Skip to content

Latest commit

 

History

History
1406 lines (729 loc) · 41.7 KB

README_EN.md

File metadata and controls

1406 lines (729 loc) · 41.7 KB
Raw

Blender Open Source AI Tools Infinite Grail Node

中文 EN

Manual

AIGODLIKE Community

Chapter One: Preface

Introduction

The Infinity Grail Tool is a blender AI tool developed by"只剩一瓶辣椒酱-幻之境开发小组"(a development team from China)based on the STABLE DIFFUISON ComfyUI core, which will be available to blender users in an open source & free fashion.

Download Address

Blender Infinity Grail Node (ComfyUI Edition)

Google Driver

Video Tutorial

Complete Collection Of Bilibili (Under Update)

Blender x ComfyUI Node-Style AI Painting Adventure -00 Installation and Deployment (including Full Blood Version integration Pack)

Chapter Two: Introduction to All Nodes

Editor:谷缘芷月、只剩一瓶辣椒酱、王也道长OS

Sampling

K Sampler

图片

K Sampler can add the noise to the latent space and then gradually remove the noise to regenerate the latent space.

Seed

  • Specifies the random seed to generate the noise
  • < Node random every run > The random seed of this sampler changes randomly every run after it is turned on
  • < Random All > The random seeds of all samplers change with each run after it is enabled

Steps

  • Specifies the number of times to remove noise

CFG

  • Control the effect of prompt on sampling

Sampler

  • Specifies the sampler to use

Scheduler

  • Specifies the scheduler to use,to adjust the degree of noise removal each time
    • karras：Nonlinear sampling
    • normal：Equidistant interval sampling
    • simple：Approximate equidistant interval sampling is the simplest scheduler
    • ddim_uniform：Official effect, recommended in combination with ddim
  • Normally, only karras is used

Denoise

  • Specifies the amount of noise to be added, generally applicable to graph generation graphs.

Correlation Node

K Sampler Advanced

图片

Add Noise

  • Whether the switch adds noise

Noise

  • Specifies the random seed to generate the noise
  • < Node random every run > The random seed of this sampler changes randomly every run after it is turned on
  • < Random All > The random seeds of all samplers change with each run after it is enabled

Starting At Step

  • Specifies the number of steps to start adding noise

Add Noise

  • Specifies the number of steps to end adding noise

Return Residual Noise

  • Specifies whether to fully return undenoised waves, as reflected in the return noise screen

Correlation Node

Loader

Checkpoint Simple Loader

Able to load model, U-net, VAE encoder/decoder in checkpoint(safetensor) file, outputs as model, CLIP, VAE.

Special Instructions

  • The loader automatically recognizes the size and version of the model, but it is not entirely accurate.
  • The default is not to limit CLIP depth, you can use to limit CLIP

Correlation Node

VAE Loader

图片

It is able to load VAE.

Correlation Node

LoRA Loader

图片

It is able to load The LoRA model.

Special Instructions

  • Can load LoHA, LoCON, LoKR and other models.

Correlation Node

CLIP Loader

图片

It is able to load CLIP model.

Correlation Node

ControlNetLoader

图片

It is able to load ControlNet model

Correlation Node

Different ControlNet Loader

图片

Another loader capable of loading the ControlNet model.

Correlation Node

Style Model Loader

It is able to load style models.

Correlation Node

CLIPVisual Loader

图片

It is able to load Clip Visual model

Correlation Node

Using Examples

图片

unCLIP Checkpoint Loader

图片

It is able to load the model in the Safecheckpoint file, the CLIP text layer, the VAE encoding/decoder and the CLIP visual layer, and the output is the model, CLIP, VAE and visual CLIP.

Correlation Node

Using Examples

图片

GLIGENLoader

图片

It is able to load GLIGEN model.

Correlation Node

Using Examples

图片

Hypernetwork Loader

图片

It is able to load Hypernetwork。

Correlation Node

Using Examples

图片

Condition

Style Model

Style Model Apply

图片

Apply the visual CLIP to the condition.

Special Instructions

  • When other style conditions already exist in the upstream node, the style is superimposed with the upstream style.
  • CLIP visual encoding is required.

Correlation Node

Using Examples

图片

GLIGEN Language Based Image Generation

GLIGEN Text Box Apply

图片

This node restricts the specified text in the condition to the specified location.

Width And Height

  • Specifies the area size.

XY

  • Specifies the area location, with the coordinates being the upper-left corner.

Special Instructions

  • Can be used superimposed to control the range of each prompt
  • If a condition exists upstream of the node tree, the condition is applied to the upstream node
  • When a condition is applied simultaneously on different scales, it can produce unpredictable results

Correlation Node

Using Examples

图片

CLIP Text Encode

图片

This node encodes the text content as a condition.

