KJNodes

Simplified skip layer guidance that only skips the uncond on selected blocks

Works as a bridge to the AudioScheduler -nodes: https://github.com/a1lazydog/ComfyUI-AudioScheduler Transforms image based on the normalized amplitude.

Works as a bridge to the AudioScheduler -nodes: https://github.com/a1lazydog/ComfyUI-AudioScheduler Creates a list of floats from the normalized amplitude.

Works as a bridge to the AudioScheduler -nodes: https://github.com/a1lazydog/ComfyUI-AudioScheduler Creates masks based on the normalized amplitude.

Works as a bridge to the AudioScheduler -nodes: https://github.com/a1lazydog/ComfyUI-AudioScheduler Offsets masks based on the normalized amplitude.

Reacts to the sound level of the input. Uses your browsers sound input options and requires. Meant to be used with realtime diffusion with autoqueue.

Creates controlnet weights from a list of floats for Advanced-ControlNet

NO LONGER NECESSARY OR FUNCTIONAL, keeping node for backwards compatibility. Use the TorchCompileModelAdvanced to use LoRA with torch.compile.

This node has been replaced with TorchCompileModelAdvanced node, please use that instead.

This node has been replaced with TorchCompileModelAdvanced node, please use that instead.

This node has been replaced with TorchCompileModelAdvanced node, please use that instead.

This node has been replaced with TorchCompileModelAdvanced node, please use that instead.

This node has been replaced with TorchCompileModelAdvanced node, please use that instead.

This node has been replaced with TorchCompileModelAdvanced node, please use that instead.

DEPRECATED, use the native EasyCache or alternative custom node that's up to date instead of this.

https://github.com/WeichenFan/CFG-Zero-star

Fast image preview using binary websocket, bypassing base64/JSON overhead.

Select individual block alpha values, value of 0 removes the block altogether

This node allows scheduling GLIGEN text box positions in a batch, to be used with AnimateDiff-Evolved. Intended to pair with the Spline Editor -node. GLIGEN model can be downloaded through the Manage's "Install Models" menu. Or directly from here: https://huggingface.co/comfyanonymous/GLIGEN_pruned_safetensors/tree/main Inputs: - **latents** input is used to calculate batch size - **clip** is your standard text encoder, use same as for the main prompt - **gligen_textbox_model** connects to GLIGEN Loader - **coordinates** takes a json string of points, directly compatible with the spline editor node. - **text** is the part of the prompt to set position for - **width** and **height** are the size of the GLIGEN bounding box Outputs: - **conditioning** goes between to clip text encode and the sampler - **coord_preview** is an optional preview of the coordinates and bounding boxes.

Interpolates coordinates based on a curve.

https://github.com/time-to-move/TTM

Adds callbacks to model to set torch settings before and after running the model.

Experimental node for patching attention mode to use NABLA sparse attention for video models, currently only works with Kadinsky5

Experimental node for patching attention to use flash attention, without the silent SDPA fallback the ComfyUI default does. Patches the attention of the model passing through this node; to disable, bypass or disconnect this node. Requires the flash_attn library to be installed.

Experimental node for patching attention mode. This doesn't use the model patching system and thus can't be disabled without running the node again with 'disabled' option.

PiD 4-step decoder color/brightness drift corrector. Subtracts a per-channel bias from x0_pred at the first sampling step, using a small linear model calibrated against the model's systematic drift (model tends to brighten dark scenes and add a blue cast).

Plots coordinates to sequence of images using Matplotlib.

# WORK IN PROGRESS Do not count on this as part of your workflow yet, probably contains lots of bugs and stability is not guaranteed!! ## Graphical editor to create coordinates **Shift + click** to add a positive (green) point. **Shift + right click** to add a negative (red) point. **Right click on a point** to delete it. **Ctrl + click** to draw a bounding box. **Drag bbox corners** to resize, **drag inside** to move. **Right click on bbox** to delete it. To add an image select the node and copy/paste or drag in the image. Or from the bg_image input on queue (first frame of the batch). **THE IMAGE IS SAVED TO THE NODE AND WORKFLOW METADATA** you can clear the image from the context menu by right clicking on the canvas

