for audio, image and video
MD_NODES
▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄
█▓▒░ ░▒▓█
█▓▒░ ▄▄▄████████████▄▄▄ ░▒▓█
█▓▒░ ▄▄██████████████████████▄▄ ░▒▓█
█▓▒░ ▄███████▓▓▒▒░░██░░▒▒▓▓███████▄ ░▒▓█
█▓▒░ ▄██████▓▓ ▄▄▄▄ ▒▒ ▄▄▄▄ ▓▓██████▄ ░▒▓█
█▓▒░ ██████▓▒ ▄██████ ██████▄ ▒▓██████ ░▒▓█
█▓▒░ ▄▄ ██████▒░ ████████ ████████ ░▒██████ ▄▄ ░▒▓█
█▓▒░ ▄████ ▐█████▒ █████▓▒░█ █░▒▓█████ ▒█████▌ ████▄ ░▒▓█
█▓▒░ ▄██████ ██████░ █████████ █████████ ░██████ ██████▄ ░▒▓█
█▓▒░ ██████▀ ██████░ █████████ █████████ ░██████ ▀██████ ░▒▓█
█▓▒░ ▀██████ ▐█████▒ ▀█████▓▒░ ░▒▓█████▀ ▒█████▌ ██████▀ ░▒▓█
█▓▒░ ▀████ ██████▒░ ▀███████ ███████▀ ░▒██████ ████▀ ░▒▓█
█▓▒░ ▀▀ ██████▓▒ ▀▀████ ████▀▀ ▒▓██████ ▀▀ ░▒▓█
█▓▒░ ▀██████▓▓ ▀▀▀ ▀▀▀ ▓▓██████▀ ░▒▓█
█▓▒░ ▀███████▓▓▒▒░ ░▒▒▓▓███████▀ ░▒▓█
█▓▒░ ▄▄▄ ▀▀██████████▄▄▄▄██████████▀▀ ▄▄▄ ░▒▓█
█▓▒░ ▄██████▄▄ ▀▀▀██████████████▀▀▀ ▄▄██████▄ ░▒▓█
█▓▒░ ▄████████████▄▄▄ ▀▀▀▀▀▀ ▄▄▄████████████▄ ░▒▓█
█▓▒░ ██████▓▓▒▒░▀███████▄▄▄ ▄▄▄███████▀░▒▒▓▓██████ ░▒▓█
█▓▒░ ██████▒░ ░▒▀█████████▄▄ ▄▄█████████▀▒░ ░▒██████ ░▒▓█
█▓▒░ ██████░ ░▒▀▀██████████████▀▀▒░ ░██████ ░▒▓█
█▓▒░ ▀█████▄ ░▒▀▀██▀▀▒░ ▄█████▀ ░▒▓█
█▓▒░ ▀█████▄ ░░ ▄█████▀ ░▒▓█
█▓▒░ ▀█████▄ ┌────────────────────────────┐ ▄█████▀ ░▒▓█
█▓▒░ ▀█████ │ M D M A N O D E S │ █████▀ ░▒▓█
█▓▒░ ▀▀██ │ P U R E D I F F U S I O N │ ██▀▀ ░▒▓█
█▓▒░ ▀ └────────────────────────────┘ ▀ ░▒▓█
█▓▒░ ┌─────────────────────────────────────────────────────────────┐ ░▒▓█
█▓▒░ │ CRACKER..: MDMAchine PACKER....: PYTHON 3.10+ │ ░▒▓█
█▓▒░ │ STATUS...: EUPHORIC NODES.....: 80+ UNLOCKED │ ░▒▓█
█▓▒░ │ SUPPLY...: 115200 BAUD PROTECT...: GPL v3 │ ░▒▓█
█▓▒░ └─────────────────────────────────────────────────────────────┘ ░▒▓█
█▓▒░ ░▒▓█
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
📟 What is MD_Nodes?
MD_Nodes is a collection of 80+ advanced ComfyUI custom nodes built by MDMAchine (A&E Concepts). Originally focused on audio generation workflows with ACE-Step, the suite has expanded to cover advanced sampling and scheduling, guidance systems, professional audio mastering, workflow automation, GGUF model loading, and system safety tools.
Everything in this repository is free to use. All code is open source GPL v3.
🏗️ About This Repository
License: GPL v3 — pure Python source, no compiled binaries in this release.
All .py files in this repo are complete, readable source code under the GNU General Public License v3. There are no hidden .pyd or .so binary cores in this initial public release — what you see is what you get.
Why GPL v3?
- ✅ Transparency — Audit exactly how every node works.
- ✅ Community protection — Improvements to the wrappers must be shared back.
- ✅ Commercial freedom — Use these nodes in personal or commercial workflows, no restrictions.