Text

  • Be able to read Texture Inversion | Embedding，The use method is"embedding:embdding_name"
  • When multiple embedding needs to be read,use“embedding:emb_1,embedding:emb_2”

Correlation Node

CLIP Set Last Layer

图片

This node is able to limit the depth of the CLIP

Stop At Clip Layer

  • When the value is -1, there is no limit
  • When the value is -2, stop before the last layer
  • When the value is -3, stop before the last 2 layers

Special Instructions

This node can cause the model to produce unpredictable results,some prompt may depend on this setting.

Correlation node

Using Examples

图片

Conditioning Average

图片

Average the two conditions by strength.

Condition To Strength

  • Specify the mixing strength value. The smaller the value, the stronger the conditional source effect

Special Instructions

  • If Condition Source contains multiple sets of conditions, only the first set of conditions is used

Conditioning Combine

图片

Combine two conditions.

Special Instructions

  • It is a simply combination.
  • Usually used in conjunction with < conditional region >.
  • The combined conditions are output separately, different from the effect of entering a set of conditions at the same time.

Correlation Node

Conditioning Set Area

图片

Constraints are sampled in a specified area, and weights can be set.

Width And Height

  • Specified area size.

XY

  • Specified position,coordinates for the "sampling from" the top left corner of the latent first pixel.

Strength

  • Designated Limit Strength

Special Instructions

  • The function of this node is not to restrict the condition area
  • If there is a region without any conditions, the calculation returns the noise
  • If a condition exists on the upstream node, the condition overrides the upstream condition

Correlation Node

Conditioning Set Mask

图片

Through the mask limit the condition sampling in the specified area, and the weight can be set.

Strength

  • Specified Conditional Strength.

Set Condition Area

  • Select whether to generate only in mask.

Special Instructions

  • The function of this node is not to restrict the condition area.
  •  there is a region without any conditions, the calculation returns the noise.
  • If a condition exists on the upstream node, the condition overrides the upstream condition.

Correlation Node

Using Examples

图片

CLIP Visual Encoder

图片

Visual CLIP is used to encode images.

Correlation Node

Using Examples

图片

unCLIP Conditioning

图片

The visual CLIP encoded output image is applied to the condition.

Special Instructions

  • This node applies only to the unCLIP model.

Correlation Node

Using Examples

图片

ControlNet Apply

图片

Apply ControlNet to conditions.

Correlation Node

Using Examples

图片

Latent Space

Internal Complement Drawing

VAE Encode For Inpaint

图片

The encoded image is latent, and areas with a color value greater than 0.5 in the mask become the target of internal filling.

Special Instructions

  • The mask input is stretched to the size of the input image.

Correlation Node

Using Example图片

图片

Set latent Noise Mask

图片

With noise overlay latent, the content outside the mask is replaced by noise.

Special Instructions

  • The noise that is replaced affects the generation of the rest of the content
  • If the upstream node already has a noise mask, the noise replaces the upstream noise

Correlation Node

Batch

Latent From Batch

图片

Select the node latent, the amount determined by < batch size >.

Batch Index

  • Choose the latent on the field, 0 denotes the first one.

Correlation Node

Using Examples

图片

Repeat Latent Batch

图片

Copy batches a specified number of times.

Amount

  • Specifies the number of times to copy
  • Keep the batch order when copying
  • lat_1, lat_2 → lat_1, lat_2, lat_1, lat_2

Rebatch Latents

图片

Respecify the batch size.

Batch Size

  • Respecify the batch size in each batch
  • When the total amount of the original lot is greater than the target amount, the original lot is split into a batch sequence of the specified size
  • When the total quantity of the original batch is not greater than the target quantity, output the original batch quantity

Special Instructions

  • The number of batches output by this node is calculated simultaneously, rather than individually

Using Examples

图片

Transformation

Latent Rotate

图片

Rotate latent.

Rotation

  • Rotate the specified angle clockwise.

Correlation Node

Latent Flip

图片

Flip latent。

Correlation Node

Latent Crop

图片

Crop latent.

Width And Height

  • Specifies the size of latent to be pruned out.

XY

  • Specified position,coordinates for the "sampling from" the top left corner of the latent first pixel.

Special Instructions

  • The output latent becomes the new latent instead of overwriting the external content with empty latent

Correlation Node

VAE Decode

图片

Decode the latent into an image.

Correlation Node

VAE Encode

图片

Encode the image to latent.

Correlation Node

Empty Latent Image

图片

Generate an empty latent.

Width And Height

  • Specify the latent size.

Batch Size

  • Specifies the number of latent generated simultaneously.

Special Instructions

  • The latent generated at the same time feeds directly to the downstream node,the first latent is automatically selected when is used.
  • The use of latent random species in the same cohort is incremental,for example, 123, 124, 125.

