Eden.art nodesuite
Maintained by Eden.art, this is a growing suite of custom nodes for building advanced pipelines.
Nodes (77)
Procedural mask reveals, no drawing required
Color-coded animation regions, 12 ways
Fit any image to an SDXL canvas, correctly
Reverse-engineer a prompt from any image
The one that respects your alpha channel
Cut a depth map into layers you can inpaint
One loader for images, folders, video, and GIFs
A boolean you can actually wire into a graph
Boolean algebra on your workflow graph
Compare any two values and get a real boolean out
Finally see what a node is actually outputting
Sweep a depth window through a scene
Pick the right number of frames, by the rules
Crop the main face out of a mask, cleanly
MediaPipe face detection, turned into a mask
A float you can wire into the graph
The three-line node that ends your type errors (and quietly truncates your numbers)
An LLM call inside your graph — the string it returns is both the point and the trap
The node that writes your first draft — and turns 'a beautiful landscape' into something generatable
LLMs mumble; this node forces them to fill in your form
A vignette you can wire straight into an image — made from pure math, no models
Paste one image onto another, mask and all — the compositor behind face swaps and region swaps
A calculator that does math on pixels — expression in, image and mask out
Padding that matches the edge color — so your outpainting doesn't start from a black slab
Crops your padded image back, and quietly fixes the divisible-by-4 rule
A constant and a type-lock in one — the unglamorous node that keeps your graph honest
The type bridge that keeps your math from rounding behind your back
Crop to your mask automatically — it hands back the cropped image AND the box coordinates
Blend up to three masks with weights, gamma, and soft clamping — the compositor for masks
A safe-eval calculator with variables a, b, c — and it hands you the answer three ways
Watch a shape grow from a seed — an organic 'fill the mask' animation, fully parameterized
The same organic fill, but with a dice roll — set a seed, get a surprise animation
Your Old Renders Still Remember Their Prompts — This Node Reads Them Back
A coin flip with a bias control — seed it, and it flips the same way every time
Pick a random file from a folder — with extension filters, subdirectories, and regex
A random number between min and max — displayed on the node, served three ways
A text file becomes a prompt roulette — one prompt per line, pick by seed
Regex find-and-replace on any text wire, without leaving the graph
Duplicate every latent in a batch — the boring node that unblocks batch workflows
Kill the alpha channel before it kills your workflow
ComfyUI Took This Node's Name — Eden's Version Still Leaves a Paper Trail
Ship the settings with the picture — save your workflow params as JSON
One seed to rule them all — and a string version of it for free
A plain text box you can actually wire into other things
Turn any string into a stable, short ID — prompt hashes for filenames and seeds
Find-and-replace for text wires, no regex required
Loop or ping-pong a frame sequence to any length — seamless animation loops
Find images still missing their IP-adapter embedding — the moodmix housekeeper
Pull your latest 'preferred' image and use it as an IP-adapter anchor
A seeded dice roll that returns a file path — for random references and more
Steal the color grade of one image and stamp it on a whole batch
The 'If' node this pack advertises is the one that doesn't ship
Get GPT-4 to caption your image — straight into a text wire
Walk a folder of images one at a time, in order — the batch pipeline workhorse
Crossfade two images with one slider — simple, batch-friendly blending
Crop your image to a friendly resolution — and read this before it eats your edges
Run any math expression on every pixel — brightness, contrast, weird curves
Keep IP-adapter weight and weight_type on wires instead of widgets
Morph keyframes into a video with denoise and IP-adapter control baked in
Squeeze Your Latents Into Half the VRAM
Morph Two IP-Adapter Styles With One Slider
Your Style Library, Back on the Graph
Feed a Shuffled Batch From a Filename List
A Random Image Batch, With Filenames You Can Actually Use
Burn a Mask Into a Video With Real Alpha
Auto-Segment an Image Into Color Regions, No Clicking
Skip It, Use the Replacement
Turn a Flat Photo Into a Deforum-Style 3D Drift
See What a Projector Would Actually Show
Roll the Dice on Style, the IP-Adapter Way
Cache Your Image Embeddings So You Never Re-Encode
The identity node you didn't know you needed
Turn sampler dropdowns into strings you can actually use
Make a Projector Ignore the Wall
Decode a whole latent batch straight to disk
Hand-pick the frames your video-to-video actually needs
The 512×512 → 512×512 Node Everyone Needs
Eden ComfyUI Pack
80+ nodes for masks, animation, color, depth, media loading and workflow logic.<br> Built and used in production by Eden.art.
