List of changes in v2:

• Added base model selector, from FP16 to Q2

• Individual activator for SageAttetion and Torch Compile

• Added “Image Style Loras” panel to change the style of the generated image. “Smartphone Snapshot Photo Reality” has been moved to this panel along with other style loras. The download links and recommended strength are available there.

• Added option to select the total steps, with automatic calculation for the steps of each KSampler.

• Added “Prompt variation helper” option to help get more variation in the result.

• Added option to use VAE or Tiled VAE

• The generated image is now upscaled to x2 by default.

• New settings in KSamplers to prevent image defects (body elongation, duplication, etc.).

• New image enhancement options, including Instagram filters.

• Additional upscaling options to x2 or x8 (up to 30k resolution).
  
  
  Hi!

I've customized a workflow to my liking with some interesting options and decided to share it.
Hope you like it.

Here are some details:

• Ready for GGUF models and MultiGPU

• Option to easily enable/disable basic Loras (Lightx2v, FusionX, Smartphone Photo Reality)

• Option to enable/disable additional Loras (characters, motions)

• Option to select a preset size or customize it manually

• Option to add sharpness and grain

• Option to enable Upscaling

• Option to enable accelerators (Sage Attention + Toch Compile)

• Descriptive text for each step

I used 2x3090Ti and the generation time at 1920x1080 is about 100 seconds.

* For the size presets you will need to copy the “custom_dimensions.json” file into /custom_nodes/comfyui-kjnodes/

If you encounter any problems or have any suggestions for improvement, please let me know.

Enjoy!

* This workflow works really great with character loras.

* The first images I uploaded got taken down, and I got a warning for using images of well-known people, so I re-uploaded images created with the WF without character loras.