Future Compiled Cores
The architecture is designed to support optional .pyd/.so compiled cores for select high-value algorithms in future releases. When that happens, the compiled cores will be free to use in workflows but protected against reverse engineering. The GPL v3 Python source layer will always remain fully open. This release ships source-only.
🔥 Node Categories
🎨 Samplers & Schedulers
The engines that drive the diffusion process.
PingPong Sampler FBG
Bidirectional ancestral sampler with feedback guidance.
- Classes:
PingPongSamplerNodeBasic,PingPongSamplerNodeFBG,PingPongSamplerNodeLite - Features: Feedback guidance logic, restart modes, look-back SNR smoothing, NaN recovery
- Use Case: Resolves hyper-detailed textures in images or complex harmonics in audio that standard samplers turn to mush. The FBG variant is the primary production sampler for ACE-Step audio workflows.
AMED Sampler
Adaptive multi-engine diffusion sampler.
- Class:
MD_AMED_Sampler - Features: Multi-mode ancestral/predictor-corrector hybrid
- Use Case: Finding the middle ground between fast render speed and high-quality output.
F-Sampler
Fast sampler using Richardson extrapolation to skip steps.
- Class:
FSampler - Features: Conservative/aggressive skip strategies, quality-preserving step reduction
- Use Case: Cuts rendering time significantly while maintaining comparable output quality.
Hybrid Sigma Scheduler
Universal precision noise scheduling with 14 curve algorithms.
- Classes:
HybridAdaptiveSigmas_Advanced,_Basic,_Lite - Modes: Karras, Poly, AYS, Fibonacci, Exponential, Tangent, LINA, and more
- Use Case: Replaces the default ComfyUI scheduler with fine-grained control over when the model focuses on broad structure vs. fine detail.
GITS Scheduler
Gradient-informed timestep scaling.
- Class:
MD_GITS_Scheduler - Use Case: Allocates steps where the latent is changing most rapidly, skipping over stable regions. Efficient for both audio and image generation.
Noise Decay Scheduler
Custom decay curves for sigma scheduling.
- Class:
NoiseDecayScheduler_Custom - Use Case: Manual control over the exact noise decay profile.
🧭 Guidance & Optimization
Tools that steer generation without destroying output quality.
APG Guider (Forked)
Adaptive Projected Gradient guidance.
- Class:
APGGuiderForked - Features: Scheduled APG scale, CFG, and momentum by sigma
- Use Case: Projects guidance math safely, allowing high prompt adherence without burning or oversaturating output.
Apply TPG
Token Perturbation Guidance — breaks repetitive patterns.
- Class:
MD_ApplyTPG - Use Case: Shuffles prompt tokens at the attention level to prevent repetitive structural artifacts.
🎵 Audio Processing
Professional audio tools built directly into the ComfyUI node graph.
Auto Master Pro
Iterative mastering to hit a target LUFS and spectral profile.
- Class:
MD_AutoMasterNode - Features: 3-band compression, stereo widening, automatic gain staging
- Use Case: Makes AI-generated audio sound broadcast-ready in a single node.
Audio Auto EQ
One-click adaptive EQ with 18+ target profiles.
- Class:
MD_AudioAutoEQ_Adaptive - Profiles: Vocal Clarity, Podcast, Cinematic, Warm Analog, and more
- Use Case: Instantly corrects muddy, muffled, or harsh AI audio.
Mastering Chain (Modular)
Individual mastering components for custom pipelines.
- Classes:
MasteringChainNode,MD_Mastering_Gain,MD_Mastering_EQ,MD_Mastering_Compressor,MD_Mastering_Limiter - Use Case: Manual control over every stage of the mastering signal chain.
Broadcast Tools
Loudness normalization for platform and broadcast standards.
- Classes:
MD_LUFS_Normalizer,MD_Stereo_Width_Controller - Use Case: Hit exact LUFS targets for Spotify, YouTube, EBU R128.
Advanced Audio Preview & Save (AAPS)
Professional audio export with metadata embedding.
- Class:
AdvancedAudioPreviewAndSave - Features: MP3/FLAC/OPUS export, LUFS normalization presets, waveform visualization, workflow JSON embedding
- Use Case: The final output node for audio workflows. Normalizes, exports, and embeds generation metadata in a single step.
Audio Simple Editor
Sample-accurate trimming and fading.
- Class:
MD_AudioSimpleEditor - Use Case: Slice, trim, and apply linear/exponential fades to audio tensors.
📦 ACE Engine
Nodes specifically for ACE-Step audio generation models.
ACE-Step XL Loader
Model loader for ACE-Step 1.5 XL variants.