Correlation Node

Latent Upscale

图片

Using the image processing algorithm to scale latent, you can specify the size after scaling.

Upscale Method

  • Nearest - Exact: Copy the pixel closest to the original image.
  • Bilinear: Select 4 pixels connected in the original image and interpolate between 2 points.
  • Area: The same as "nearest - exact".
  • area：和临近-精确相同

Width and Height

  • Specifies the size of latent to scale to.

Crop

  • Whether to crop the original image.
  • If not opened, the latent stretches to match the target width and height.

Correlation Node

Latent Composite

图片

Composite "sample from" to "sample to".

XY

  • Specified position,coordinates for the "sampling from" the top left corner of the latent first pixel.

Special Instructions

Blend two latent based on the color value of the mask,the output is (1-mask) destination+masksource.

Correlation Node

Latent Composite Masked

图片

Similar to ,eclosion is not available, masks can be used.

XY

  • Specified position,coordinates for the "sampling from" the top left corner of the latent first pixel.

Special Instructions

  • The two latent values were mixed based on the color value of the mask and the output was （1-mask）destination+masksource.

Correlation Node

Image Scale

图片

Scale the image using an image processing algorithm,ability to specify the size after scaling.

Amplification Method

  • Nearest - Exact: Copy the pixel closest to the original image.
  • Bilinear: Select 4 pixels connected in the original image and interpolate between 2 points.
  • Area: The same as "Nearest - Exact".

Width and Height

  • Specifies the size to which the image is to be scaled

Crop

  • Whether to crop the original image
  • If not enabled, the image will be stretched to fit the target width and height

Correlation Node

Image Upscale With Model

图片

Upscale the image using a upscale model.

Correlation Node

Post-Processing

Image Blend

图片

Using the algorithm to blend two images.

Blend Mode

  • Soft Light:
    • img2 <= 0.5,
    • img1 - (1 - 2 * img2) * img1 * (1 - img1),
    • img2 > 0.5,
    • (1 - 2 * img2)* img1 * (1 - img1) + img1^2
  • Overlay:
    • img1 <= 0.5,
    • 2 * img1 * img2,
    • img1 > 0.5
    • 1 - 2 * (1 - img1) * (1 - img2)
  • Screen：1 - (1 - img1) * (1 - img2)
  • Multiply：img1 * img2
  • Normal：img2

Image Blur

图片

Use algorithms to blur images.

Blur Radius

  • The larger the value, the stronger the blur effect

Sigma Coefficient

  • The larger the value, the greater the blur range.

Image Quantize

图片

Use an algorithm to quantify the image.

Colors

  • Specifies the number of colors used to represent the image

Dither

  • Whether to reduce the ribbon by adding a small amount of noise

Image Sharpen

图片

Use algorithms to sharpen images.

Sharpen Radius

  • The greater the value, the greater the sharpening range

Alpha

  • The greater the value, the greater the sharpening intensity

Save Image

图片

Save the image on the web side.

Correlation Node

Preview Image

图片

Preview the image on the web side.

blender interface not applicable.

Correlation Node

Load Image

图片

Load the image on the web side.

blender interface not applicable.

Correlation Node

Image Invert

图片

Set all colors in the image to invert.

Image Pad For Outpaint

图片

Image extension node suitable for supplementary painting.

Image output complement is 0, the original image unchanged; Mask output is 1, the original image is 0.

Up, Down, Left, Right

  • Extend outward in the specified direction.

Mask

Load Image Mask

图片

Load the mask on the web side.

blender interface not applicable.

Correlation Node

Mask To Image

图片

Convert the mask to a image.

Image To Mask

图片

Convert the image to a mask according to the specified channel,masks are 0-1 grayscale maps that do not retain color.

Solid Mask

图片

Generates a solid mask.

Value

  • Specifies the mask color value.

Width And Height

  • Specifies the mask size.

Invert Mask

图片

Invert the mask color value.

The output is (1-value).

Crop Mask

图片

Crop the mask.

XY

  • Specified position,coordinates for the "sampling from" the top left corner of the latent first pixel.

Width And Height

  • Specifies the size of the mask to crop out.

Mask Composite

图片

Mix the two masks according to the algorithm.

XY

  • Specified position,coordinates for the "sampling from" the top left corner of the latent first pixel.

Operation

  • multiply：target * source
  • add：target + source
  • substract：target - source

Feather Mask

图片

Feather from edge to center.

Up, Down, Left ,Right

  • The amount of feathering in the specified direction.

Special Instructions

  • Feathering is feathering from the original color value to 0.

Test

VAE Decode Tiled

图片

Similar to . It takes longer, has lower quality, and requires less VRAM.

Correlation Node

VAE Encode Tiled

图片

Similar to <VAE Encode>. It takes longer, has lower quality, and requires less VRAM.