StyleModelApply but with strength parameter

Allow scheduling of the azimuth and elevation conditions for SV3D. Note that SV3D is still a video model and the schedule needs to always go forward https://huggingface.co/stabilityai/sv3d

Extends the potential frame count of HunyuanVideo using this method: https://github.com/thu-ml/RIFLEx

Select individual block alpha values, value of 0 removes the block altogether

color-matcher enables color transfer across images which comes in handy for automatic color-grading of photographs, paintings and film sequences as well as light-field and stopmotion corrections. The methods behind the mappings are based on the approach from Reinhard et al., the Monge-Kantorovich Linearization (MKL) as proposed by Pitie et al. and our analytical solution to a Multi-Variate Gaussian Distribution (MVGD) transfer in conjunction with classical histogram matching. As shown below our HM-MVGD-HM compound outperforms existing methods. https://github.com/hahnec/color-matcher/

Creates a gradient image from coordinates.

Creates an image or batch of images with the specified shape. Locations are center locations.

Cuts the masked area from the image, and drags it along the path. If inpaint is enabled, and no bg_image is provided, the cut area is filled using cv2 TELEA algorithm.

Returns a range of images from a batch.

Returns width, height and batch size of the image, and passes it through unchanged.

Selects and returns the images at the specified indices as an image batch.

Returns latent tensor dimensions, and passes the latent through unchanged.

Calculates list of floats from image.

Add blends multiple images together. You can set how many inputs the node has, with the **inputcount** and clicking update.

Helper node for video generation extension First input source and overlap amount to get the starting frames for the extension. Then on another copy of the node provide the newly generated frames and choose how to overlap them.

Removes empty images from a batch

Transitions between two batches of images, starting at a specified index in the first batch. During the transition, frames from both batches are blended frame-by-frame, so the video keeps playing.

Creates an image batch from multiple images. You can set how many inputs the node has, with the **inputcount** and clicking update.

Concatenates images from a batch into a grid with a specified number of columns.

Crops the input images based on the provided mask.

Crops the input images based on the provided masks.

Captures an area specified by screen coordinates. Can be used for realtime diffusion with autoqueue.

Concatenates the 4 input images into a 2x2 grid.

Concatenates the 9 input images into a 3x3 grid.

Converts a grid of images to a batch of images.

Normalize the images to be in the range [-1, 1]

Creates an image list from the input images.

Same as ComfyUI native model upscaling node, but allows setting sub-batches for reduced VRAM usage. Optionally downscale the result with a ratio.

Repeats each image in a batch by the specified number of times. Example batch of 5 images: 0, 1 ,2, 3, 4 with repeats 2 becomes batch of 10 images: 0, 0, 1, 1, 2, 2, 3, 3, 4, 4

Pad the input image and optionally mask with the specified padding.

Passes the image through without modifying it.

This node is designed to be use with node FilterZeroMasksAndCorrespondingImages It inserts the images_to_insert into images according to insert_indexes Returns: images_after_insert: updated original images with origonal sequence order

Inserts images at the specified indices into the original image batch.

Merges channel data into an image.

Inserts empty frames between the images in a batch.

Remaps the image values to the specified range.

Replaces the images in a batch, starting from the specified start index, with the replacement images.

DEPRECATED! Due to ComfyUI frontend changes, this node should no longer be used, please check the v2 of the node. This node is only kept to not completely break older workflows.

Resizes the image to the specified width and height. Size can be retrieved from the input. Keep proportions keeps the aspect ratio of the image, by highest dimension.

Reverses the order of the images in a batch.

Saves an image and mask as .PNG with the mask as the alpha channel.

Captures an area specified by screen coordinates. Can be used for realtime diffusion with autoqueue.

Splits image channels into images where the selected channel is repeated for all channels, and the alpha as a mask.

Creates transitions between images in a batch.

Creates transitions between images.

Appends tracking data to be used with InstanceDiffusion: https://github.com/logtd/ComfyUI-InstanceDiffusion

Creates tracking data to be used with InstanceDiffusion: https://github.com/logtd/ComfyUI-InstanceDiffusion InstanceDiffusion prompt format: "class_id.class_name": "prompt", for example: "1.head": "((head))",

Draws the tracking data from CreateInstanceDiffusionTracking -node.

Generates an empty latent image with the specified dimensions. The choices are loaded from 'custom_dimensions.json' in the nodes folder.

