Extensions/ComfyUI 1hewNodes
ComfyUI Extension

ComfyUI 1hewNodes

This is a custom node collection for ComfyUI that provides some utility nodes.

By 1hew·Created 2 years ago·Updated 5 days ago· 35
1hew/ComfyUI-1hewNodes
Nodes156
On cloudLocal install
Category1hewNodes/logic, 1hewNodes/conversion
Stars35
Updated5 days ago

Nodes (156)

Any Empty Bool

'Is there actually anything here?' — the empty-checker for any type

1hewNodes/logic
Any Empty Int

Turn 'is it empty' into a number you can do math with

1hewNodes/logic
Any Switch Bool

An if/else for the graph, with lazy evaluation that skips dead branches

1hewNodes/logic
Any Switch Select

One integer to route between ten workflow branches

1hewNodes/logic
Any to Any
1hewNodes/conversion
Audio Duration

How long is this audio, in seconds? One output, that's the whole node

1hewNodes/audio
Detect Guide Line

Find the vanishing point and draw the composition's guide lines

1hewNodes/detect
Detect Remove BG

Five background-removal backends behind one node, RMBG by default

1hewNodes/detect
Detect Remove BG Refine

Fix the fuzzy edges your background-removal mask left behind

1hewNodes/detect
Detect Remove SolidBG UnMult

The background remover with no model inside it

1hewNodes/detect
Detect Yolo

YOLO detection with masks you can feed straight into inpainting

1hewNodes/detect
Float Number Compare

Compare two floats and hand the decision downstream as a boolean

1hewNodes/logic
Get File Count

Count the images in a folder, then drive a batch loop off the number

1hewNodes/io
Image Add Label

Burn text labels onto images — with real Chinese font support

1hewNodes/image
Image Align Change Mask

Get a mask of only what your edit model actually changed

1hewNodes/image
Image Alpha Clean

Sweep the speckles and fringe out of your alpha channel

1hewNodes/image
Image Alpha Edge

Kill the matte fringe, keep the hair

1hewNodes/image
Image Alpha Join

Rebuild an RGBA image from RGB plus a separate mask

1hewNodes/image
Image Alpha Split

Get your alpha back without wrecking the RGB

1hewNodes/image
Image Area Compare

One boolean that saves you a whole branch

1hewNodes/logic
Image Batch Group

Chunk a long batch into overlapping windows without losing the tail

1hewNodes/batch
Image Batch Index

Grab exactly the frames you want out of a batch

1hewNodes/batch
Image Batch Interleave

Riffle-shuffle your image batch, on purpose

1hewNodes/image
Image Batch Range

Slice any image batch like it's a Python list

1hewNodes/batch
Image Batch Split

Split a batch in two — first N frames, or last N frames

1hewNodes/batch
Image Batch to List

The tiny bridge that feeds a batch to list-only nodes

1hewNodes/conversion
Image Batch Uniform

Pull a representative film strip out of any batch

1hewNodes/batch
Image BBox Overlay by Mask

Turn a segmentation mask into a drawn bounding box

1hewNodes/image
Image Blend Mode by Alpha

Photoshop-style blend modes, driven by alpha

1hewNodes/image/blend
Image Blend Mode by CSS

Every CSS blend mode, inside your ComfyUI graph

1hewNodes/image/blend
Image Blur

A Gaussian with a decimal radius and an honest off switch

1hewNodes/image
Image BW Matte

A cheap automatic cutout matte when you don't have a matting model

1hewNodes/image
Image Color Match

Stop re-rolling generations because the colors are off

1hewNodes/color
Image Color Restore

Put the original palette back after an AI edit

1hewNodes/color
Image Crop Square

Square crop around the subject, without eyeballing it

1hewNodes/image/crop
Image Crop With BBox Mask

Crop around your subject's mask — and remember where it came from

1hewNodes/image/crop
Image Edge Crop/Pad

The one node that pads to divisible and tells you where the original was

1hewNodes/image/crop
Image Edit Stitch

Build the reference | edit layout that edit models actually want

1hewNodes/image
Image Grid Split

Cut one image into a grid of tiles and get them back as a batch

1hewNodes/image/crop
Image HL Freq Combine

Put the frequency layers back together, with dials

1hewNodes/image/hlfreq
Image HL Freq Separate

Frequency separation, the retoucher's trick, as a node

1hewNodes/image/hlfreq
Image HL Freq Transform

Steal one image's texture and give it to another

1hewNodes/image/hlfreq
Image List Append

Merge image lists from two branches without losing order

1hewNodes/batch
Image List Interleave

The list version of the riffle shuffle, for mismatched sizes

1hewNodes/image
Image List to Batch

Flatten a list of images back into one batch

1hewNodes/conversion
Image Main Stitch

One clean strip out of your main shot plus extras

1hewNodes/image
Image Mask Blend

Stick a subject on a new background without the gray halo

1hewNodes/image/blend
Image Mask Crop

Crop to the mask, keep the alpha, and get the mask back too

1hewNodes/image/crop
Image Minimum Area

Upscale to a floor, never below it, and branch on whether it fired

1hewNodes/logic
Image Pad By BBox Mask

Drop a local crop back onto the full canvas, gaps and all

1hewNodes/image/crop
Image Paste By BBox Mask

Paste the inpainted crop back where it belongs

1hewNodes/image/crop
Image PingPong

Loop a clip forever with a bounce, not a hard cut

1hewNodes/image
Image Plot

Turn a batch into a contact sheet you can actually inspect

1hewNodes/image
Image Resize Area

Let sqrt(W×H) pick your resolution instead of you

1hewNodes/image/resize
Image Resize Doubao Seedream 5.0 Pro

Stop the API guessing your output size

1hewNodes/image/resize
Image Resize Flux Kontext

Put your reference image on Flux Kontext's native grid

1hewNodes/image/resize
Image Resize Gemini 3.0 Pro Image

Resize for Nano Banana Pro without calling any API

1hewNodes/image/resize
Image Resize Gemini 3.1 Flash Image

The Gemini 3.1 Flash size table, including the panoramas

1hewNodes/image/resize
Image Resize GPT Image 2.0

A resize node that speaks GPT Image 2.0's dimension language — no API, no key

1hewNodes/image/resize
Image Resize Jimeng

Size your images for Jimeng's exact buckets — locally, in two clicks

1hewNodes/image/resize
Image Resize Qwen Image

Snap your input to QwenImage's exact resolution buckets

1hewNodes/image/resize
Image Resize Qwen Image 3.0 Pro

Hand the model the shape you want back

1hewNodes/image/resize
Image Resize Square

Square images for models that demand exact sizes

1hewNodes/image/resize
Image Resize Universal

The one resizer that doesn't care which model you're feeding

1hewNodes/image/resize
Image Rotate with Mask

Rotate the subject and its mask together, or your cutout drifts

1hewNodes/image
Image Shift
1hewNodes/image
Image Solid

A solid color image that also matches your reference's dimensions

1hewNodes/image
Image Stroke by Mask

Put a colored outline around anything in your mask — the comic-label effect

1hewNodes/image
Image Tile Merge

The other half of the tiling duo — feather-blend your tiles back into one image

1hewNodes/image/tile
Image Tile Split

Slice big images into overlapping tiles so you can upscale or diffuse them on limited VRAM

1hewNodes/image/tile
Image to Mask

Make a mask from an image's brightness, with a safe fallback

1hewNodes/conversion
Image to URL

Turn an image into a URL you can paste anywhere

1hewNodes/conversion
Int Audio Count
1hewNodes/int
Int Image Count

Count the real images in a batch — ignoring empty frames and pure-black junk

1hewNodes/int
Int Image Side Length

Grab one number out of an image's dimensions — longest, shortest, width, or height

1hewNodes/int
Int Image Size

Need an image's width and height as integers? Here's the two-wire answer

1hewNodes/int
Int Mask Count
1hewNodes/int
Int Mask Side Length

The mask version of 'what's the shortest side' — for when masks carry the dimensions

1hewNodes/int
Int Number Compare

Compare two integers and get a real boolean out of it

1hewNodes/logic
Int Split

Split one integer budget into two parts — by percentage or by exact cut point

1hewNodes/int
Int Text Count
1hewNodes/int
Int Video Count

Count how many videos you actually plugged in

1hewNodes/int
Int Wan

A Wan-flavored integer node — the 4n+1 frame-count nudge

1hewNodes/int
List Custom Float

A float-list builder that speaks ranges, strides, and even full-width commas

1hewNodes/text
List Custom Int

5' and get an int list — ranges and strides in one box

1hewNodes/text
List Custom Seed

Generate N unique random seeds from one seed — deterministic, then rerunnable

1hewNodes/text
List Custom String

Turn one text box into a real list that batch loops can eat

1hewNodes/text
Load Image

Load a folder of images as a batch — and get their filenames out

1hewNodes/io
Load PS

Load Photoshop PSD layers straight into your graph — one layer, all layers, or the flattened doc