- Class:
MD_ACE_XLLoader - Features: Adapter support, tunable AuraFlow shift
- Use Case: Loads ACE-Step base, SFT, and turbo UNet variants with correct architecture detection.
ACE Sigma Denoise Patcher
Slices an existing sigma schedule for audio-to-audio workflows.
- Class:
MD_ACE_SigmaDenoisePatcher - Use Case: Feed it an existing audio file and re-generate with altered style or instrumentation while preserving the original structure.
AceStep Inpaint (Generative Fill)
Time-based generative fill mask for audio latents.
- Class:
MD_ACE_LatentInpaintMask - Use Case: Mask a specific time region in an audio latent and regenerate just that section.
🛡️ Guardian Suite
Crash prevention and output quality protection.
NaN / Image / Audio Guardians
Multi-modal protection against math errors and artifacts.
- Classes:
MD_NaN_Guardian,MD_Image_Guardian,MD_Audio_Guardian - Use Case: Detects and repairs NaN/Inf values in tensors before they crash ComfyUI or produce garbage output.
Universal Latent Sanitizer
Deep-level latent repair.
- Class:
MD_LatentSanitizer - Use Case: Clamps wild outlier values before VAE decode, eliminating static pops and decoding artifacts.
🛠️ Workflow Automation & Utilities
Prompting & Wildcards
Wildcard expansion, LLM integration, and scene automation.
- Classes:
WildcardPromptBuilder,UniversalWildcardOrchestrator,SceneGeniusAutocreator - Use Case: Automate prompt generation with nested wildcard logic, local LLM routing (Ollama/LM Studio), and genre/style preset libraries.
YAML Configuration
YAML-driven parameter systems for complex nodes.
- Classes:
MD_YAML_Generator,MD_YAML_Utils - Use Case: Load and manage complex node configurations from human-readable YAML files. Used extensively by the PingPong sampler nodes.
Smart Filenames & Saving
Intelligent filename generation with metadata embedding.
- Classes:
SmartFilenameBuilder,AdvancedMediaSave - Use Case: Auto-generate organized filenames and embed generation metadata.
Hardware & System Management
Real-time VRAM/GPU monitoring and protection.
- Classes:
MD_VRAMCanary,LLMVRAMManager,GPUTemperatureProtectionEnhanced - Use Case: Monitor GPU temperature and VRAM headroom, pause the queue if limits are hit.
Seeds & Conditioning
Seed management and conditioning cache.
- Classes:
EnhancedSeedSaver,MD_AdvancedSeedGenerator,MD_LoadConditioning,MD_SaveConditioning - Use Case: Save favorite seeds to disk, cache expensive text encoding to speed up subsequent runs.
Maintenance Tools
Node-based repo management.
- Classes:
MD_RepoMaintenance,MD_GlobalUpdateManager,MD_ModelStateReset - Use Case: Update custom nodes, clear VRAM caches, roll back broken updates.
Math, Logic & Modulators
Conditional routing and signal modulation.
- Classes:
MD_Math_Add,MD_Math_Subtract,MD_MultiSwitch,MD_LFO_Generator,MD_CustomNoiseGenerator - Use Case: Build decision-branching workflows, modulate parameters over time with LFOs, generate custom noise patterns.
🔗 Coming Soon
These algorithms are being released as standalone repositories and ComfyUI node packages. Each pairs directly with MD_Nodes workflows.
| Repo | Description | |------|-------------| | MDMAchine/STORM-Sampler | STORM adaptive hybrid ODE solver (STORK4 + DPM++3M) with LookBack SNR smoother | | MDMAchine/MD-Causal-Scheduler | 14-mode causal sigma scheduler with LINA time-axis warp | | MDMAchine/MD-HAP-Scheduler | Hamiltonian potential well sigma scheduler | | MDMAchine/MD-Audio-VAE-Tiled | Tiled VAE decoder for long-form audio with LSS, HPC, SCE |
🌐 HOT-Step-CPP
MDMAchine has contributed several plugins to HOT-Step-CPP, a C++ inference runtime with a Lua plugin system maintained by scragnog.
Contributions merged upstream include the PingPong solver, Causal and HAP schedulers, STORM Guidance V2, the STORM sampler core, Seed Manager UI, and DSP improvements to the tiled decoder. More in progress including negative prompt support.
If you're running HOT-Step-CPP, these plugins ship with it — no separate install needed. Check the HOT-Step-CPP repo for the full list and release notes.
🧰 Installation
cd path/to/ComfyUI/custom_nodes
git clone https://github.com/MDMAchine/ComfyUI_MD_Nodes.git
cd ComfyUI_MD_Nodes
pip install -r requirements.txt
Or via ComfyUI Manager: search for MD_Nodes and click Install.
Restart ComfyUI after installation.
