Steaked-nodes
A bunch of tools ported from the workshop tools site to ComfyUI nodes along with a few new ones.
Nodes (23)
The don't-care-about-types router every workflow ends up with
A mini Lightroom panel that never leaves the graph
A LoRA stack you can actually read, save, and reuse
Edge detection with texture work in mind, not ControlNet prep
The node that sneaks past hotlink protection
Cheap newspaper-comic texture, straight out of a print shop
Krita's 88 blend modes, now in your ComfyUI graph
Twelve ways to break an image on purpose, none of them AI
ImageScaleToMegapixels — resize an image to a target pixel budget, not a fixed width
Make your upscale land on the right resolution, not a guess
Pick your reference without leaving the node
A whole space scene from noise and math, no model required
Pose your character in 3D, feed the result to ControlNet
Four prompts, four boxes, one coherent image
Regional prompting on anything upstream produces
Regional prompting on a photo you already have
The masked-conditioning approach with sharper boundaries
Strict masked regions on a live image feed
Masked regional edits with hard isolation
File your generations into the character's own gallery
One loader for .safetensors and GGUF, with the name as a string output
Your whole character stack behind one node
The two-string joiner that respects the Oxford comma question
Steaked Nodes
A collection of custom nodes for ComfyUI: model and LoRA loading with modern UIs, regional prompting, image processing effects, and library tooling.
Installation
- Clone into your ComfyUI
custom_nodesfolder:
cd ComfyUI/custom_nodes
git clone https://github.com/StealthNinja1O1/Steaked-nodes.git
- Restart ComfyUI
Or install via Comfy Manager -> Install via Git URL
Loaders
Smart Model Loader (SmartModelLoader)
One loader for every model type. Browses checkpoints and diffusion_models folders together, detects what each file contains from its safetensors header, and outputs MODEL, CLIP, and VAE.
- Combined searchable listing (Anima diffusion models and SDXL checkpoints side by side)
- Detects whether CLIP/VAE are baked in; external overrides shown only when needed
- Optional external CLIP/VAE override pickers (text_encoders / vae folders)
- Weight dtype option (fp8 variants) for diffusion models
- Model info modal with civitai sidecar data when available
Combo Lora Loader (ComboLoraLoader)
A LoRA stack with a full DOM widget UI: drag rows to reorder, scrub strengths with the mouse, multi-select with ctrl/shift, and activation words surfaced automatically.
- Free tab: ad-hoc LoRA stack, drop
.safetensorsfiles straight onto the node - Combos: save stacks as named combos organized in folders, re-use and re-sync them across nodes
- Activation words: read from civitai sidecars or safetensors training metadata, exposed as a node output and a copyable badge per row
- Info modal: training parameters, tag frequency, civitai data, raw metadata
- Outputs
MODEL,CLIP, andACTIVATION_WORDS(comma-joined string, ready for prompt concat)
Regional Prompting
Regional Prompts (RegionalPromptsLatent / RegionalPromptsAttention + Img2Img and Image Input variants)
Define up to 4 rectangular regions on a canvas, each with its own prompt, weight, and timestep schedule.
- Latent variants: standard latent masking, better isolation between regions
- Attention variants: attention hooks, faster with more coherent blending. Recommended for most use cases
- Interactive canvas: drag boxes to move, drag edges/corners to resize
- Per-region weight (0.0-2.0) and start/end timesteps
- Overlapping regions are normalized automatically; unmasked areas fall back to the base prompt
Image Processing
Load and Crop Image (LoadAndCrop)
Load images from the input folder (with subfolder browsing) and crop with adjustable dimensions and position.
Color Blender (ColorBlender)
Color overlay with multiple blend modes, hue/saturation/brightness adjustments, temperature, tint, and vibrance controls.
Image Distortion (ImageDistortion)
12 distortion effects: Wave, Swirl, Kaleidoscope, Pixel Sort, Displacement, Fisheye, Ripple, Twist, Spherize, Glitch, Mosaic, and Warp. Parameters: intensity (0-100), frequency (0-100), offset x/y.
Edge Detection (EdgeDetection)
Edge detection with Sobel, Prewitt, Scharr, Roberts, Laplacian, and Canny algorithms. Adjustable threshold, thickness, edge/background colors, invert.
Image Blender (ImageBlender)
Blend two images with 14 professional blend modes (Normal, Multiply, Screen, Overlay, Soft Light, Hard Light, Color Dodge, Color Burn, Darken, Lighten, Difference, Exclusion, Add, Subtract) with opacity and optional mask.
Halftone Effect (HalftoneEffect)
Retro printing and dithering effects: Halftone Dots, Halftone Lines, Bayer Dithering, Floyd-Steinberg, Ordered Dithering, Newspaper, Crosshatch.
Nebula Generator (NebulaGenerator)
Procedural nebula/texture generation.
Scale to Megapixels (ImageScaleToMegapixels / ImageScaleToMegapixelsAdvanced)
Scale to a target megapixel count while snapping dimensions to multiples (64 for SDXL, 8 for SD1.5) to avoid generation artifacts.
Utility
Prompt Library (SteakedLibrary)
Prompt and character library with civitai integration: browse your lora collection, fetch civitai metadata and images, and store prompts/characters in a shared library. Booru tag proxy included for tag lookup and autocomplete.
Save Image to Library (SaveImageToLibrary)
Save generated images into the library for later reference.
Text Concat (SteakedTextConcat)
Concatenate text inputs with a configurable separator.
Any Switch (AnySwitch)
Pass through the first connected input of any type.
OpenPose Editor (OpenPoseEditor)
Edit OpenPose skeletons directly in the node.
Get Image from URL (GetImageFromURL)
Load an image from a URL with referer support (for sites that block hotlinking).
License
MIT