1hewNodes/io
Load Text

Pull prompts out of .txt files — with an encoding guesser that actually tries

1hewNodes/io
Load Video

Load a video as a VIDEO object, not a frame explosion

1hewNodes/io
Load Video to Image

Pull a video apart frame by frame, audio and all

1hewNodes/io
Mask Alpha Clean

Scrub the speckles off a matte before it ruins your composite

1hewNodes/mask
Mask Batch Math Ops

Collapse a whole mask batch into one mask — union or intersection

1hewNodes/batch
Mask Batch Range

Slice a mask batch with start, step, and a frame budget

1hewNodes/batch
Mask Batch Split

Split your mask batch the same way you split the images

1hewNodes/batch
Mask Batch to List

Un-batch your masks — one batch in, a list of individual masks out

1hewNodes/conversion
Mask Crop by BBox Mask

Crop a mask to another mask's bounding box — the clean cutout helper

1hewNodes/mask
Mask Fill Hole

Fill the donut holes in your masks — auto-cutout masks are full of them

1hewNodes/mask
Mask Levels

Two numbers between a mushy matte and a usable mask

1hewNodes/mask
Mask List to Batch

When your masks come back as a list, this is the glue

1hewNodes/conversion
Mask Paste by BBox Mask

Paste a mask back where it came from, no coordinates needed

1hewNodes/mask
Mask Repeat

One mask, N frames — for when a video needs the same region every time

1hewNodes/mask
Mask Separate

Cut a messy mask into its individual objects — this is the one you'll keep

1hewNodes/mask
Mask Stroke

Outlines and sticker borders, straight from a mask

1hewNodes/mask
Mask To BBox Mask

Turn a mask into its bounding boxes — but as masks, and 8-aligned

1hewNodes/mask
Mask to Image

Turn an invisible mask into a color image you can actually see

1hewNodes/conversion
Mask to SAM3 Box

Feed SAM3 the boxes it wants, straight from a mask you already have

1hewNodes/conversion
Match Brightness Contrast

Make one image's light match another's — and keep it stable across frames

1hewNodes/color
Memory Cleanup

Give VRAM a breather between queued runs

1hewNodes/util
Multi Image Batch

Multi Image Batch unifies sizes for you

1hewNodes/multi
Multi Image List

A batch is not a list — and this node knows the difference

1hewNodes/multi
Multi Image Overlay

Stack image layers in the graph, no Photoshop tab needed

1hewNodes/multi
Multi Image Stitch

Stitch a before/after comparison or a collage, cleanly

1hewNodes/multi
Multi Mask Batch

Several masks of different sizes, one tidy batch

1hewNodes/multi
Multi Mask Math Ops

Boolean masks in one node instead of five

1hewNodes/multi
Multi String Join

Join prompt blocks without the separator fiddling

1hewNodes/multi
Multi Switch Select

Route one of ten inputs to one of ten outputs with a single number

1hewNodes/logic
Multi Video Stitch

Four generated clips, four different sizes, one file

1hewNodes/multi
Range Mapping

Turn a 0-1 slider into the range you actually need

1hewNodes/util
Save Audio

The 1hewNodes Save Audio node does FLAC, WAV, MP3 and Opus

1hewNodes/io
Save Image

Save files where you can actually find them — and get the path back

1hewNodes/io
Save Text

Save anything as a .txt — prompts, coordinates, debug logs

1hewNodes/io
Save Video

Save a VIDEO object without losing its container or metadata

1hewNodes/io
Save Video by Image

You made 60 frames. Now turn them into an actual video.

1hewNodes/io
Seedream BBox to Mask
1hewNodes/conversion
String Coordinate to BBoxes

Paste coordinates as text, get real BBOXES out

1hewNodes/conversion
String Coordinate to BBox Mask

Text coordinates straight into actual masks

1hewNodes/conversion
String Filter

Clean the junk out of your prompt text

1hewNodes/text
String Join Multi

Assemble multi-part prompts without the 'seams' showing

1hewNodes/text
String Random

One prompt, five options, and a seed that lets you go back

1hewNodes/logic
String Ratio Doubao Seedream 5.0 Pro

1 Crop — Snap It to One of 8 Shapes Instead

1hewNodes/text
String Ratio Gemini 3.1 Flash Image

The name says Gemini, but it never calls an API — no key needed

1hewNodes/text
String Ratio GPT Image 2.0

Snap any image to the nearest GPT Image 2.0 ratio, locally

1hewNodes/text
String Ratio Jimeng

Hand the API a ratio it will actually accept

1hewNodes/text
String Ratio Qwen Image 3.0 Pro

Qwen Image 3.0 Pro Takes 15 Ratios — Let the Image Pick Yours

1hewNodes/text
String Resolution Doubao Seedream 5.0 Pro

1080p Counts as 1.5k on Seedream — Resolution Matching, Explained

1hewNodes/text
String Resolution Gemini 3.1 Flash Image

Gemini 3.1 Flash's Ladder Has a 512px Rung — Picking the Right Tier

1hewNodes/text
String Resolution GPT Image 2.0

The 4k Tier Isn't 4096 — Reading GPT Image 2.0's Resolution Ladder

1hewNodes/text
String Resolution Jimeng

Jimeng's Three Rungs Won't Fit Every Jimeng Node — Match Them Properly

1hewNodes/text
String Resolution Qwen Image 3.0 Pro

Two Rungs, Two Megapixels, and a 23,552-Pixel Coin Flip

1hewNodes/text
String Resolution Seedance 2.5 Multimodal Video

A node that only outputs a string, and why you'd want it

1hewNodes/text
Text Compare

A boolean for 'is this prompt about X?' — without writing a regex

1hewNodes/logic
Text Custom Extract

Pull one key out of messy LLM JSON without fighting the parser

1hewNodes/text
Text Encode Qwen Image Edit

The Qwen image-edit conditioning encoder, minus the workflow surgery

1hewNodes/condition
Text List to String

Turn a list of strings into one string without a headache

1hewNodes/conversion
Text Match Rownum

Which line does this match? Ask, and get a number

1hewNodes/logic
Text Match Value

A lookup table that lives inside your workflow

1hewNodes/logic
Text Path Extract
1hewNodes/text
Text Prefix Suffix

Wrap every line of text in a prefix and suffix, in one node

1hewNodes/text
Text to Any

The node that lies about its type, and why that's useful

1hewNodes/conversion
URL to Video

Point at a video URL and get a real VIDEO object back

1hewNodes/conversion
Video Cut Group

Scene-cut detection that hands you the keyframes

1hewNodes/batch
Video to Image

Unpack a VIDEO object into frames, audio, fps, and a frame count

1hewNodes/conversion
Video to URL

Hand your generated video a public URL — no server, no API key

1hewNodes/conversion
Workflow Name

Name your output files after the workflow that made them

1hewNodes/util
Readme
<div align="center"> <a href="./README.md"><img src="https://img.shields.io/badge/🇬🇧English-0b8cf5"></a> <a href="./README.ZH_CN.md"><img src="https://img.shields.io/badge/🇨🇳中文简体-e9e9e9"></a> </div>

ComfyUI-1hewNodes

This is a custom node collection for ComfyUI that provides a series of practical image processing, mask operations, and utility nodes.

📦 Installation

Clone the repository into ComfyUI's custom_nodes directory:

git clone https://github.com/1hew/ComfyUI-1hewNodes

📜 Changelog

v4.5.0

  • Breaking (node IDs): Rename 1hew_SaveTxt → 1hew_SaveText and 1hew_LoadTxt → 1hew_LoadText so the IDs match the SaveText / LoadText classes, the Save Text / Load Text display names and the full-word IDs used by every other I/O node; the abbreviation was also misleading now that Save Text writes txt / csv / md / json. Display names, ports, defaults and file formats are unchanged.
  • refactor(io): Register both renames through ComfyUI's Node replacement API (utils/_node_replacements.py, wired from on_load): a workflow saved with an old ID can be converted from the missing-node panel (links and widget values carry over, old_widget_ids maps the old widgets_values order), and API prompts that still send the old class types are rewritten before execution (input_mapping is explicit because the backend rebuilds inputs from it). ComfyUI builds without node-replacement support still need the node re-added by hand.
  • test: Pin both new node IDs and cross-check the replacement table against the declared schema inputs and widget order.
  • docs: Note the rename and migration path in the Save Text / Load Text docs and document the rename rule in nodes/__init__.py.