Installation
Search for Eden Comfy Pack in ComfyUI Manager, or:
comfy node install eden_comfy_pipelines
cd ComfyUI/custom_nodes
git clone https://github.com/edenartlab/eden_comfy_pipelines.git
pip install -r eden_comfy_pipelines/requirements.txt
Requires ComfyUI 0.3.48+. All nodes are under Eden 🌱 in the node library, with tooltips on nearly every input.
Highlights
🎨 Mask From RGB (KMeans)
Splits an image or a whole video into up to 8 soft color-region masks, plus a combined grayscale map (top image). Clustering runs once over all frames, so each mask tracks the same colors through a video. This is the node behind Eden's TextureFlow workflow.
n_color_clusters: number of masks.softness: how feathered the edges are.equalize_areas: evens out mask sizes.
🌿 Organic Fill Animation
Grows a noisy, organic fill through the dark shape of any image and returns a mask video. loop plays it back out again. Organic Fill Random 🎲 samples every growth parameter from a seed, for quick variations.
🌀 Animated masks
Animated Shape Mask makes moving bands, sine waves and growing or shrinking circles, with soft gradient edges. Animation RGB Mask makes looping multi-region patterns (rotating or pushing segments, concentric circles or rectangles, stripes). Both are useful as masks for regional prompting and for AnimateDiff or IP-Adapter attention masks.
<img src="assets/screenshots/shape_masks.jpg" alt="Animated Shape Mask and Animation RGB Mask nodes" width="100%"> <img src="assets/screenshots/strip_shape_circle.jpg" alt="Expanding circle frames" width="100%"> <img src="assets/screenshots/strip_rgb_mask.jpg" alt="Rotating segments frames" width="100%">🌈 Histogram Matching
Transfers the color distribution of a reference image onto any batch of images. matching_fraction blends between the original and the fully matched result.
➗ Image / Mask Math
Write any expression over the inputs a, b, c, such as a * b, sin(a*pi) * b + c or max(a, b). Images and masks broadcast to each other automatically, and the result comes out as both an IMAGE and a MASK. Here a KMeans mask cuts one color region out of the image:
🔀 Logic, math and randomness
- If ANY Execute A Else B is lazy: only the branch it picks runs, so the other branch's nodes never execute.
- Compare, Math Expression, Int / Float / Bool / String and Bool Binary Operation cover everyday control flow.
- Random Number Sampler shows the last value it drew right on the node.
- Every random node uses its own seeded generator, so none of them resets the global random state for other nodes.
📝 Prompt from Image Folder
Point it at a folder of ComfyUI renders and it returns the positive prompt that made one of them. It traces the workflow embedded in the PNG from the sampler back to its text encoder. Use seed to step through the folder or pick at random.
📐 Depth & parallax
- Depth Slicer: splits an image into depth layers.
- Parallax Zoom: turns those layers into a Deforum-style 2.5D zoom/pan video.
- Depth Slice Mask Video: sweeps a thin depth band through the scene, for reveal animations.
📁 Loaders & savers
- All Media Loader: one node for an image, a folder, a glob pattern, a video (mp4/mov/webm/mkv/avi), a GIF or a zip/tar/7z of images, with frame-rate subsampling and a max resolution.
- Load Random Image(s): a seeded batch from a folder.
- Save Image Advanced (Eden): adds a timestamp and a sidecar JSON of the workflow.
- Folder loaders re-run when the files in the folder change, and stay cached otherwise.
🤖 AI helpers
- GPT Prompt Enhancer, GPT-4 Completion, GPT Structured Output (JSON) and GPT Image Description use your OpenAI key.
