ComfyDeploy: How Compositor Node works in ComfyUI?

Create complex compositions the FAST and EASY way

How many times do you need to create something like this?

Well, This node was created to make the composition process quick and easy!

The Compositor Node

With the Compositor Node you can:

• Pass up to 8 images and visually place, rotate and scale them to build the perfect composition.
• Group move and group rescale the images. choose your transform center (corner, or center by using ctrl)
• Remember the position, scaling values and z-index across generations and easily easy swap images.
• Use the buffer zone to park an asset you don't want to use or easily reach transformations controls
• Clearly see the exported area through a green overlay
• Easily re-frame your shot via multiple selection scaling, resizing and re-positioning
• Flip an image via negative scaling (drag a corner towards and past the inside of the image)
• Mask your images quickly
• Precisely move selections with keyboard
• Use the information about transforms in other nodes (like conditioning set area)

Changelog

• head over to the wiki tab for more workflows and information!
• v 3.1.0 - 20.09.2024

  • new configuration feature: onConfigChange action toggle when you change the configuration (or any of the attached nodes) you can now choose if:

    • you want to stop the flow to allow edits
    • or you want to grab a capture and continue the flow
      $${\color{red}Important}$$! this option stops, uploads the composition and re-enqueues your prompt (there is no real pause) so careful with random seeds on the left of the node or it will loop!

  • new configuration feature: normalize height when this is activated your images will all be the same height of the canvas (this can lower image quality)

  • transforms enhancement: output transforms now give you back the angle and bounding box coordinates

  • transforms enhancement: you can force transform outputput values to be integers (as some nodes requires it)

  • new feature: (experimental and limited): Tools ! this is an experimental feature. it allows controlling some aspects of the compositor.

    • precise selection toggle ignore transparent pixels and select the first image below the mouse
    • center selected puts the selected images in the center of canvas
    • reset transforms zeroes out the changes to images bringing them to their original size, angle and 0,0 location (top left)
    • limitations: as saving discards the selection, and it happens on mouse out
      you might need to re-select to use centering and reset

Previous versions changelog

<details> <summary><b>click here to expand</b> the changelog...</summary>

• v 3.0.8 - 18.09.2024
  • new feature: invert mask option. the implementation of mask was not correct. now it's possible to invert the mask via toggle.
  • new feature: angle output the angle of rotation is now accessible in the output (and soon the bounding box x,y, width and height).
  • bugfix: fix cut images on swap due to wrongly preserved width and height
  • _new feature: added force int to allow the outputs to be used with set area conditioning (that requires int)
• v 3.0.4 - 18.09.2024 - bugfix: the width and height stored in transforms were swapped and the output node would report them incorrectly. thanks @sky958958 for spotting it
• v 3.0.2 - 17.09.2024 - friendly transforms for area prompting!
  With the goal of being able to do regional area prompting,
  now you can easily output each input x,y coordinates and their scaled width and height with the help of the new Transform Output node! select the channel corresponding the input and the node will output the values for you.
  • enhancement: a new node outputs x,y,width,height other images into a convenient node to be attached to the transforms output
  • enhancement: save and restore skew from transform (now you can distort your images to help fake perspective)
• v 3.0.0 - 16.09.2024 - this release is a full rewrite of the code and fixes:
  • issues #45 , #34, #18 also, and adds new features:
  • enhancement: simplified control panel (cature on queue, save transform, pause are removed as not needed anymore)
  • new feature: automatic upload of the output on mouse out of the canvas area (no need to click capture)
  • new feature: flash on save (once the image is uploaded the composition area green border briefly flashes in orange)
  • new feature: preliminary work for optional control panels (they will contain alignment controls, and other tools)
  • enhancement: enqueue with continue, on the first run, if necessary information is missing (like output) the flow will stop, make your composition, and click continue to re-enqueue the flash finishes.
  • v 2.0.4 - 06.09.2024 - enhancement: You can now scale the selected image via mouse wheel!
  • v 2.0.1 - 05.09.2024 - V2 is HERE!
    • enhancement: An all new widget layout with maximized working area and less clutter
    • new feature: A new companion configuration widget to allow more control and easier maintenance
    • enhancement: More control! it's now possible to select an image or group and then "alt+drag" to center scale and rotate
    • new feature: More control! it's now possible to nudge a selection by one pixel by using keyboard arrows, and while holding shift the movement is 10px! pixel perfect alignments!
    • new feature: the node now remembers the transforms you have applied, on the new run it will re-apply the stored transforms (storing transforms is controlled in the config)
    • new feature: masks are here! you can now pass masks, and they will be applied automatically! (depending on the results you might want still to invert them)
    • regression: a bit annoying but is_changed is not being observed so flows are re-triggered even on fixed
    • regression: img in workflow saved is not visible anymore
  • V 1.0.9 - 30.08.2024 - Huge refactoring!
    • new feature: multiple instances are now possible
    • bugfix: zooming out does not hide the compositor images anymore
    • bugfix: when saving a png with the workflow the compositor content is now visible (will not be restored...yet)
    • enhancement: the node does not re-trigger the execution of the flow if the image is not changed
    • performance: the node is now more efficient and correctly implements the is_changed check via checksum, avoiding re-triggering flows downstream if the composition has not changed
    • maintainability: the node is now refactored and better engineered, with a lot of comments. could be a good use case for those learning to code comfy extensions.
  • V 1.0.8 - 28.08.2024 - new feature: safe area indication - a green border is overlaid on top of the composition to indicate the exported area
  • V 1.0.7 - 28.08.2024 - new feature: preserve stacking order. when selecting a node, it's z-order is preserved image1 being the background/farthest and image8 the foreground/closest.
    • the first connected node will be the most distant from camera (background)
    • the last will be the closest to camera (subject/foreground)
  • V 1.0.4 - 27.08.2024 - new feature: now it's possible to pause the flow with a switch to avoid processing an unfinished composition