v4.4.0

  • feat(float): Add Float Mask Area Ratio — report how much of the frame a mask covers as a 0–1 float, weighted by the soft mask values (feathered and anti-aliased edges count their real coverage instead of a whole pixel), averaged across a batch, clipped to 0–1, safe on empty / non-mask input, and rounded by round (0–10, default 2, the same name and range as Range Mapping); the float output label shows the measured value in the front end.
  • feat(text): Add Text System Prompt — offer every .txt under utils/system_prompt/ as one flat <model>/[<lang>] <name> dropdown and emit the selected body verbatim on a string output; the key list is scanned once at startup, an edited body is re-read when its mtime changes (no restart needed), _ / .-prefixed files stay out of the dropdown, and an empty folder or an unknown key returns "" instead of blocking the graph. The bundled Qwen-Image-2.1 t2i / edit prompts are adapted from benjiyaya/ComfyUI-Qwen-Image-2.1-Prompt-Enhancer (MIT) and rewritten to answer in plain text.
  • feat(image): Rework Image Crop With BBox Mask's auto preset into adaptive framing — sparse specks are trimmed first (only edges 10x lighter than the densest row/column and within 1% of the total mass), the de-speckled bbox ratio snaps to the nearest standard ratio only within a 30% linear mismatch, the aligned box always fits inside the image instead of silently returning the whole frame, and scale_strength becomes a geometric framing weight from mask-tight (0) to almost the whole image (1) with width and height blending independently; crop_to_side / crop_to_length still take over the size and scale_strength is then ignored. Workflows saved with the old auto (image ratio snapped to a standard ratio) render differently by design.
  • feat(util): Range Mapping's decimal input is renamed rounding → round (same 0–10 range and default 3), matching the new float node; saved workflows fall back to the default value.
  • refactor(shared): Move the paired-pixel colour estimator into utils/_color_fit.py and use it in Image Align Change Mask, so structure-gated sampling, capped gain/bias fitting and held-out validation live in one place; the applied correction keeps the previous magnitude envelope, so the rendered result is unchanged.
  • refactor(pad): Unify pad_color's empty-value semantics across Image Pad By BBox Mask, Image Edge Crop/Pad, Multi Image Batch, Multi Image Stitch, Multi Video Stitch, Image Main Stitch, Image Edit Stitch, Image Resize Universal / Area and Image Minimum Area — an empty value (or transparent) now always produces RGBA: 3-channel input is expanded to 4 channels, the original content stays opaque and the padded area is (0,0,0,0) (the resize family expands in pad mode; crop / stretch add no padding and stay RGB). Explicit colours keep their opaque alpha and the resize family's empty colour no longer falls back to black. This changes the output channel count for RGB graphs that left the field empty, so re-check saved workflows.
  • test: Add regression coverage for the colour-fit kernel, Float Mask Area Ratio, Text System Prompt, the auto framing ladder (specks, long strips, ratio snapping, divisible_by, scale_strength blending, edge-subject shift, batches), and empty / transparent pad fills.
  • docs: Add bilingual docs for the two new nodes, update the Image Crop With BBox Mask / Range Mapping docs, add a Float Nodes section to the README node list, credit the bundled Qwen-Image-2.1 prompts in the acknowledgments, and synchronize the changelog.

v4.3.0

  • feat(image): Add Image Shift — translate a whole image batch by shift_x / shift_y, either wrapping the overflow (cyclic) or filling it with background_color (the same Image Alpha Split syntax, sm included), and return the shifted image, the matching mask and a blurred seam border_mask for seamless-tile work; channel count (alpha included) is preserved, masks are matched per frame, the non-cyclic bounds are pixel-exact, and an optional pad wrap-extends the shifted outputs (with a pad_mask marking the added frame) so seams that cross the canvas border can be inpainted with outside context.
  • test: Add regression coverage for cyclic/non-cyclic bounds, zero shift, alpha masks, per-frame mask matching, seam-band geometry, wrap padding / pad_mask, and empty batches.
  • docs: Add bilingual Image Shift docs and synchronize the README node lists and changelog.

v4.2.1

  • fix(text): Text Path Extract now returns fallback on the value port as well when the path parser raises (it used to emit None), so downstream nodes no longer receive a type the port never declared; the error path gains its own regression test.
  • docs: Document the parser-error fallback in the Text Path Extract docs.

v4.2.0

  • feat(text): Add Text Path Extract — read a value out of the text widget (JSON, fenced JSON, a Python literal, relaxed JSON or plain text) with a path such as [0].content, [-1].bbox[0] or [*].content, then emit it as auto / json / text / csv text, the raw value and a match count; the default path [*] lists every item, [*].content / [*].bbox collect whole lists (json returns one JSON array), and .count / .length / [#] count inside the path; the count port label shows the number in the front end (e.g. 3 count).
  • refactor(shared): Move the JSON-like parser used by Text Custom Extract into utils/_json_path.py, which also hosts the new path query and value formatting helpers.
  • test: Add regression coverage for the path grammar, the loose parsing fallbacks, the output formats and the shared Text Custom Extract parser.
  • docs: Add bilingual docs for the new node and synchronize the README node lists and changelog.
  • feat(conversion): Add Seedream BBox to Mask, which inverts Seedream <bbox>x1 y1 x2 y2</bbox> per-mille coordinates (x = round(px / width * 1000)) or a layer bounding_box back to pixels and rasterizes them into rectangular masks (bboxes only, <point> is ignored): separate emits one mask per box, merge unions them, plus a count output; parsing is format-agnostic (tags, bbox(...) calls, JSON, bare numbers, full-width punctuation, 图1 / Image 1 prefixes); the unit is a plain coord_mode choice between normalized (per-mille 0-1000) and absolute (pixels), boxes are always xyxy (right/bottom are edges); a layer bounding_box can be pasted as-is, the field is picked by coord_mode when both exist and by its own unit when only one does, so each layer yields a single box; pixel values above 1000 in normalized mode raise an explicit error instead of producing a wrong mask.
  • test: Add regression coverage for the normalized inverse mapping, multi-box batches, point dots, xywh and auto spaces, out-of-range clamping and empty input.

v4.1.0

  • feat(conversion): Add Any to Any, a wildcard pass-through that forwards any value unchanged through * ports, so outputs and inputs with different declared types can be linked without a real conversion.
  • feat(int): Add Int Mask Count, Int Audio Count, and Int Text Count for dynamic mask_X / audio_X / wildcard text_X ports; emptiness follows Any Empty Bool, image/mask batches count per frame, list outputs are flattened before counting, and Int Image Count / Int Video Count now share the same utils/_count_ops.py helpers.
  • feat(logic): Multi Switch Select's select accepts a standalone a / all (case-insensitive) to route every connected input lazily; Image Batch Index keeps its numeric-only grammar.
  • feat(io): Load PS now defaults preview=true, and psd_tools INFO chatter from unknown image resources or tagged blocks is suppressed by setting the psd_tools logger to WARNING (WARNING/ERROR stay visible).
  • feat(io): Save Audio defaults to mp3 (320 kbps) for new nodes and unknown legacy values; its in-node player uses a dedicated DOM container so it keeps full width on both the classic canvas and Nodes 2.0, and keeps its last preview when an unrelated execution message arrives.
  • fix(multi): Multi String Join's string_1 is now an optional socket-only input, so dynamically appended string_N sockets stay below the settings instead of shifting to the top of the node.
  • refactor(shared): Reuse utils/_image_ops.py's parse_pad_color / pad_to_rgb in Multi Image Batch, Multi Video Stitch, and the shared resize helpers instead of node-local copies.
  • test: Add regression coverage for Any to Any, the three new count nodes, Multi Switch Select select-all, and Save Audio's mp3 fallback.
  • docs: Add bilingual docs for the four new nodes, update the Load PS / Save Audio / Multi Switch Select / Multi String Join and count-node docs, and synchronize the README node lists and changelog.

v4.0.0

  • Major I/O upgrade: unify text reading/writing with metadata-aware Load Text / Save Text, add Save Audio, make Save Video container-safe, and switch Load PS to on-demand cached layer previews.
  • Logic and shared foundation: Multi Switch Select shares Image Batch Index's range grammar and supports multi-selection; duplicated image/video/text/file helpers are centralized in utils/_*.py, covered by the expanded consistency checker and regression suite.
  • Compatibility: all node IDs stay 1hew_*; Save Text uses the optional STRING text port (formerly any) and Load Text drops encode, with the front end migrating old widget values automatically; Save Audio requires PyAV.
  • Docs: bilingual node docs and README are synchronized; release notes before v4.0.0 are archived below.
<details> <summary><b>3.x releases</b></summary>

v3.31.0

  • fix(io): Save Video now names WebM/VP8 and other sources after their own container (.webm / .mp4 / .mkv, Matroska when the codecs do not fit) instead of forcing MP4, which failed with 'mp4' format does not support 'vp8' codec; metadata remux and silent-audio remux also work on current PyAV builds
  • feat(io): Add Save Audio — save AUDIO to FLAC / WAV / MP3 / Opus with output/temp routing, auto-increment or fixed filenames, optional metadata, batch splitting, and empty-input pass-through (nothing is written when no audio is connected); the node embeds an audio player that previews the saved file, matching the native Save Audio (Advanced) preview
  • feat(io): Save Text now supports txt / csv / md / json, always writes UTF-8, can append workflow/prompt metadata, and clears its in-node preview when the input is empty; the input port is now the optional STRING text (formerly any)
  • feat(io): Load Text adds file upload, live in-node preview, automatic encoding detection, and strips the metadata block written by Save Text; the encode input is removed and the front end migrates old workflows
  • feat(io): Load PS previews are now loaded on demand — new /1hew/view_ps_info and /1hew/view_ps_layer endpoints list layers and render individual PNG previews with an LRU cache, merged previews reuse the embedded composite, and all_layers uses the native batch preview on classic or a single-layer viewer on Nodes 2.0; the node frame follows the image aspect ratio
  • feat(logic): Multi Switch Select accepts the same 1-based range/list syntax as Image Batch Index (1,3, 1-3, [1,3], :step, negatives) and routes multiple inputs; both nodes share the new utils/_index_range.py parser
  • refactor(utils): Move duplicated image/video/text/file helpers into utils/_*.py, and enforce one-node-per-file plus no cross-category node imports in scripts/check_node_consistency.py

