Modular - Every big module of this workflow is split into groups with a toggle to enable/disable.

Easy Lora management - Done using Lora Manager, all the trigger words are automatically added to the positive prompt, with a node for trigger word managing.

Drag and Drop to CivitAI - The images hold the metadata and are easy to upload to CivitAI.

I2I Image Reference Loader - A Image Loader group is present for easier use of reference images for copying poses, composition.

WD14 + Florence for easier prompting - A group of nodes dedicated for I2P.

Explanation:

Got tired of the mess, that my old workflow was and decided to tidy it up/improve some things after some testings.

This workflow will produce high-detail images because of the combination of the RTX upscale and Mixture of Diffusions. The later can add a lot of details to the image, but when working with high pixel count and long prompts, it will start to add unnecessary artifacts to the image, especially on places with low-detail count, such as a simple white background.

So to solve this, i've come up with the idea, to use the Mixture of Diffusions with the Anima LLLite-ControlNet, specifically Canny. Anima LLLite is set to high-power but low percentage, to give the model the understanding, what needs to be detailed and what not.

Additionally to this combo, a Turbo Lora is injected into the inference, before the MoD will begin working. This speeds up the proccess, the upscale+MoD is working with a high pixel count image and Turbo Lora adds more details and better quality at low steps.

Necessary Node Packs:

• comfyui_controlnet_aux

• rgthree-comfy

• ComfyUI-KJNodes

• ComfyUI-Impact-Pack

• ComfyUI-Easy-Use

• ComfyUI-Florence2

• ComfyUI-Lora-Manager

• ComfyUI_Comfyroll_CustomNodes

• ComfyUI-WD14-Tagger

• Nvidia_RTX_Nodes_ComfyUI

• JPS Custom Nodes for ComfyUI

• ComfyUI Anima Remap

• ComfyUI-Anima-LLLite

• ComfyUI_toyxyz_test_nodes