ComfyUI-MaxedOut
Custom ComfyUI nodes used in Maxed Out workflows (SDXL, Flux, etc.)
Nodes (70)
Run your Impact Pack detector over a whole batch in one node
Stitch two clips into one video, end to end
One node where you used to need two
Crop to your subject's bounding box in one node
Never hand-copy a workflow JSON again
Flux 2's weird latent format, tamed
Flux resolutions, finally in a dropdown
Scale to a megapixel target and skip Flux's bad sizes
Flux resolution presets, picked for you across three tiers
One dropdown, eighteen Flux resolutions, zero typing
Chop the junk frames off the front of your clip
Chop the mushy ends off a clip without opening an editor
Grab the exact frame (or run) you need from either end of a clip
What resolution is this latent, anyway?
Split a long frame sequence into chunk-sized groups
See your mask tinted over the image before you waste a sample
A/B your images and save the winner in one node
Scale to a megapixel count and land on an SDXL-safe size
Give the instruction something to look at
Making Krea 2 stop redrawing the face you liked
Rotate the conditioning, not the seed
The lazy Krea 2 variety hack
Left mask, right mask — split a latent down the middle
The node that unwraps a WAN 2.2 latent pipe into real wires
The checkpoint loader that stops you guessing what you loaded
Feed a folder of your own outputs back into the graph
Load an image, get its prompts back too — and batch the whole folder
Load an image and its own prompts in one go
Resume a WAN 2.2 run with every setting intact
Re-run a WAN 2.2 I2V latent without re-wiring a dozen wires
Resume every WAN 2.2 run in a folder, one after another
Load a whole folder of WAN 2.2 latents, settings included
The batch latent loader that remembers what your Wan run used
Resume a saved Wan 2.2 latent with its original prompt and settings
The batch video loader
The two-node video loader, collapsed into one
Load one video — or flip a switch and drain the whole folder
Load a video into WAN 2.2 without the audio drama
The Power LoRA loader for people who got tired of one-LoRA-per-node
The scaler that knows LTX wants a 32-aligned Stage 1 and a /64 final
The refine pass that makes LTX 2.3's two-stage flow work
The Stage 1 LTX sampler with the trained 8-step schedule already in it
The LTX empty latent that doesn't make you do the 8n+1 math
Grow your canvas for outpainting — with the mask built in
Paste an image exactly where a mask says, no hand-aligning
Look at a video in the viewer without writing it to disk
One node for the Flux text-encode + guidance combo
Mask-free image editing, compressed into one node
Three reference images, one prompt — Qwen-Image-Edit conditioning in a box
Pick an aspect ratio and be done
Hand your graph a width and height, not a latent
Save Checkpoint keeps crashing on fp8 CLIP? This is the fix
Save images under their real names — for datasets, not galleries
Save, preview, or save quietly — the output node that stops filling your viewer
The checkpoint for your WAN 2.2 workflow that isn't a checkpoint
Save a latent so you can finish it later without reloading the model
Save a new video version next to the original without clobbering it
SDXL resolutions, in a dropdown, for once
SDXL resolution, matched for you
Don't waste a KSampler pass on an empty mask
Crop to your subject, not to the center of the frame
A/B your videos side by side without juggling tabs
An empty WAN 2.2 video latent without the mental math
The image latent WAN 2.2 I2V actually wants
Pin the first AND last frame of your WAN 2.2 clip
Drop your I2V start image into a Wan 2.2 bucket without the math
Make your second image match your first — the WAN 2.2 way
Feed WAN 2.2 the bucket size it actually wants
WAN 2.2 I2V conditioning without the scaling headache
Z-Image Turbo's 64-block rule, handled for you
ComfyUI-MaxedOut
Custom ComfyUI nodes I use in my own Maxed Out workflows (SDXL, Flux, WAN 2.2, and more).
Mission
My main goal is to make existing nodes easier to use in the day-to-day especially for my own workflows.
Overview
- Cleaner UX on frequently used nodes.
- Time-saving presets for common resolutions.
- Image/Video Comparer nodes with easy save.
- Advanced Wan 2.2 nodes for my Patreon exclusive workflows.
