78 lines
4.1 KiB
Markdown
78 lines
4.1 KiB
Markdown
# WLSH ComfyUI Nodes
|
|
A set of custom nodes for ComfyUI created for personal use to solve minor annoyances or implement various features.
|
|
|
|
## Node descriptions
|
|
### Checkpoint Loader with Name
|
|
The regular checkpoint loader with an output that provides the name of the loaded model as a string for use in saving filenames
|
|
|
|
### KSamplerAdvanced (WLSH)
|
|
Modified version of the advanced KSampler to return the `denoising` option for experimentation
|
|
|
|
### Seed to Number
|
|
Converts a `SEED` type to an `INT`, which was needed for some third party nodes
|
|
|
|
### SDXL Steps
|
|
A single node that outputs three different integers for use as step count in SDXL workflows.
|
|
|
|
### SDX Resolutions
|
|
Outputs height and width integers from a list of SDXL optimal resoultions
|
|
|
|
### Multiple Integer
|
|
Simple math node to multiple an integer by an integer.
|
|
|
|
### Time String
|
|
Returns a formatted string of the current time in the selected format. Useful for filenames or debugging
|
|
|
|
### Empty Latent by Ratio
|
|
Quickly generate an empty latent by selecting an aspect ratio, an orientation (landscape or portrait) and the length of the 'shorter' side
|
|
|
|
### Resolutions by Ratio
|
|
Similar to Empty Latent by Ratio, but returns integer `width` and `height` for use with other nodes.
|
|
|
|
### SDXL Quick Empty Latent
|
|
Quickly generate an empty latent of a size from a list of ideal SDXL image sizes
|
|
|
|
### CLIP Positive-Negative
|
|
Simple CLIP box containing positive and negative prompt boxes. Mostly a space saver
|
|
|
|
### Clip Postive-Negative w/Text
|
|
Same as the above, but with two output ndoes to provide the positive and negative inputs to other nodes
|
|
|
|
### Outpaint to Image
|
|
Extends an image in a selected direction by a number of pixels and outputs the expanded image and a mask of the outpainted region with some blurred border padding.
|
|
|
|
### Generate Edge Mask
|
|
Used to generate a 'stripe' of mask with blurred edges that overlaps an original iamge with the outpainted new space. Use on a second pass of an image to 'redo' the masked stripe region and help smooth out the edges of an outpainted image. Pass output to a `Convert Image to Mask` node using the `green` channel.
|
|
|
|
### VAE Encode for Inpaint Padding
|
|
A combined node that takes an image and mask and encodes for the sampler. Can apply some padding/blur to improve inpainting behavior.
|
|
|
|
### Image Scale by Factor
|
|
Scale an image by a predefined factor without a model
|
|
|
|
### Upscale by Factor with Model
|
|
Does what it says on the tin. Scales using an upscale model, but lets you define the multiplier factor rather than take from the model.
|
|
|
|
### SDXL Quick Image Scale
|
|
Take an input image and do a quick simple scale (or scale & crop) to one of the ideal SDXL resolutions
|
|
|
|
### Image Save with Prompt Data
|
|
Save an image with prompt information embedded as a comment or exif data similar to how Auto1111 does it. Need to feed it various pieces of information and then specify the filename syntax for replacement if desired. Filename ptions include `%time` for timestamp, `%model` for model name (via input node or text box), `%seed` for the seed (via input node). Writes the above as well as positive and negative prompts (convert their widgets to inputs to connect to a prompt string box for easy automation). As far as I can tell, does not remove the ComfyUI 'embed workflow' feature.
|
|
|
|
### Save Prompt Info
|
|
As the above, but to a txt file
|
|
|
|
### Image Save with Prompt File
|
|
Combination of the above two
|
|
|
|
### Save Positive Prompt File
|
|
Saves just the connected text info (expected to be the positive prompt) to help with training. Make sure to do matching filenames!
|
|
|
|
## Disabled or Non-Functional Nodes
|
|
|
|
### Alternating KSampler
|
|
Work in progress. Attempting to recreate the alternating prompts of Auto1111. Provide a prompt with the syntax "<foo|bar>" for it to alternate between "foo" and "bar" (and any number of others) before looping back around to the first entry. Currently not functional due to sampling issues, but the prompt alternating logic works.
|
|
|
|
### Generate Face Mask
|
|
Use a face detection library to create a mask box on a desired color channel to then be passed to the `Convert Image to Mask` node, but currently not working.
|