This workflow uses Stable Diffusion 1.5 but can be adapted to use any other version, it also uses LLMs (vision) to automatically generate prompts. The latest version of the IPAdapter is used and also the pose controlnet, feel free to use it or ask me if you need help.

Checkpoint that I recommend: https://civitai.com/models/132632/epicphotogasm

Ollama Describer - https://github.com/alisson-anjos/ComfyUI-Ollama-Describer (To install ollama you need to enter this repository and follow the steps, it's very simple)

IPAdapter Plus -  https://github.com/cubiq/ComfyUI_IPAdapter_plus

Layer Style -  https://github.com/chflame163/ComfyUI_LayerStyle

KJNodes - https://github.com/kijai/ComfyUI-KJNodes (Get, Set)

   [Required] *Download all model files from https://huggingface.co/mattmdjaga/segformer_b2_clothes (or git clone https://huggingface.co/mattmdjaga/segformer_b2_clothes) to ComfyUI/models/segformer_b2_clothes folder.

LoRa Weight Slider -  https://civitai.com/models/180008/weight-slider

Note: If you think that using LLM is overloading your GPU, you can use your manual references, there are some controls where you can choose to use manual references and manual prompt, now an important detail, your prompt needs to force the generation of an image of a single full-body person and that the clothing generated for that person has a similar format to the reference clothing (Top, Bottom), because as an attention mask is used, if it is not this way it can generate deformed images since the format is which will define the attention mask used in the IPAdapter.

If you need anything, you can message me on discord: alissonerdx