Returns a range of latents from a batch.

Selects and returns the latents at the specified indices as an latent batch.

Inserts a latent at the specified index into the original latent batch.

Select individual block alpha values, value of 0 removes the block altogether

https://github.com/NUS-HPC-AI-Lab/Enhance-A-Video

Downloads and loads CLIPSeg model with huggingface_hub, to ComfyUI/models/clip_seg

Segments an image or batch of images using CLIPSeg.

Returns selected index from bounding box list as integers.

Visualizes the specified bbox on the image.

Creates a block mask by dividing the bounding box of each mask into blocks of the specified size and filling in blocks that contain any part of the original mask.

Converts chosen RGB value to a mask. With batch inputs, the **per_batch** controls the number of images processed at once.

Consolidates a batch of separate masks by finding the largest group of masks that fit inside a tile of the given width and height (including the padding), and repeating until no more masks can be combined.

Applies the provided masks to the input images with Alpha Blending support.

Filter out all the empty (i.e. all zero) mask in masks Also filter out all the corresponding images in original_images by indexes if provide original_images (optional): If provided, need have same length as masks.

Returns the width, height and batch size of the mask, and passes it through unchanged.

# GrowMaskWithBlur - mask: Input mask or mask batch - expand: Expand or contract mask or mask batch by a given amount - incremental_expandrate: increase expand rate by a given amount per frame - tapered_corners: use tapered corners - flip_input: flip input mask - blur_radius: value higher than 0 will blur the mask - lerp_alpha: alpha value for interpolation between frames - decay_factor: decay value for interpolation between frames - fill_holes: fill holes in the mask (slow)

Creates an image batch from multiple masks. You can set how many inputs the node has, with the **inputcount** and clicking update.

Offsets the mask by the specified amount. - mask: Input mask or mask batch - x: Horizontal offset - y: Vertical offset - angle: Angle in degrees - roll: roll edge wrapping - duplication_factor: Number of times to duplicate the mask to form a batch - border padding_mode: Padding mode for the mask

Sets new min and max values for the mask.

Resizes the mask or batch of masks to the specified width and height.

Rounds the mask or batch of masks to a binary mask. <img src="https://github.com/kijai/ComfyUI-KJNodes/assets/40791699/52c85202-f74e-4b96-9dac-c8bda5ddcc40" width="300" height="250" alt="RoundMask example">

Separates a mask into multiple masks based on the size of the connected components.

Splits the specified bbox list at the given index into two lists.

Combines multiple conditioning nodes into one

Bundles multiple conditioning mask and combine nodes into one,functionality is identical to ComfyUI native nodes

Bundles multiple conditioning mask and combine nodes into one,functionality is identical to ComfyUI native nodes

Bundles multiple conditioning mask and combine nodes into one,functionality is identical to ComfyUI native nodes

Bundles multiple conditioning mask and combine nodes into one,functionality is identical to ComfyUI native nodes

Create a batch of masks interpolated between given frames and values. Uses same syntax as Fizz' BatchValueSchedule. First value is the frame index (not that this starts from 0, not 1) and the second value inside the brackets is the float value of the mask in range 0.0 - 1.0 For example the default values: 0:(0.0) 7:(1.0) 15:(0.0) Would create a mask batch fo 16 frames, starting from black, interpolating with the chosen curve to fully white at the 8th frame, and interpolating from that to fully black at the 16th frame.

Creates a mask or batch of masks with the specified shape. Locations are center locations. Grow value is the amount to grow the shape on each frame, creating animated masks.

Creates a mask or batch of masks with the specified shape. Locations are center locations.

Creates a mask or batch of masks with the specified text. Locations are center locations.

Generates a batch of masks based on the input float values. The batch size is determined by the length of the input float values. Each mask is generated with the specified width and height.

Overrides the memory usage factor of the model during sampling.

Adds callbacks to model to report memory usage during after sampling

Returns the inputs unchanged, they are only used as triggers, and performs comfy model management functions and garbage collection, reports free VRAM before and after the operations.

Simply passes through the positive and negative conditioning, workaround for Set node not allowing bypassed inputs.

Does nothing, used to trigger generic workflow output. A way to get previews in the UI without saving anything to disk.

Controls flow of execution based on a boolean switch.