v3.30.1

  • refactor(logic): Move Image Minimum Area's divisible_by input after fit / pad_color so the widget order matches Image Resize Universal; existing workflows should recheck that node's parameters
  • fix(detect): Align Detect Remove SolidBG UnMult's gamut-safe unpremultiply with the UnMult (Auto Background) 2026-10-15 fix — negative channels are clipped and positive overflow is jointly scaled, so one bad channel no longer reverts the whole pixel to the observed RGB
  • fix(color): Image Color Restore's hole filling now bails out when every border pixel is foreground (e.g. a 1–2px transparent fringe) instead of turning the whole image into foreground

v3.30.0

  • feat(multi): Add Multi Video Stitch — stitch dynamic video_X inputs end-to-end with crop / pad / stretch size unification, an optional get_video_size reference video, and configurable pad_color / fps (video_1 reuses the first segment; a number or auto also works); the dynamic-port table now pins get_video_size below every video_X port
  • feat(text): Add String Resolution Seedance 2.5 Multimodal Video — infer the nearest Seedance 2.5 video resolution tier (480p / 720p / 1080p / 2k / 4k) from the input video by pixel area, or pass through the selected label when no video is connected
  • feat(detect): Add Detect Remove SolidBG UnMult — estimate a flat background from the image border, unmultiply to alpha, then optionally clean low alpha with threshold and composite onto background_color (or sm smart background); outputs a foreground image plus alpha mask
  • refactor(image_resize): Make Image Resize Area's divisible_by / fit behave like Image Resize Universal — align both output sides to the multiple in every branch (no more failure when the range contains no multiple) and rename the int output to resolution
  • fix(io): Save Video by Image now raises a clear error when audio preparation or encoding fails instead of silently dropping the audio track, and removes placeholder or partial output files when a save fails

v3.29.2

  • fix(image): Refine Image Alpha Split sm smart-background selection to fully replicate the stroke-aware refiner scoring and selection strategy (using the required #808080 gray candidate), including OKLab/luminance contrast, alpha boundaries, grouped sampling, low-tail scoring, and the gray margin rule; grayscale+alpha inputs are supported as well

v3.29.1

  • feat(image): Add sm smart-background mode to Image Alpha Split background_color; for each alpha image it selects black, white, or #808080 to improve subject and translucent-edge visibility

v3.29.0

  • feat(image_resize): Add Image Resize Area, an aspect-preserving area normalizer: clamp and align sqrt(width × height) to configurable bounds, synchronize optional masks, provide Image Resize Universal-compatible fit / pad_color handling, and lazily use an optional upscale_model only for undersized inputs
  • fix(image): Make Image Alpha Join return an empty output whenever the image input is empty (unconnected or a zero-length batch) instead of a transparent canvas, and treat a zero-length mask batch as no mask
  • fix(image): Make Image Alpha Split return empty image / mask outputs when the image input is empty (unconnected or a zero-length batch)
  • docs: Add bilingual Image Resize Area documentation and synchronize the README node list

v3.28.2

  • chore(tooling): Extend scripts/check_node_consistency.py to fail on any quoted 1hew_* literal in web/**/*.js that is not a current node_id, so a renamed or removed node cannot leave a stale front-end reference

v3.28.1

  • chore(tooling): Extend scripts/check_node_consistency.py to validate the web/js/dynamic_port.js port table against the node schemas — stale node_ids, duplicate keys, base / addType / select / outputBase / outputType mismatches, and invalid initial / max now fail

v3.28.0

  • feat(text): Add model-specific String Resolution selectors — String Resolution GPT Image 2.0 (1k / 2k / 4k), String Resolution Gemini 3.1 Flash Image (0.5k / 1k / 2k / 4k), String Resolution Jimeng (1k / 2k / 4k), String Resolution Doubao Seedream 5.0 Pro (1k / 1.5k / 2k) and String Resolution Qwen Image 3.0 Pro (1k / 2k) — replacing the generic String Resolution node; switch existing workflows to the matching model node
  • feat(text): Add String Ratio Doubao Seedream 5.0 Pro (8 ratios) and String Ratio Qwen Image 3.0 Pro (15 ratios) to infer the nearest supported aspect ratio from input images
  • refactor(image_resize): Reorder Image Resize Doubao Seedream 5.0 Pro presets widest → square → tallest to match the API node's aspect_ratio option order
  • docs: Add bilingual docs for the 7 new text nodes, remove the generic String Resolution docs, and document the per-node/per-model Jimeng resolution tiers for String Resolution Jimeng / Image Resize Jimeng in the node docs
  • refactor(naming): Unify model node display names to the spaced form (Image Resize Gemini 3.1 Flash Image, String Ratio GPT Image 2.0, Text Encode Qwen Image Edit, …); node_ids are unchanged, so existing workflows keep working
  • feat(image): Give Image Align Change Mask an optional mask input — alignment, color compensation, and noise fitting then use only the redrawn region's background, and only change islands overlapping that region are kept, whole (the mask is used exactly as drawn)
  • chore(tooling): Add scripts/check_node_consistency.py and document the naming rule in nodes/__init__.py; nodes/**/*.py is the single source of truth, and the checker fails on duplicate node_id/display_name, missing or orphan web/docs entries, README node-list drift, and a pyproject.toml / changelog version mismatch

v3.27.0

  • feat(logic): Add Image Area Compare to compare the pixel area (width × height) of two IMAGE inputs with a selected operator and output a boolean
  • fix(image_resize): Remove the duplicate Image Resize Qwen Image 3.0 node, which was identical to Image Resize Qwen Image 3.0 Pro; switch existing workflows to the Pro node
  • refactor(image): Rework Image Alpha Edge so edge/feather modify the alpha channel only and never blur or recolour the subject's RGB; only newly visible pixels receive an alpha-aware colour extension instead of the old solid-region RGB bleed
  • docs: Add bilingual docs for Image Area Compare, update the Image Alpha Edge docs, and sync the node list

v3.26.0

  • feat(image): Add Image Alpha Edge to shrink/expand an alpha image's edge and feather it, with built-in RGB bleed for cleaning matte fringe (hole filling stays in the mask nodes)
  • feat(image_resize): Add Image Resize Doubao Seedream 5.0 Pro with exact official 1k/1.5k/2k presets, plus Image Resize Qwen Image 3.0 and Image Resize Qwen Image 3.0 Pro with 1k/2k presets and dynamic sizing
  • refactor(image_resize): Rebuild Image Resize GPT Image 2.0 1k/2k presets as exact multiples of 16 and make the 4k tier exactly proportional within the official limits and the 3840 longest-edge cap
  • refactor(color): Rename Image Color Restore outputs to image / unchanged_mask and default unchanged_threshold to 0 (auto)
  • docs: Add bilingual docs for the new image and resize nodes, and sync the node list

v3.25.0

  • feat(color): Add Image Color Restore to restore an aligned AI edit's colors to the reference while smoothing removal/occlusion seams
  • feat(mask): Add Mask Levels for grayscale levels (black/white point) adjustment with inversion and binary fallback
  • refactor(image): Adjust Image Align Change Mask expand/feather defaults to 16/8
  • docs: Add bilingual docs for the two new nodes and sync the node list

v3.24.0

  • feat(image_resize): Unify auto (tier by input area → closest aspect within tier) across GPT Image 2.0 / Gemini30Pro / Gemini31Flash / Jimeng resize nodes
  • feat(image_resize): Jimeng auto excludes [2.0_pro] presets
  • docs: Sync bilingual docs and node list for the four resize nodes

v3.23.1

  • fix(io): Make Load Video and Load Video to Image resolve absolute paths and paths relative to the ComfyUI input directory, including directory scans and cache invalidation
  • fix(io): Preserve alpha-capable MOV/video streams when loading and saving, including ProRes 4444 output for processed transparent MOV files
  • fix(io): Keep source video streams when Save Video receives a source_path, avoiding unnecessary re-encoding that can drop transparency

v3.23.0

  • feat(int): Add Int Video Count to count connected videos from dynamic video_X inputs
  • docs: Add bilingual documentation for Int Video Count and synchronize the node list
  • feat(conversion): Add Video to URL to upload ComfyUI VIDEO objects to a public URL with content-based caching
  • feat(conversion): Make Image to URL's image input optional, returning an empty string when unconnected

v3.22.0

  • feat(conversion): Add Video to Image for extracting image, audio, FPS, and frame-count outputs from ComfyUI VIDEO objects

v3.21.2

  • feat(image_resize): Update Image Resize GPT Image 2.0 with 1:3–3:1 aspect-ratio presets and revised 1k/2k/4k resolutions aligned with GPT Image 2 size limits