Correlation Node

Tome Patch Model

图片

Use extra tokens in the "Tome model" to speed up generation.

Ratio

  • Indicates the ratio of the combined tokens.

Load Latent

图片

Load the latent space on the web side.

blender interface not applicable.

Save Latent

图片

Save the latent space on the web side.

Advanced

Loader

CheckpointLoader

图片

The node that will load the model, U-net and VAE encoding/decoder from the Safecheckpoint file, and output them as the model, CLIP and VAE, is capable of defining the configuration file of the model.

This node is not recommended unless you understand the model you are working with.

Diffusers Loader

图片

Ability to load diffusers models.

Preprocessor

Edge Line

Canny Edge Preprocessor

图片

The Canny algorithm is used to process the image.

Low And High Threshold

  • Low Threshold : Control the weak edge detection threshold
  • High Threshold : Control the strong edge detection threshold

L2 Gradient

  • Select whether to use L2 operation.
  • L2 is more accurate than L1, but it reduces the speed of calculation.

M-LSD Preprocessor

图片

Using the M - LSD algorithm to process the image.

Score Threshold

  • Control the threshold of detection lines. Lines with confidence lower than the threshold will not be detected as edges

Distance Threshold

  • Control the threshold of line distance. When the distance between lines is lower than the threshold, it will be merged

HED Preprocessor

图片

Use HED algorithm to process images.

Version

  • Selection algorithm version

Safe

  • Choose whether to increase the stability of the line, specifically for clearer lines

Scribble Preprocessor

图片

Use Scribble algorithm to process images.

FakeScribble Preprocessor

图片

Use FakeScribble algorithm to process images.

Binary Preprocessor

图片

Use Binary algorithm to process images.

Threshold

  • Pixels with a color value above the threshold are treated as 1, and pixels with a color value below the threshold are treated as 0
  • When the threshold is 0, the effect is the same as 255

PiDiNet Preprocessor

图片

Use PiDiNet algorithm to process images.

Increase Stability

  • Choose whether to increase the stability of the line, specifically for clearer lines.

Normal Depth Map

Midas-DepthMap Preprocessor

图片

Using Midas DepthMap algorithm processing image.

A

  • The Angle of inclination of the depth projection onto the image

Background Threshold

  • Specifies the threshold for removing the background

Correlation Node

MiDaS-NormalMap Preprocessor

图片

Using Midas NormalMap algorithm processing image.

A

  • Specifies the rotation Angle of the normal graph

Background Threshold

  • Specifies the threshold for removing the background

Correlation Node

LeRes-DepthMap Preprocessor

图片

Using LeRes DepthMap algorithm processing image.

Removal Nearest

  • Specifies the threshold at which the identified depth nearest is removed, as if the portion larger than this value is replaced by 1.

Remove Background

  • Specifies the threshold for removing the recognized depth background, as if the parts less than this value are replaced by 0.

Correlation Node

Zoe-DepthMap Preprocessor

图片

Using Zoe DepthMap algorithm processing image.

Correlation Node

BAE-NormalMap Preprocessor

图片

Using BAE NormalMap algorithm processing image.

Correlation Node

Pose

Openpose Preprocessor

图片

Using Openpose algorithm processing image.

Detection

  • Select whether to detect the corresponding part

Version

  • Selection algorithm version

Correlation Node

MediaPipe-HandPose Preprocessor

图片

Using MediaPipe HandPose algorithm processing image.

Detection Posture

  • Select whether to detect torso posture.

Detecting Hand

  • Choose whether to test the hand.

Correlation Node

Semantic Segmentation

SemSeg Preprocessor

图片

Using SemSeg algorithm processing image.

Face Mesh

MediaPipe-FaceMesh Preprocessor

图片

Using MediaPipe HandPose algorithm processing image.

Maximum Faces

  • Specifies the maximum number of faces to detect.

Minimum Confidence

  • Specifies the minimum threshold for determining the face.

Color Style

Color Preprocessor

图片

Using Color algorithm processing image.

Tile Preprcessor

图片

Blender

Input

图片

Load images in blender interface.

Not applicable to the web side.

Input Mode

  • Input : Directly input the picture and process it into a recognizable form
  • Render: The input image may contain render elements or transparent channels that require secondary preprocessing

Correlation Node

Mask

图片

Load the mask in blender interface.

Not applicable to the web side.

Special Instructions

  • Objects in the scene can be selected to enter the mask directly

Correlation Node

Save

图片

Save the image in blender interface.

Not applicable to the web side.

Correlation Node

Preview

图片

Preview the image in blender interface.

Not applicable to the web side.

Special Instructions

  • Preview images do not contain image metadata
  • The preview folder is emptied at each startup

Correlation Node