Simply passes through the model, workaround for Set node not allowing bypassed inputs.

Scales a batch schedule from Fizz' nodes BatchPromptSchedule to a different frame count.

Delays the execution for the input amount of time.

Experimental node for patching torch.nn.Linear with CublasLinear.

Node for patching torch.nn.Linear with CublasLinear.

Returns the path to the model as a string.

Creates a sigmas tensor from a string of comma separated values. Examples: Nvidia's optimized AYS 10 step schedule for SD 1.5: 14.615, 6.475, 3.861, 2.697, 1.886, 1.396, 0.963, 0.652, 0.399, 0.152, 0.029 SDXL: 14.615, 6.315, 3.771, 2.181, 1.342, 0.862, 0.555, 0.380, 0.234, 0.113, 0.029 SVD: 700.00, 54.5, 15.886, 7.977, 4.248, 1.789, 0.981, 0.403, 0.173, 0.034, 0.002

Creates a sigmas tensor from list of float values.

Generates noise for injection or to be used as empty latents on samplers with add_noise off.

Creates a float list from sigmas tensors.

Creates a new with the given text, and concatenates it to either above or below the input image. Note that this changes the input image's height! Fonts are loaded from this folder: ComfyUI/custom_nodes/ComfyUI-KJNodes/fonts

Creates a text image and mask. Looks for fonts from this folder: ComfyUI/custom_nodes/ComfyUI-KJNodes/fonts If start_rotation and/or end_rotation are different values, creates animation between them.

Creates single string, or a list of strings, from multiple input strings. You can set how many inputs the node has, with the **inputcount** and clicking update.

Converts any type to a string.

# SuperPrompt A T5 model fine-tuned on the SuperPrompt dataset for upsampling text prompts to more detailed descriptions. Meant to be used as a pre-generation step for text-to-image models that benefit from more detailed prompts. https://huggingface.co/roborovski/superprompt-v1

Selects a node and it's specified widget and outputs the value as a string. If no node id or title is provided it will use the 'any_input' link and use that node. To see node id's, enable "Node ID Badge Mode" in main settings. Alternatively you can search with the node title. Node titles ONLY exist if they are manually edited! 'widget_name' can be a comma separated list. The 'any_input' is required for making sure the node you want the value from exists in the workflow.

Advanced torch.compile patching for diffusion models.

Deprecated, use TorchCompileModelAdvanced instead.

Deprecated, use TorchCompileModelAdvanced instead.

Video latent VAE decoding to fix artifacts on loop seams.

Extends the potential frame count of HunyuanVideo using this method: https://github.com/thu-ml/RIFLEx

Select individual block alpha values, value of 0 removes the block altogether

https://github.com/NUS-HPC-AI-Lab/Enhance-A-Video

https://github.com/ChenDarYen/Normalized-Attention-Guidance

Gets the mean values from mask or image batch and returns that as the selected output type.

# WORK IN PROGRESS Do not count on this as part of your workflow yet, probably contains lots of bugs and stability is not guaranteed!! ## Graphical editor to create values for various ## schedules and/or mask batches. **Shift + click** to add control point at end. **Ctrl + click** to add control point (subdivide) between two points. **Right click on a point** to delete it. Note that you can't delete from start/end. Right click on canvas for context menu: NEW!: - Add new spline - Creates a new spline on same canvas, currently these paths are only outputed as coordinates. - Add single point - Creates a single point that only returns it's current position coords - Delete spline - Deletes the currently selected spline, you can select a spline by clicking on it's path, or cycle through them with the 'Next spline' -option. These are purely visual options, doesn't affect the output: - Toggle handles visibility - Display sample points: display the points to be returned. **points_to_sample** value sets the number of samples returned from the **drawn spline itself**, this is independent from the actual control points, so the interpolation type matters. sampling_method: - time: samples along the time axis, used for schedules - path: samples along the path itself, useful for coordinates - controlpoints: samples only the control points themselves output types: - mask batch example compatible nodes: anything that takes masks - list of floats example compatible nodes: IPAdapter weights - pandas series example compatible nodes: anything that takes Fizz' nodes Batch Value Schedule - torch tensor example compatible nodes: unknown

Converts different value lists/series to another type.

Extends, and converts if needed, different value lists/series