v3.21.1

  • feat(image_crop): Image Edge Crop/Pad now supports RGBA (transparent) input — pad defaults to transparency, crop preserves alpha; supports transparent keyword and 4-component color strings (e.g. 255,0,0,128)

v3.21.0

  • feat(logic): Add Text Compare for string equality, containment, prefix/suffix, and regex boolean matching
  • feat(text): Expand String Ratio Gpt20Image and String Ratio Gemini31FlashImage supported ratio sets to match the latest selector presets
  • refactor(image_resize): Align Image Resize Gemini30ProImage and Image Resize Gemini31FlashImage preset families with the updated ratio ordering

v3.20.0

  • feat(image): Add a simplified Image BW Matte that fixes the pipeline to auto + soft and keeps only gamma, inward edge shrink, and blur controls
  • feat(mask): Add Mask Stroke for outer stroke mask with percentage width and union output

v3.19.1

  • feat(conversion): Add Image to Mask for optional image luminance-to-mask conversion with a 64x64 black fallback
  • refactor(conversion): Move Mask to Image from mask operations into the conversion group

v3.18.1

  • fix(image_resize): Correct Image Resize Gemini31FlashImage [1k] 4:1 preset to 2064x512

v3.18.0

  • feat(logic): Add Int Number Compare and Float Number Compare for boolean results from two numeric inputs and a selected comparison operator

v3.17.1

  • feat(image_blend): Add edge/e edge-average color and extend/ex edge-extension filling to Image Mask Blend.background_color

v3.17.0

  • feat(image_crop): Add Image Pad By BBox Mask for restoring a local image into a full mask canvas by bbox and filling uncovered areas with a configurable color strategy

v3.16.0

  • feat(image_resize): Add Image Resize GPT Image 2.0, a GPT Image 2.0 size adapter with Gemini-style auto preset matching, dynamic aspect-preserving tiers, and 1k/2k/4k fixed presets

v3.15.1

  • refactor(io): Add psd scanning, list-form filename / file_path outputs, and filename_suffix control to Get File Count; align Load Image.filename and Load PS.layer_name to string-list outputs
  • refactor(logic): Add automatic typed output for simple numeric and boolean values in text_match_value

v3.15.0

  • feat(io): Add Load PS for loading a single PSD/PSB as one indexed layer, all layer batches, or the merged document image, with drag-and-drop upload and node preview
  • refactor(batch): Reshape batch sampling nodes by adding step and num_frame=0 support to Image Batch Range / Mask Batch Range, and replacing Image Batch Extract with Image Batch Index and Image Batch Uniform
  • refactor(text): Enhance List Custom Int and List Custom Float with more flexible multiline value/range/stride parsing

v3.14.0

  • feat: Add utility nodes for batch image lists, image counting, list interleaving, and mask bounding boxes
  • fix: Improve Image Mask Crop channel output and alpha handling

v3.13.0

  • feat(text): Add String Ratio Gpt20Image with the GPT 2.0 image ratio set (3:2, 1:1, 2:3, 5:4, 4:5, 16:9, 9:16, 21:9, 3:4, 4:3)

v3.12.0

  • feat(text): Rename String Image Ratio Gemini31FlashImage to String Ratio Gemini31FlashImage, rename String Image Resolution to String Resolution, and align related file/node/doc names

v3.11.0

  • feat(image): Add Image Batch Interleave for contiguous segment splitting followed by column-first interleaved batch reordering

v3.10.0

  • feat(image): Add Image Alpha Join

v3.9.0

  • feat(image): Add Image Alpha Split for extracting alpha masks while flattening transparent images onto configurable solid backgrounds

v3.8.0

  • feat(conversion): Add Image to URL with auto, kefan, and data URL output modes

v3.7.2

  • build(repo): rewrite legacy fonts/*.otf history into Git LFS to reduce clone size; no node behavior change

v3.7.1

  • feat(image_resize): Add Image Resize Square with square presets for 256/512/1024/2048/4096, plus auto and auto (0.5k | 1k) matching

v3.6.0

  • feat(mask): Add Mask to Image with configurable black/white area colors, fill_hole, and optional RGBA output

v3.5.2

  • refactor(io): Adjust Save Text naming settings and filename generation logic

v3.5.1

  • refactor(io): Optimize video load/save compatibility and keep Save Image preview available when save_output is disabled

v3.5.0

  • feat(io): Add Load Text and Save Text nodes

v3.4.0

  • feat(nodes): Add Image Resize Gemini30ProImage, Image Resize Gemini31FlashImage, Detect Remove BG, Detect Remove BG Refine, and text_match_rownum; reorganize resize-related capabilities into the dedicated image_resize group, and sync related docs plus README node list updates.

v3.3.0

  • feat(conversion): Add Mask to SAM3 Box and Text to Any nodes

v3.2.6

  • refactor(condition): Optimize Text Encode QwenImageEdit node

v3.2.3

  • refactor(io): Optimize IO group

v3.2.1

  • feat(io): Add video loading progress bar

v3.2.0

  • refactor(io): Refactor IO group

v3.1.0

  • feat(io): Add Save Video by Image node for turning IMAGE batches into video files
  • refactor(color): Enhance Match Brightness Contrast sequence consistency options

v3.0.8

  • refactor(image_crop): Optimize Image Crop With BBox Mask node
  • refactor(color): Optimize Match Brightness Contrast node

v3.0.7

  • feat(color): Add Match Brightness Contrast node

v3.0.6

  • feat(mask): Add Mask Repeat node, supporting batch mask repetition and inversion
  • feat(io): Add Get File Count, Load Image From Folder, and Load Video From Folder nodes to optimize file loading logic and stability

v3.0.5

  • feat(image_tile): Merged Image Tile Split and Image Tile Split Preset nodes into a unified Image Tile Split node

v3.0.2

  • feat(text): Add String Filter node
  • feat(text): Add String Join Multi node
  • feat(conversion): Add Text List to String node

v3.0.1

  • feat(image): Add Image PingPong node for bidirectional batch frame generation with pre-reverse, link-frame removal, and frame truncation
  • feat(audio): Add Audio Duration node to get audio length (seconds)

v3.0.0

  • build: bump version to 3.0.0
</details> <details> <summary><b> 2.x releases</b></summary>

v2.0.5

  • feat(multi): Added Multi String Join, Multi Image Batch, Multi Mask Batch, Multi Image Stitch
  • feat(image): Added Image Three Stitch node
  • feat(condition): Added Text Encode QwenImageEdit Keep Size node

v2.0.3

  • feat(image_blend): Added output_mask_invert parameter to Image Mask Blend for output-only mask inversion

v2.0.0

  • breaking: Major update. Existing workflows built with previous nodes require reconfiguration after upgrading to function correctly. Please review nodes and parameters carefully.
</details> <details> <summary><b> 1.x releases</b></summary>

v1.2.46

  • feat(detection): Added DetectGuideLine node

v1.2.45

  • feat(detection): Added DetectYolo node for YOLO model object detection

v1.2.44

  • feat(util): Added Workflow Name node

v1.2.43

  • feat(logic): Added Video Cut Group node

v1.2.42

  • feat(logic): Added Image Batch Extract node for extracting specific images from batches with multiple modes including custom indices, step intervals, and uniform distribution

