AITK-FLUX-modular multipass workflow, Txt2img, img2img, inpainting, lora, controlnet detailers, upscale, faceswap, and more

AITK Workflow

• Workflow Version 9:

In Version 9, I’ve introduced two new features and a fun little addition for good measure. First, I’ve added a new method for the first pass called split sigmas noise injection. This technique renders your image with three samplers, injecting noise between the last two, giving you more control over texture and detail during the initial pass.

I’ve also added an LLM prompt inspiration feature. You can run this with the Grok API for free, or use the one included in the workflow—it’s all set up, and I don’t mind if you run it as-is!

As a bonus, I included the Comfy Pet extension—it’s just a playful meme feature where you can mess around with a virtual doge while waiting for your image to generate. It’s not essential for running the workflow, but I had fun with it, so I figured, why not share the fun?

• Workflow Version 8:

In Version 8, I’ve introduced some significant upgrades to make the workflow even more efficient and user-friendly. One of the major additions is an image compare pass, allowing you to directly compare the first pass with the final image within Comfy, making it easier to see the progression and changes at a glance.

I also added the upscale latent tile pass, which was originally created by Latent Vision, but I’ve adjusted it to fit into this workflow seamlessly. Additionally, there’s a new upscale latent "classic" pass, which provides faster upscaling while maintaining added detail without significantly altering the composition. This is a great option when you want to preserve the core elements of your image but still enhance the finer points.

One of the more exciting additions is the Vton (Virtual Try-On) feature, which comes with automatic segmentation, though you can still mask manually if you prefer. This pass allows for easy virtual try-ons in your images, giving you even more creative control.

I’ve also improved the group bypasser, making it much easier to avoid accidental activation of passes like the first pass and inpainting simultaneously—something that used to trip me up often. Now, the workflow is smoother, and mistakes like that are much less likely.

Lastly, I fixed some issues with how images were distributed in the pipes, ensuring a more reliable and consistent operation throughout the process.

• Workflow Version 7:

In Version 7, I’ve made two important additions. First, I added a simple image sender to easily save your images within the workflow. But the big feature in this update is the new 8X upscale pass, which will give you crisp details and high resolution—perfect for when you need those ultra-sharp final outputs.

• Workflow Version 6:

This latest update brings several powerful new features. First, I’ve added taggers for image captioning and integrated 3 different models for more versatility. One of the key additions is a new pass called "detailer no upscale"—this pass allows you to add more detail to your image without upscaling, which is ideal if you want to keep things sharp in a smaller format. I highly recommend using AITKcine flux for this pass—it will, as a friend said i'm citing , "make your shit dramatically real."

I’ve also revamped the user interface a bit. The control colors have been updated to be more intuitive, making it easier to navigate and adjust your settings. Lastly, I’ve added two block notes to recap the workflow and guide you through how to use it effectively.

• Version 5:
  The latest version adds more control and flexibility to the workflow. I introduced a Flux enhance prompt toggle, allowing you to turn it on or off as needed. Additionally, a CLIP attention multiplier was added to better control how emphasis is placed in the images. I also replaced the Lora stack with Power Lora, improving the performance of the face swaps. On the technical side, I fixed the inpainting functionality and resolved issues with the image pipe, ensuring a smoother operation.

• Version 4:
  In this most recent update, I added a Lora faceswap feature, improving the versatility of face manipulation. I also solved minor bugs related to the pipes and expressions, ensuring a smoother and more efficient workflow.

• Version 3:
  The latest version refines expression control with LivePortrait nodes, offering better handling of facial expressions and the overall flow of image edits. Additionally, I’ve enhanced input image management, allowing you to resize images according to your specific needs, making the workflow even more adaptable.

• Version 2:
  In this update, I integrated ControlNet for Flux, giving you greater flexibility and control over how you work with your inputs, making the workflow more dynamic and powerful for various creative tasks.

• Version 1:
  After the Comfy update, many of my previous workflows became outdated, so I transitioned to Flux. The initial setup was designed to mimic a simple UI, with all controls in front of you, allowing you to choose which groups to bypass or activate. This first version included minimalistic edits for face, eyes, skin, clothes, hand detailing, an inpainting tab, and an upscaler.

This workflow was built to meet my own creative needs, but I’m always open to your suggestions. If you have any ideas for further improvements, feel free to comment!