- Optional animated sampler previews in a dockable panel.
Install
ComfyUI Manager (Recommended):
Open Manager (top-right in ComfyUI), search Maxed Out, install, then restart ComfyUI.
Manually:
cd ComfyUI\custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git
Restart ComfyUI after install.
Model Storage Auto-Registration
If you keep models outside ComfyUI/models (e.g. on another drive), point MaxedOut at that folder and every subfolder inside it (loras, checkpoints, vae, ...) gets auto-registered with ComfyUI at startup, just like ComfyUI/models/<type>. No extra_model_paths.yaml edits needed, and any new subfolder you add just works after a restart.
Set it any one of these ways:
- ComfyUI Settings panel (easiest): Settings →
MXD→Model Storage→ set Root Folder, then restart ComfyUI. - Config file: copy
model_storage_config.json.exampletomodel_storage_config.jsonin this repo's folder and setmodel_storage_root. This file is gitignored, so it's yours to keep. - Environment variable: set
MAXEDOUT_MODEL_STORAGEto the folder path before launching ComfyUI.
Leave all three unset and nothing changes -- this is entirely opt-in.
Live Sampling Previews
MaxedOut can animate image and video latents while any KSampler-family node is
running, then keep the finished preview in a docked panel. This is opt-in:
open ComfyUI Settings → MXD → Sampling and enable Display animated
previews when sampling. It is off by default, so installing MaxedOut adds no
preview decoding or saving overhead until you enable it.
LTX 2.3 is supported through the tiny taeltx decoder. The model is downloaded to the configured VAE folder only when the feature is enabled and an LTX 2.3 preview actually needs it.
Featured Nodes
| Node | What it does |
|---|---|
| Lora Loader MXD | Based on rgthree Power LoRA Loader. Fixes copy/paste issues, surfaces useful info/remove buttons, and keeps local LoRA info files organized. |
| Image Comparer MXD | Compare original vs new images in-node and save the new result easily. |
| Video Comparer MXD | Similar to Image Comparer but for video. |
| Flux Empty Latent Image MXD / ZIT Empty Latent Image MXD / SDXL Empty Latent Image MXD | Resolution presets plus vertical toggle to avoid retyping the same sizes repeatedly. |
| Save Image MXD | Simple save modes (Save + Preview, Save Only, Preview Only). |
| WAN 2.2 MXD nodes | Helpers for WAN 2.2 latent/video prep, frame tools, and I2V-focused workflows. |
| Krea 2 Edit MXD / Krea2 Edit MXD | Krea 2 source-preservation and grounded prompt encoding. No separate custom-node pack is required; model weights are still downloaded separately. |
| Prompt spellcheck | Right-click a misspelled word in any prompt box for suggestions. Works offline, no node to add. |
Companion Packs
Some things that used to live in here now ship separately, so you can install only what you want. Nothing below is required.
| Pack | What it is | |---|---| | Prompt-Library-MXD | Save named prompt snippets by category and pull them in by name, or let a wildcard pick one at random. | | Spell-Check-MXD | The prompt spellchecker on its own, if you don't want the rest of this pack. Safe to install alongside — it won't double up. | | Smart-Model-Loaders-MXD | Loader nodes that accept safetensors or GGUF in the same slot. Only needed if you actually use GGUF quants. |
Free Workflows
Free workflow releases are posted on Patreon (no sign up needed).
Credits
Huge thanks to these projects. I have learned a lot from them and built on many of their ideas:
- https://github.com/rgthree/rgthree-comfy Inspiration for LoRA Loader, Image/Video Comparer, and more.
- https://github.com/kijai/ComfyUI-KJNodes Major reference and inspiration for my own nodes.
- https://github.com/lbouaraba/comfyui-krea2edit The Krea 2 Edit MXD implementation is adapted from Conrad Locke's Apache-2.0-licensed ComfyUI-Krea2Edit project.
If you star this repo, definitely consider starring theirs too.
License
GNU General Public License v3.0 — see LICENSE. Third-party notices and license details are recorded in THIRD_PARTY_LICENSES.md and alongside vendored code.
This pack has no pip dependencies — everything it needs ships with ComfyUI.