v1.2.40

  • feat(text): Added IntWan node for generating 4n+1 arithmetic progression sequences with configurable step control and range validation
  • refactor(logic): Enhanced Image Batch Split and Mask Batch Split nodes with improved boundary condition handling and comprehensive error recovery
<details> <summary><b>v1.2.39</b></summary>
  • feat(logic): Added Image Batch Group node for intelligent image batch splitting with overlap support and flexible padding strategies
</details> <details> <summary><b>v1.2.38</b></summary>
  • refactor(image): Enhanced ImageResizeUniversal node with comprehensive mask processing logic
</details> <details> <summary><b>v1.2.37</b></summary>
  • feat(image): Enhanced Image Solid node with advanced color parameter
  • feat(image): Added the ImageGridSplit node for splitting images into grid layouts with flexible output options
</details> <details> <summary><b>v1.2.36</b></summary>
  • feat(conversion): Enhanced URL to Video node
</details> <details> <summary><b>v1.2.35</b></summary>
  • feat(image): Added Image Resize Qwen Image node for Qwen vision model optimized image resizing with 7 preset resolutions and automatic aspect ratio selection
</details> <details> <summary><b>v1.2.32</b></summary>
  • feat(image): Added Image Solid Flux Kontext node for generating solid color images with Flux Kontext dimension presets
  • feat(image): Added Image Solid Qwen Image node for generating solid color images with QwenImage dimension presets
</details> <details> <summary><b>v1.2.31</b></summary>
  • fix: Fixed various bugs and improved stability
</details> <details> <summary><b>v1.2.28</b></summary>
  • feat(mask): Added Mask Paste by BBox Mask node for simplified mask pasting with automatic base mask creation and bounding box detection
  • feat(image_tile): Added Image Tile Split Preset node with predefined resolution presets and intelligent tile size selection
  • feat(image): Added Image Rotate with Mask node for advanced image rotation with mask support and multiple fill modes
  • feat(text): Enhanced Text Load Local node with user_prompt parameter for combining JSON content with additional user prompts
</details> <details> <summary><b>v1.2.26</b></summary>
  • feat(image_crop): Enhanced Image Crop with BBox Mask node with precise dimension control, added crop_to_side and crop_to_length parameters
</details> <details> <summary><b>v1.2.25</b></summary>
  • feat(image_crop): Added apply_paste_mask parameter to Image Paste by BBox Mask node for controlling smart scaling behavior
</details> <details> <summary><b>v1.2.24</b></summary>
  • feat(image_crop): Added opacity parameter to Image Paste by BBox Mask node for controlling paste image transparency
  • feat(image): Enhanced Image Stroke by Mask node with batch processing support for handling multiple images and masks
</details> <details> <summary><b>v1.2.23</b></summary>
  • fix(image): Enhanced Image Stroke by Mask node color parsing logic, supporting RGB string formats and improved default fallback to white color
  • fix(image): Enhanced Image Paste by BBox Mask node rotation parameter
</details> <details> <summary><b>v1.2.21</b></summary>
  • feat(text): Added Text Filter Comment node for filtering single-line comments (starting with #) and multi-line comments (wrapped in triple quotes), preserving non-comment blank lines
  • feat(text): Added Text Join by Text List node for merging any type of list into a string with support for prefix, suffix, and custom separators
  • refactor(text): Refactored Text Format node to Text Prefix Suffix, optimizing wildcard input processing and formatting functionality
</details> <details> <summary><b>v1.2.18</b></summary>
  • feat(sample): Added Step Split node for high-low frequency sampling step separation with support for percentage and integer input modes
</details> <details> <summary><b>v1.2.17</b></summary>
  • feat(image_crop): Optimized Image Crop with BBox Mask node
</details> <details> <summary><b>v1.2.15</b></summary>
  • feat(text): Added Text Join Multi node for concatenating multiple text inputs with dynamic variable referencing
  • feat(image_crop): Added Image Edge Crop Pad node with smart edge cropping and padding capabilities, featuring mask output functionality
  • feat(image_blend): Enhanced Image Luma Matte node with feathering and alpha output features, supporting multiple color formats and edge processing
</details> <details> <summary><b>v1.2.13</b></summary>
  • feat(text): Added Text Load Local node for loading JSON format prompt files from prompt directory with bilingual Chinese and English output
</details> <details> <summary><b>v1.2.12</b></summary>
  • feat(text): Added Text Format node for flexible text formatting with wildcard input support
</details> <details> <summary><b>v1.2.9</b></summary>
  • feat(image_crop): Refactored Image Crop with BBox Mask node
</details> <details> <summary><b>v1.2.8</b></summary>
  • feat(image): Added Image Resize Flux Kontext node with support for automatic and manual size selection for images and masks
  • feat(image): Enhanced Image Edit Stitch node with improved stitching algorithms and parameter handling
</details> <details> <summary><b>v1.2.7</b></summary>
  • feat(text): Added List Custom Seed node for creating unique random seed lists with control after generate functionality
</details> <details> <summary><b>v1.2.6</b></summary>
  • feat(image_hlfreq): Added high-low frequency separation node group, including Image HLFreq Separate, Image HLFreq Combine, and Image HLFreq Transform nodes with support for RGB, HSV, and IGBI frequency separation methods
</details> <details> <summary><b>v1.2.5</b></summary>
  • feat(mask): Added the Mask Fill Hole node, which fills holes in enclosed areas of masks with support for batch processing.
</details> <details> <summary><b>v1.2.3</b></summary>
  • fix(image_blend): Fixed the issue of inconsistency between devices for the Image Blend Modes by Alpha node.
</details> <details> <summary><b>v1.2.2</b></summary>
  • feat(image): Added the Image BBox Overlay by Mask node, which overlays the image bounding box based on a mask.
</details> <details> <summary><b>v1.2.1</b></summary>
  • refactor(image/crop): Renamed node classes and updated related documentation
  • feat(image_crop): Enhanced the functionality and output options of the ImageCropByMaskAlpha node
</details> <details> <summary><b>v1.2.0</b></summary>
  • feat: Added conversion, and restructured image mixing and masking processing
</details> <details> <summary><b>v1.1.6</b></summary>
  • feat(ImageEditStitch): Add a “spacing” parameter to control the distance between stitched images
</details> <details> <summary><b>v1.1.5</b></summary>
  • feat: Added text processing and logic nodes, optimized existing node functions
  • refactor(util): Refactored utility nodes, renamed nodes RangeMapping and PathBuild
  • feat(logic): Added ImageListAppend node for image list merging
  • feat(text): Added TextCustomList and TextCustomExtract text processing nodes
  • style: Cleaned up node parameter labels to maintain simplicity and consistency
</details> <details> <summary><b>v1.1.2</b></summary>
  • feat(image_tile): Improved the Image Tile Merge algorithm, using weight masks and cosine gradients to achieve perfect seamless stitching
</details> <details> <summary><b>v1.1.1</b></summary>
  • feat (image_crop): Added intelligent batch processing for Image BBox Paste
</details> <details> <summary><b>v1.1.0</b></summary>
  • build: Add new tile nodes
  • feat: Update node functionality
  • docs: Add bilingual documentation, improve node descriptions
</details> <details> <summary><b>v1.0.5</b></summary>
  • Add Path Select
</details> <details> <summary><b>v1.0.4</b></summary>
  • Fix Image Cropped Paste error, add batch processing feature.
</details> </details>

📋 Node List

🖼️ Image Processing Nodes

| Node Name | Description | |-----------|--------------| | Image Solid | Generate solid color images with enhanced color parameter supporting multiple input formats and dimension presets | | Image Alpha Join | Combine optional image and mask into a 4-channel image; an empty image yields an empty output, and a missing mask falls back to full alpha | | Image Alpha Split | Extract alpha as mask while flattening RGBA or alpha images onto a configurable solid background; an empty image yields empty outputs | | Image Rotate with Mask | Advanced image rotation with mask integration, multiple fill modes, and mask center rotation options | | Image Edit Stitch | Image stitching and merging with multiple stitching modes | | Image Main Stitch | Main-image stitcher supporting dynamic image_2..image_N, direction, size matching, spacing, and padding | | Image Add Label | Add text labels to images with automatic alpha preservation for RGBA inputs | | Image Plot | Image plotting and visualization tools with automatic RGBA preservation for alpha inputs | | Image Stroke by Mask | Apply stroke effects to mask regions with customizable width and color | | Image BBox Overlay by Mask | Mask-based image bounding box overlay with independent and merge modes | | Image Alpha Clean | Clean image alpha edge noise with simple strength presets and optional detection-only output | | Image Alpha Edge | Shrink or expand an alpha image's edge and feather it, with built-in RGB bleed to remove matte fringe; hole filling is left to the mask nodes | | Image BW Matte | Generate a cutout-friendly mask with a fixed auto + soft pipeline, plus gamma, inward shrink, and blur controls | | Image Align Change Mask | Align an AI-edited image to its original and extract clean/raw change masks, a change score, and an overlay while rejecting small global color drift | | Image Blur | Whole-image Gaussian blur with fractional radius, per-frame batch processing, and pass-through when blur is 0 | | Image Shift | Shift a whole image batch by shift_x / shift_y with cyclic wrap or a background_color fill (same syntax as Image Alpha Split, sm included), and return the shifted mask plus a blurred seam border_mask for seamless tiling; an optional wrap pad (0 = off, with a pad_mask marking the added frame) gives seams that run off the canvas outside context before inpainting, and channel count is preserved |

📐 Image Resize Nodes

| Node Name | Description | |-----------|-------------| | Image Resize Flux Kontext | Resize images/masks to Flux Kontext preset sizes with auto/manual target selection | | Image Resize Qwen Image | Qwen-oriented image resizing with preset resolutions and auto matching | | Image Resize Universal | Universal image resizing with ratio inference, fit strategies, and synchronized mask output | | Image Resize Jimeng | Resize for Jimeng-oriented presets with area-tiered auto matching and 1k/2k/4k fixed presets | | Image Resize Gemini 3.0 Pro Image | Gemini 3.0 Pro preset size adapter with area-tiered auto matching and synchronized image/mask transforms | | Image Resize Gemini 3.1 Flash Image | Gemini 3.1 Flash preset size adapter with area-tiered auto matching across 0.5k/1k/2k/4k tiers and extreme aspect ratios | | Image Resize GPT Image 2.0 | GPT Image 2.0 size adapter with area-tiered auto preset matching, dynamic aspect-preserving tiers, and 1k/2k/4k fixed presets | | Image Resize Doubao Seedream 5.0 Pro | Doubao Seedream 5.0 Pro size adapter with official 1k/1.5k/2k presets, tiered auto matching, dynamic sizing, and synchronized image/mask transforms | | Image Resize Qwen Image 3.0 Pro | Qwen Image 3.0 Pro size adapter with 1k/2k presets, tiered auto matching, dynamic sizing, and synchronized image/mask transforms | | Image Resize Square | General square size adapter with 256/512/1024/2048/4096 plus auto and auto (0.5k \| 1k) | | Image Resize Area | Area-based automatic resize that keeps the source aspect ratio: N = sqrt(area) is clamped to min/max and aligned to a multiple, and the output is scaled to area N^2 with both sides rounded to that multiple; reuses Image Resize Universal's fit / pad_color / mask behaviour and supports an optional lazy upscale_model |