- CLIP Interrogator runs locally and turns an image into an SD-style prompt.
The GPT nodes read OPENAI_API_KEY from the environment, or from a .env file in the ComfyUI root.
All nodes
The name in code is the node id stored in saved workflows.
| Node | What it does |
|---|---|
| Mask From RGB (KMeans) 🎨 MaskFromRGB_KMeans | Color-region masks for images and video |
| Mask Combiner Eden_MaskCombiner | Blends up to 3 masks with signed strengths and a soft percentile clamp |
| Mask Bounding Box Crop Eden_MaskBoundingBox | Crops a mask (and image) to the mask's bounding box, after removing speckles |
| Face to Mask (MediaPipe) Eden_FaceToMask | A rectangle mask per detected face (needs the legacy mediapipe solutions API) |
| Animated Shape Mask AnimatedShapeMaskNode | Moving band, sine wave or growing/shrinking circle masks |
| Animation RGB Mask Animation_RGB_Mask | Looping multi-region band animations |
| Organic Fill Animation 🌿 Eden_OrganicFillAnimation | Organic growth filling a shape |
| Organic Fill Random 🎲 Eden_OrganicFillRandom | The same, with parameters sampled from a seed |
| Gradient Border Mask Eden_GradientBorderMask | White image with edges fading to black |
| Node | What it does |
|---|---|
| Image / Mask Math Eden_Image_Math | Expression over image/mask tensors a, b, c |
| Image Math (Pixel Expression) IMG_scaler | Expression over every pixel value x |
| Image Blender IMG_blender | Weighted blend of two batches |
| Histogram Matching HistogramMatching | Color transfer from a reference image |
| Convert to Grayscale ConvertToGrayscale | 3-channel grayscale, RGBA-aware |
| RGBA to RGB Eden_RGBA_to_RGB | Flattens alpha onto a background |
| Image Padder / Unpadder Eden_IMG_padder / Eden_IMG_unpadder | Adds or removes an edge-colored border |
| Crop to Resolution Multiple IMG_resolution_multiple_of | Crops to multiples of N |
| Aspect Pad Image for Outpainting AspectPadImageForOutpainting | Pads to an SDXL aspect ratio |
| Image Mask Composite Eden_ImageMaskComposite | Pastes a source onto a destination through a mask |
| Face Crop Eden_Face_Crop | Crops around a face mask and returns paste-back info |
| Width/Height Picker WidthHeightPicker | Scales a resolution and rounds it to a multiple |
| Projection Preview (Additive) ProjectionPreview | Simulates a projection on a textured surface |
| Surface Radiometric Compensation SurfaceRadiometricCompensation | Projector image that compensates for the surface |
| Node | What it does |
|---|---|
| Depth Slicer DepthSlicer | Depth layers via k-means |
| Parallax Zoom ParallaxZoom | 2.5D zoom/pan video from layers |
| Depth Slice Mask Video Eden_DepthSlice_MaskVideo | Sweeping depth-band masks |
| Video Frame Selector VideoFrameSelector | Evenly spaced frames and an interpolation multiplier for a target fps |
| Keyframe Blender 🎞️ KeyframeBlender | Crossfades keyframes and IP-Adapter embeds with per-frame masks |
| Extend Sequence (Loop / Ping-Pong) Extend_Sequence | Loops or ping-pongs a sequence to N frames |
| Determine Frame Count Eden_DetermineFrameCount | Snaps a frame count to a multiple of the source length |
| Node | What it does |
|---|---|
| All Media Loader 📁 Eden_AllMediaLoader | Images, folders, globs, videos, GIFs, archives |
| Load Random Image(s) 🎲 LoadRandomImage | A seeded batch from a folder |
| Image Folder Iterator ImageFolderIterator | The image at an index in a folder |
| Load Images by Filename LoadImagesByFilename | Loads images from a list of paths |
| Get Random File 🎲 GetRandomFile | A seeded random file path |
| Load Prefixed Images Get_Prefixed_Imgs | Latest images whose name contains a prefix |