</details>

Setup

Method 1: git clone open the custom nodes directory in your editor and

git clone https://github.com/erosDiffusion/ComfyUI-enricos-nodes.git

like all other custom nodes (that are not integrated with manager)

Method 2: ComfyUi Manager In Comfy UI Manager search "Compositor" and select the node from erosDiffusion and press install.

Method 3: via manager's button open ComfyUI manager click on Install via Git URL and paste this url

https://github.com/erosDiffusion/ComfyUI-enricos-nodes.git

if you get: "This action is not allowed with this security level configuration" then check your manager config.ini as discussed here: and set the security to weak (at your risk)

Reasons and How To use

Why this node ?

• I wanted to learn how to create custom nodes with a GUI in ComfyUI
• be able to composite visually images in ComfyUI
• be able to have image inputs that are generated on the fly in the composition
• be able to remember sizing and position across usages/generations
• have more room to manipulate objects around/outside the generated image

Alternatives ?

• the painter node is great and works better and does a million things more, but it misses some of these features.
• continue compositing your image like caveman using pixel coordinates
• well...photoshop ** if you have it** and import via million clicks or with a plugin
• finally use Krita which is good powerful and free
• oh and Blender also has a great plugin but you need to know/learn blender

How to use

Method1:

• search "compositor" (v3) in the dropdown, connect with config (V3) by dragging from the node config slot.
• configure width, height and padding around node (it's used to be able to move beyond the generated image) the node should will resize when you run
• connect the inputs (suggested setup is to always have a fixed size via resize and rembg where needed)
• important: connect the output (save image, preview image,...)
• run once to get the inputs in the compositor (the flow will stop if there is no output)
• create your composition (see below)
• mouse out the composition area (green border flashes to orange as the image uploads)
• click continue to enqueue again (or enqueue)
• use the output ! (suggestion is to feed it to a depth anything v2 node and use it in a depth controlnet to guide your image)

Create your composition details:

• put your images in the dark gray area
• you can connect any flow (generation with fixed, static rgba, full rgb)
• anything in the dark gray area is rendered
• use up till 8 images, optionally pass masks
• background will be at first slot on top
• in v 1.0.9 and later the z-index is fixed, reconnect an input or move stuff around.
  it should be simpler to handle depth stacking

Advanced

• click to select
• drag (from a clear area) to select multiple
• use controls to rotate and scale
• drag selected to move (can also rescale the group)
• shift click to select multiple
• shift click to unselect selected in a group select
• if you choose to stop on config change, hit continue to re-enqueue. capture happens on mouse out from the composition or
• if you choose grabAndContinue then it will be automatic
• scroll up or down to scale a single image selection

Aupporting nodes I use with this one

• Rembg(batch) -> from https://github.com/Mamaaaamooooo/batchImg-rembg-ComfyUI-nodes.git -> extracts the subject and returns a rgba image
• any other technique to create masks (grounding dino, sam, florence2...)
• any controlnet depth for your model - works well with depth anything v2 preprocessor for both 1.5 (regular controlnet) and xl (via union controlnet) or lineart (like anylineart), for flux you can try x-labs controlnet (but it does not work well for me)

Demo Workflow for v3.1

Just throw the worst possible images you find on the internet or that you can generate... ...scale and align quick, give a depth controlnet, describe the full scene and style, render... and you will get:

with the V3.1 workflow in json format you are in pixel perfect positioning control of your scene and content ! Images to replicate are in the assets folder.

Final words and limitations

• limitation you need to run the flow once for the compositor to show images
• limitation careful on random values on the left of the node, the node stops the execution on config change to be able to grab a capture and re-enqueues the flow. if the cache is invalidated you not be able to go next see here https://github.com/erosDiffusion/ComfyUI-enricos-nodes/issues/63
  when I tried implementing threading pause it was not reliable, so I resorted to stop / restart. another option would be a while loop...but that feels not right.
• tools new tools only show up on load, so if you add them, reload page with browser reload
• known issue: the compositing is not scaled, so if you want a 5k image well... I hope you have a big enough monitor, but it's not (yet) the goal of this node...

Now go put a fairy in a forest!

yours, ErosDiffusion 💜