🌈 Color Nodes

| Node Name | Description | |-----------|-------------| | Match Brightness Contrast | Adjusts the brightness and contrast of the source image to match the reference image | | Image Color Match | Transfer the color of a reference image onto a source image with wavelet/adain (native) or mkl/hm/reinhard/mvgd plus hybrids (color-matcher), supporting batch cycling | | Image Color Restore | Restore an aligned AI edit's colors to its original reference with local occlusion-seam smoothing; outputs the corrected image and a trusted unchanged mask |

🎨 Image Blending Nodes

| Node Name | Description | |-----------|-------------| | Image Mask Blend | Luminance-based image mask compositing with feathering, alpha output, and edge-extension background strategies | | Image Blend Mode by Alpha | Alpha-based image blending with multiple Photoshop-style blend modes | | Image Blend Mode by CSS | CSS standard blend modes implemented natively in PyTorch |

✂️ Image Cropping Nodes

| Node Name | Description | |-----------|-------------| | Image Mask Crop | Batch mask-based cropping with RGB/RGBA channel preservation and optional mask alpha output | | Image Crop Square | Square cropping with mask guidance and scaling support | | Image Crop With BBox Mask | Smart bounding box cropping with aspect ratio presets, an auto adaptive-framing mode (0 = mask-tight, 1 = whole image) that also survives long strips and dirty masks, scale strength and divisible-by control | | Image Paste By BBox Mask | Paste cropped images back with multiple blend modes | | Image Pad By BBox Mask | Restore a local image into the full bbox mask canvas and fill uncovered areas with a configurable color or strategy | | Image Edge Crop/Pad | Smart edge cropping and padding with multiple padding modes and mask output | | Image Grid Split | Split images into grid layouts with flexible row/column configuration and selective output options |

🧩 Image Tiling Nodes

| Node Name | Description | |-----------|-------------| | Image Tile Split | Image tile splitting with auto/grid/preset modes, overlap support, and reference image sizing | | Image Tile Merge | Image tile merging with intelligent stitching |

🌊 High-Low Frequency Separation Nodes

| Node Name | Description | |-----------|-------------| | Image HL Freq Separate | Advanced frequency separation node supporting RGB, HSV, and IGBI separation methods with precise high-low frequency image separation and automatic recombination | | Image HL Freq Combine | Advanced frequency recombination node supporting RGB, HSV, and IGBI recombination modes with intensity adjustment and intelligent batch matching | | Image HL Freq Transform | Advanced detail transfer node supporting IGBI, RGB, and HSV transfer methods for precise texture detail migration from detail images to generated images |

🎭 Mask Operation Nodes

| Node Name | Description | |-----------|-------------| | Mask Separate | Separate mask into individual connected regions with sorting and area filtering | | Mask To BBox Mask | Convert mask foregrounds into minimum bounding rectangle masks, with merged/separate output and outward divisible_by alignment | | Mask Fill Hole | Fill holes in enclosed areas of masks with batch processing support | | Mask Stroke | Create an optional hole-filled outer stroke mask with percentage width support and union output | | Mask Crop by BBox Mask | Mask bounding box cropping based on mask regions | | Mask Paste by BBox Mask | Simplified mask pasting with automatic base mask creation and bounding box detection | | Mask Repeat | Batch repeat masks with optional inversion support | | Mask Alpha Clean | Clean mask alpha noise and tiny islands with strength presets and optional detection-only output | | Mask Levels | Apply a levels adjustment to a grayscale mask, remapping the range via black/white points with inversion and binary fallback |

🔍 Detection Nodes

| Node Name | Description | |-----------|-------------| | Detect Remove BG | Multi-backend background removal (RMBG/BiRefNet/InSPyReNet) with image + alpha mask outputs | | Detect Remove BG Refine | RMBG mask post-refinement using original image + mask, outputs refined RGBA and alpha | | Detect Remove SolidBG UnMult | Solid-background UnMult keying: estimate the flat background from image borders, unmultiply to alpha, with optional low-alpha threshold cleanup and background composition | | Detect Yolo | YOLO model object detection with subfolder model support, customizable confidence thresholds, and optional label display control | | Detect Guide Line | Guide line detection combining Canny, HoughLinesP, and DBSCAN vanishing-point clustering |

🔧 Utility Nodes

| Node Name | Description | |-----------|-------------| | Workflow Name | Automatically retrieves current workflow filename with path control, custom prefixes/suffixes, and date formatting | | Range Mapping | Value range mapping tool supporting linear transformation and precision control for slider values | | Memory Cleanup | Requests ComfyUI to release execution cache and optionally unload models after the current task completes |

🔄 Conversion Nodes

| Node Name | Description | |-----------|-------------| | URL to Video | Convert video URLs to ComfyUI VIDEO objects with improved error handling, timeout control, and support for both synchronous and asynchronous download methods | | Video to Image | Extract image frames, audio, FPS, and frame count from ComfyUI VIDEO objects | | Image to URL | Convert image tensors into URL strings with auto, kefan, and data modes | | Video to URL | Upload ComfyUI VIDEO objects to a public URL with content-based caching | | Image to Mask | Convert an optional image into a luminance mask, falling back to a 64x64 black mask when empty | | Image Batch to List | Convert batch images to image lists for individual processing | | Image List to Batch | Convert image lists to batch images with automatic size normalization | | Mask to Image | Convert a mask into RGB/RGBA image output with hole filling, configurable black/white color mapping, and optional transparent background | | Mask Batch to List | Convert batch masks to mask lists for individual processing | | Mask List to Batch | Convert mask lists to batch masks with automatic size normalization | | Mask to SAM3 Box | Convert masks to SAM3 box prompts (normalized cx, cy, bw, bh), supporting positive/negative and merge/separate modes | | String Coordinate to BBoxes | Convert string format coordinates to BBOXES format with enhanced format support and improved SAM2 compatibility | | String Coordinate to BBox Mask | Convert string format coordinates to BBoxMask format with image dimension support and flexible output modes | | Seedream BBox to Mask | Invert Seedream <bbox> per-mille coordinates or a layer bounding_box into rectangular masks (bboxes only): coord_mode picks per-mille or pixel, separate emits per-box batches, merge unions them, always xyxy | | Text List to String | Merge a text list by applying per-item prefix/suffix and joining with a separator; supports escape sequences and composite separators | | Text to Any | Output text as a wildcard payload for flexible connections (e.g., COMBO inputs) | | Any to Any | Pass any value through unchanged as a wildcard (*), so outputs and inputs with different declared types can be connected |

🧠 Logic Nodes

| Node Name | Description | |-----------|-------------| | Any Empty Bool | Universal empty value checker (boolean output version) that checks if any type of input is empty and returns a boolean value | | Any Empty Int | Universal empty value checker (integer output version) that checks if any type of input is empty and returns custom integer values | | Any Switch Bool | Universal boolean switch node supporting any type input with lazy evaluation, selecting output based on boolean condition | | Any Switch Select | Multi-way selector node supporting multiple input options, lazily routing the chosen input_k to output and returning the effective select value | | Float Number Compare | Float comparison node that evaluates two FLOAT values with a selected operator and outputs a boolean | | Int Number Compare | Integer comparison node that evaluates two INT values with a selected operator and outputs a boolean | | Multi Switch Select | Multi-output selector for input_1..input_N; a 1-based select using the same range syntax as Image Batch Index (e.g. 1-3, [1,3], or a/all for every connected input) lazily routes each chosen input to its matching output_k, leaving the rest None | | Text Compare | Compares two strings with equality, containment, prefix/suffix, or regex operators and outputs a boolean | | Text Match Rownum | Match a single text against multi-line text and return first matched row number (1-based), or 0 when not found | | Text Match Value | Match single-line text against multi-line key-value pairs and return corresponding value with simple numeric/boolean auto-typing | | Image Area Compare | Compare the pixel area (width × height) of two IMAGE inputs with a selected operator and output a boolean | | Image Minimum Area | Upscale an image until its area reaches a minimum square reference area (N²), with crop/pad/stretch fit, divisible_by rounding, mask output, and a resize_bool flag | | String Random | Randomly pick one item from multiline custom text (split by dash lines, newlines, or punctuation) using a fixed seed |

