ComfyUI Flux Dev FP8 AIO Workflow
Workflow Name
flux-dev-fp8-aio
Workflow Description
This workflow is a basic workflow for use with ONLY the Flux Dev FP8 AIO checkpoint.
Default Workflow Dependencies
Download Flux FP8 AIO checkpoint, save in ComfyUI\models\checkpoints:
https://huggingface.co/Comfy-Org/flux1-dev/blob/main/flux1-dev-fp8.safetensors
Workflow Details
CheckpointLoaderSimple
Purpose: This node loads a Stable Diffusion checkpoint file, which contains the model, CLIP, and VAE components necessary for image generation.
Customizable Settings:
ckpt_name: This setting allows the user to select the specific checkpoint file to load. Changing this will alter the model used for image generation, resulting in different styles, subjects, and overall image characteristics.
CLIPTextEncode (Positive Prompt)
Purpose: This node encodes a positive text prompt into conditioning information, which guides the image generation process towards the desired content.
Customizable Settings:
text: This setting allows the user to input the positive text prompt. Changing this prompt modifies the content and style of the generated image.
CLIPTextEncode (Negative Prompt)
Purpose: This node encodes a negative text prompt into conditioning information, which guides the image generation process away from undesired content.
Customizable Settings:
text: This setting allows the user to input the negative text prompt. Changing this prompt modifies the content and style of the generated image by excluding elements.
FluxGuidance
Purpose: This node applies flux guidance to the positive conditioning, enhancing the prompt’s influence on the generated image.
Customizable Settings:
guidance: This setting adjusts the strength of the flux guidance. Increasing this value amplifies the prompt’s impact, potentially leading to more accurate or exaggerated results, while decreasing it reduces the prompt’s influence.
EmptySD3LatentImage
Purpose: This node generates an empty latent space representation, which serves as the initial noise for the diffusion process.
Customizable Settings:
width: This setting defines the width of the latent image, which directly influences the final image’s width. Increasing or decreasing this value changes the horizontal resolution of the generated image.
height: This setting defines the height of the latent image, which directly influences the final image’s height. Increasing or decreasing this value changes the vertical resolution of the generated image.
batch_size: This setting determines the number of latent images generated in a single batch. Increasing this value produces multiple images at once, while decreasing it reduces the number of generated images.
KSampler
Purpose: This node performs the core diffusion process, iteratively refining the latent image based on the model, conditioning, and sampler settings.
Customizable Settings:
seed: This setting initializes the random number generator, affecting the noise pattern and ultimately the generated image. Changing this seed produces different image variations. Setting it to “randomize” will generate a different image every time.
sampler_name: This setting selects the sampling algorithm used for the diffusion process. Different samplers have varying speed, quality, and style characteristics.
steps: This setting controls the number of diffusion steps performed. Increasing this value generally improves image quality but increases processing time.
cfg: This setting adjusts the classifier-free guidance scale, controlling the strength of the prompt’s influence. Higher values increase adherence to the prompt but may introduce artifacts.
scheduler: This setting chooses the noise schedule used during sampling, impacting the diffusion process’s speed and quality.
denoise: This setting controls how much of the initial noise is removed during the sampling process. A value of 1.0 means full denoising, while lower values result in a noisier image.
VAEDecode
Purpose: This node decodes the latent image generated by the KSampler into a pixel-based image.
Customizable Settings: There are no customizable settings on this node beyond the VAE model that is input.
SaveImage
Purpose: This node saves the generated image to a file.
Customizable Settings:
filename_prefix: This setting allows the user to specify a prefix for the saved image file name. Changing this prefix helps organize and identify generated images.
Credits
This workflow has been modified from the ComfyUI workflow:
https://comfyanonymous.github.io/ComfyUI_examples/flux/