| Save Image Advanced (Eden) 💾 Eden_SaveImageAdvanced | PNG with a timestamp and a workflow JSON |
| VAE Decode to Folder VAEDecode_to_folder | Decodes latents straight to disk, frame by frame |
| Masked Region Video Export (Alpha) MaskedRegionVideoExport | Transparent webm/ProRes video (needs ffmpeg) |
| Save Param Dict 📁 Eden_Save_Param_Dict | Saves up to 10 key/value pairs as JSON |
| Node | What it does |
|---|---|
| If ANY Execute A Else B 🔀 If ANY execute A else B | Lazy if/else on any value |
| Compare (a ? b) Eden_Compare | ==, !=, <, >, <=, >= on any type |
| Bool Binary Operation Eden_BoolBinaryOperation | And, Or, Xor, Nand, … |
| Math Expression Eden_Math | Expression in a, b, c; returns FLOAT, INT and STRING |
| Int / Float / Bool / String Eden_Int / Eden_Float / Eden_Bool / Eden_String | Constants |
| Int to Float / Float to Int Eden_IntToFloat / Eden_FloatToInt | Conversions |
| SD Type to String / SD Any-Type Converter SDTypeConverter / SDAnyConverter | Combo → string, or any → wildcard |
| Random Number Sampler 🎲 Eden_RandomNumberSampler | A seeded number, shown on the node |
| Random Bool 🎲 Eden_randbool | True with a given probability |
| Random Filepath Sampler 🎲 Eden_RandomFilepathSampler | A seeded file, with filters |
| Prompt From File (by Seed) 🎲 Eden_RandomPromptFromFile | Line seed % n of a text file |
| Seed 🎲 Eden_Seed | A seed as INT and STRING |
| Prompt from Image Folder 🎲 Eden_PromptFromImageFolder | Recovers the prompt of a render in a folder |
| String Replace Eden_StringReplace | Plain text replacement |
| Regex Replace Eden_Regex_Replace | re.sub with a count and case sensitivity |
| String Hash Eden_StringHash | Deterministic hash of a string |
| Node | What it does |
|---|---|
| GPT-4 Completion 🤖 Eden_gpt4_node | Prompt in, reply out |
| GPT Prompt Enhancer 🤖 Eden_GPTPromptEnhancer | Rewrites a prompt following instructions |
| GPT Structured Output (JSON) 🤖 Eden_GPTStructuredOutput | JSON reply following a schema |
| GPT Image Description 🤖 ImageDescriptionNode | Captions an image with GPT-4o |
| CLIP Interrogator 🔍 CLIP_Interrogator | Local image-to-prompt |
| IP-Adapter Settings IP_Adapter_Settings_Distribution | Shared weight and weight type |
| Random Style Mixture 🎲 Random_Style_Mixture | Random weighted mixes of style embeds |
| Linear Combine IP Embeds Linear_Combine_IP_Embeds | Interpolates two embeds |
| Save / Load IP-Adapter Embeds SavePosEmbeds / Load_Embeddings_From_Folder / FolderScanner | Caches embeds as .pth next to their images |
| Latent Type Conversion (fp16/fp32) LatentTypeConversion | Casts latents to halve memory |
| Repeat Latent Batch Eden_RepeatLatentBatch | Repeats a latent batch and its noise mask |
| Node | What it does |
|---|---|
| Debug Anything 🐞 Eden_Debug_Anything | Logs type, shape, stats and a preview of any value |
| Organic Fill Mask Animation (Deprecated) OrganicFillNode | Kept for old workflows; use Organic Fill Animation |
Workflow compatibility
Node ids, inputs, defaults and outputs never change between versions, so workflows saved with older versions keep working.
Since May 2026, ComfyUI core has its own node with the id SaveImageAdvanced, and core ids always win. This pack's saver is now Eden_SaveImageAdvanced. Old workflows are migrated automatically: opening one in the UI renames the node, and API prompts that use Eden's inputs are routed to the Eden node on the server.
More
- Example graphs:
example_workflows/ - Eden's production workflows: edenartlab/workflows
- Issues and ideas: GitHub issues
MIT licensed. 🌱 Made by Eden.art.