📋 Requirements
Python: 3.10+ ComfyUI: Latest (Nodes 2.0 compatible) PyTorch: 2.0+ with CUDA (2.11+cu130 recommended for full feature set)
Audio nodes: soundfile, librosa, pyloudnorm, pedalboard
Visualization: matplotlib
See requirements.txt for the full list.
🔐 License
GPL v3 — full source, no compiled binaries in this release.
You are free to:
- ✅ Use in personal or commercial projects
- ✅ Read, modify, and learn from the source code
- ✅ Redistribute with attribution
- ✅ Fork and create derivative works (GPL v3 terms apply)
Your generated content (audio, images, video) is always 100% yours.
See LICENSE.md for full details.
🚀 Roadmap
Current: June 2026 Public Release
- [x] Full audit — GPL headers, VERSION constants, unit tests
- [x] torchaudio fully migrated to soundfile
- [x] IP firewall verified — no proprietary code in public repo
- [x] HOT-Step-CPP contributions documented
- [ ] Tooltip coverage pass (currently 66%, targeting 100%)
- [ ] Full node documentation
Next Phase
- [ ] Compiled core releases for select algorithms (.pyd/.so)
- [ ] API service endpoints (Captain Quantum, SCT Analyzer, FidelityX)
- [ ] Additional ACE-Step nodes
- [ ] Video generation utilities
💾 Credits
| Handle | Contribution | |--------|-------------| | MDMAchine (Alex) | Core architecture, all nodes, HOT-Step-CPP plugins | | scragnog | HOT-Step-CPP maintainer, upstream collaboration | | blepping | Original PingPong/APG concepts | | c0ffymachyne | Audio I/O and signal processing research | | Community | Bug reports, testing, feedback |
🐛 Support
Issues: GitHub Issues Discussions: GitHub Discussions Consulting / custom development: [email protected]
▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄
█▓▒░ ░▒▓█
█▓▒░ ▄▄███████▄ ▄███████▄▄ ░▒▓█
█▓▒░ ▄█████████████▄ ▄█████████████▄ ░▒▓█
█▓▒░ ▄██████▀▀ ▀▀████▄ ▄████▀▀ ▀▀██████▄ ░▒▓█
█▓▒░ ▄██████▓▒░ ░▒▓████▄ ▄████▓▒░ ░▒▓██████▄ ░▒▓█
█▓▒░ ██████▓▒░ ░▒▓███████▓▒░ ░▒▓██████ ░▒▓█
█▓▒░ ██████▒░ ▄▄▄▄▄ ░▒█████▒░ ▄▄▄▄▄ ░▒██████ ░▒▓█
█▓▒░ ██████░ ▄█████████▄ ░███░ ▄█████████▄ ░██████ ░▒▓█
█▓▒░ ▐█████▌ ▄█████████████▄ ▐█▌ ▄█████████████▄ ▐█████▌ ░▒▓█
█▓▒░ ██████ ▐██████▓▒░░▒▓████ █ ████▓▒░░▒▓██████▌ ██████ ░▒▓█
█▓▒░ ██████ ███████▒░ ░▒█████ █████▒░ ░▒███████ ██████ ░▒▓█
█▓▒░ ██████ ███████▓▒░░▒▓█████ █████▓▒░░▒▓███████ ██████ ░▒▓█
█▓▒░ ▐█████▌ ▀██████████████▀ ▄ ▀██████████████▀ ▐█████▌ ░▒▓█
█▓▒░ ██████░ ▀██████████▀ ▄█▄ ▀██████████▀ ░██████ ░▒▓█
█▓▒░ ██████▒░ ▀▀▀▀▀▀ ▄█████▄ ▀▀▀▀▀▀ ░▒██████ ░▒▓█
█▓▒░ ██████▓▒░ ▄████▀████▄ ░▒▓██████ ░▒▓█
█▓▒░ ▀███████▓▓▒▒▒▒▒▓▓█████▀ ▀█████▓▓▒▒▒▒▒▓▓███████▀ ░▒▓█
█▓▒░ ▀██████████████████▀ ▀██████████████████▀ ░▒▓█
█▓▒░ ▀▀███████████▀▀ ▀▀███████████▀▀ ░▒▓█
█▓▒░ ░▒▓█
█▓▒░ ┌──[ SYSTEM OFFLINE ]─────────────────────────[ STAY PURE ]──┐ ░▒▓█
█▓▒░ │ │ ░▒▓█
█▓▒░ │ >_ TERMINAL DISCONNECTED. │ ░▒▓█
█▓▒░ │ >_ DATA DUMP SUCCESSFUL. NO CARRIER. │ ░▒▓█
█▓▒░ └────────────────────────────────────────────────────────────┘ ░▒▓█
█▓▒░ ░▒▓█
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