🔢 Integer Nodes

| Node Name | Description | |-----------|-------------| | Int Image Count | Count valid images from dynamic image_X inputs, ignoring empty inputs, empty tensors, and all-zero images | | Int Mask Count | Count valid masks from dynamic mask_X inputs, ignoring empty inputs, empty tensors, and all-zero masks | | Int Image Side Length | Output selected side length (longest/shortest/width/height) from image dimensions | | Int Image Size | Output width and height integers from image dimensions | | Int Mask Side Length | Output selected side length (longest/shortest/width/height) from mask dimensions | | Int Video Count | Count non-empty videos from dynamic video_X inputs (list outputs counted per item) | | Int Audio Count | Count non-empty audio inputs from dynamic audio_X ports (list outputs counted per item) | | Int Text Count | Count non-empty STRING / INT / FLOAT inputs (BOOLEAN is treated as INT) from dynamic wildcard text_X ports | | Int Split | Split a total value into two parts, supporting percentage and integer split point | | Int Wan | Generate 4n+1 arithmetic sequences with step and range validation |

🧮 Float Nodes

| Node Name | Description | |-----------|-------------| | Float Mask Area Ratio | Output how much of the whole image a mask covers as a 0-1 float; the round input defaults to 2 decimals |

📦 Batch Processing Nodes

| Node Name | Description | |-----------|-------------| | Image Batch Index | Extract images from a batch using multiline index text with flexible list/range/stride syntax and reverse-order support | | Image Batch Split | Intelligent image batch splitter with forward/backward splitting modes and enhanced boundary condition handling | | Image Batch Group | Intelligent image batch grouper with configurable batch sizes, overlap handling, and flexible padding strategies | | Image Batch Range | Sample an image batch by start_index, step, and final num_frame; num_frame=0 takes all remaining images | | Image Batch Uniform | Uniformly sample an image batch by final num_frame; num_frame=0 returns all images | | Image Batch Interleave | Reorder an image batch by splitting it into contiguous segments and reading them back in column-first interleaved order | | Image List Interleave | Reorder an image list by contiguous segment interleaving while preserving each image's original size | | Image PingPong | Bidirectional frame repeat over batch, supports pre-reverse, link-frame removal, and frame truncation | | Image List Append | Image list appender for intelligently merging images into lists | | Mask Batch Math Ops | Batch mask mathematical operations | | Mask Batch Range | Sample a mask batch by start_index, step, and final num_frame; num_frame=0 takes all remaining masks | | Mask Batch Split | Intelligent mask batch splitter with forward/backward splitting modes and enhanced boundary condition handling | | Video Cut Group | Video scene cut detector that identifies scene transitions by analyzing frame similarity, supporting both fast and precise modes |

📝 Text Processing Nodes

| Node Name | Description | |-----------|-------------| | Text Prefix Suffix | Text prefix suffix formatter with wildcard input support for flexible data formatting with custom prefix and suffix | | Text Custom Extract | Text custom extractor for extracting specified key values from JSON | | Text Path Extract | Read values from JSON text, dicts or plain text by path such as [0].content, [-1].bbox[0] or [*].content, collect every match ([*].bbox), count with .count / .length / [#], and emit auto/json/text/csv text plus a raw value and a count output | | Text System Prompt | Pick one builtin system prompt from utils/system_prompt/<model>/[<lang>] <name>.txt and output its body verbatim on a single string output; bundled Qwen-Image-2.1 t2i / edit prompts, and any new .txt you drop in shows up after a reload | | String Ratio GPT Image 2.0 | Infer the nearest GPT 2.0 image supported aspect ratio from input images, or pass through the selected ratio when no image is connected | | String Ratio Gemini 3.1 Flash Image | Infer the nearest Gemini 3.1 Flash supported aspect ratio from input images, or pass through the selected ratio when no image is connected | | String Ratio Jimeng | Infer the nearest Jimeng supported aspect ratio from input images, or pass through the selected ratio when no image is connected | | String Ratio Doubao Seedream 5.0 Pro | Infer the nearest Doubao Seedream 5.0 Pro supported aspect ratio from input images, or pass through the selected ratio when no image is connected | | String Ratio Qwen Image 3.0 Pro | Infer the nearest Qwen Image 3.0 Pro supported aspect ratio from input images, or pass through the selected ratio when no image is connected | | String Resolution GPT Image 2.0 | Infer the nearest GPT Image 2.0 resolution tier (1k / 2k / 4k) from input images, or pass through the selected label when no image is connected | | String Resolution Gemini 3.1 Flash Image | Infer the nearest Gemini 3.1 Flash Image resolution tier (0.5k / 1k / 2k / 4k) from input images, or pass through the selected label when no image is connected | | String Resolution Jimeng | Infer the nearest Jimeng resolution tier (1k / 2k / 4k) from input images, or pass through the selected label when no image is connected | | String Resolution Doubao Seedream 5.0 Pro | Infer the nearest Doubao Seedream 5.0 Pro resolution tier (1k / 1.5k / 2k) from input images, or pass through the selected label when no image is connected | | String Resolution Qwen Image 3.0 Pro | Infer the nearest Qwen Image 3.0 Pro resolution tier (1k / 2k) from input images, or pass through the selected label when no image is connected | | String Resolution Seedance 2.5 Multimodal Video | Infer the nearest Seedance 2.5 video resolution tier (480p / 720p / 1080p / 2k / 4k) from the input video, or pass through the selected label when no video is connected | | String Filter | Text cleaner supporting {input} / {n} substitution from the input value, comment filtering (# and triple quotes), and optional empty-line removal | | String Join Multi | Join up to 5 text blocks with {input} substitution, comment/empty-line filtering, and composite separators | | List Custom Int | Build integer lists from flexible multiline value/range/stride text with reverse-order and mixed punctuation support | | List Custom Float | Build float lists from flexible multiline value/range/stride text with reverse-order and mixed punctuation support | | List Custom String | Custom string list generator with dash separator and multiple delimiter support | | List Custom Seed | Custom seed list generator for creating unique random seed lists with control after generate functionality |

🔗 Multi Nodes

| Node Name | Description | |-----------|-------------| | Multi String Join | Concatenate dynamic string_X sockets (string_1 is socket-only) with {input} variable support and comment/triple-quote filtering; customizable separator | | Multi Image Batch | Build image batches from dynamic image_X with crop/pad/stretch size unification and automatic RGBA preservation when any input has alpha | | Multi Image List | Build image_list from dynamic image_X batch or list inputs, splitting them into single-image list items | | Multi Image Overlay | Overlay multiple image layers sequentially with alpha compositing, fit modes, and automatic RGBA preservation when any input has alpha | | Multi Image Stitch | Dynamic multi-image stitcher with direction, size matching, spacing, padding color control, and automatic RGBA preservation when any input has alpha | | Multi Mask Batch | Build mask batch from dynamic mask_X with crop/pad/stretch size unification and configurable gray padding | | Multi Mask Math Ops | Dynamic multi-mask operations (union/intersection/difference/XOR) with batch broadcasting and size alignment | | Multi Video Stitch | Stitch dynamic video_X inputs end-to-end into one video with crop/pad/stretch size unification, optional get_video_size reference size, and configurable pad color / fps |

📁 IO Nodes

| Node Name | Description | |-----------|-------------| | Get File Count | Count image/video/psd/text files in a folder with optional recursive scanning, list-form filename / file_path outputs, and configurable filename extension retention | | Load Image | Load images from a file or folder with batch mode, size unification, derived masks, and list-form filename output | | Load PS | Load a PSD/PSB file as one indexed layer, all layer batches, or the merged document image, with drag-and-drop upload, lazy cached layer previews (enabled by default), hidden-layer, group-mode controls, and list-form layer_name output | | Load Text | Load .txt content from a file or folder with upload, live preview, auto encoding detection, and indexed folder picking | | Load Video | Select a video from a file or folder and apply trimming/FPS settings during decoding | | Load Video to Image | Decode a video into an image batch, audio, fps, and frame count | | Save Audio | Save an AUDIO object as FLAC/WAV/MP3/Opus (default mp3) with output/temp routing, optional metadata, batch splitting, empty-input pass-through, and an in-node audio player preview | | Save Image | Save image batches to output/temp and return absolute saved file paths | | Save Text | Save text input as txt / csv / md / json (UTF-8) with auto-increment naming, metadata, and preview-only mode | | Save Video by Image | Encode an image batch into a video with optional audio muxing and alpha support | | Save Video | Save a VIDEO object to disk with container extension preservation and alpha preview |

🎛️ Conditioning Nodes

| Node Name | Description | |-----------|-------------| | Text Encode Qwen Image Edit | Qwen image-edit conditioning encoder combining vision inputs and text, supports size preservation modes and reference latents |

🔊 Audio Nodes

| Node Name | Description | |-----------|-------------| | Audio Duration | Get audio length (seconds) |

🙆 Acknowledgments

ComfyUI

ComfyUI-KJNodes

ComfUI-Easy-Use

ComfyUI_LayerStyle

Comfyui_TTP_Toolset

comfyUI_FrequencySeparation_RGB-HSV

comfyui_extractstoryboards

ComfyUI-VideoHelperSuite

ComfyUI-RH-Nodes

ComfyUI-Qwen-Image-2.1-Prompt-Enhancer (bundled Qwen-Image-2.1 system prompts, MIT)

🌟 Star

My gratitude extends to the generous souls who bestow a star.

<img src="imgs/Stargazers.png" alt="Stargazers" style="zoom:80%;" />