Here's a workflow using Qwen to generate base image and SDXL to refine it. It's the result of my long hours of practice on ComfyUI, applying tiled diffusion upscaling and detailers for various body parts. I tried to keep it as simple as possible, but it's a complex workflow that isn't necessarily easy for a new user to grasp. Qwen is very ressource intensive so you can use quantized versions <a target="_blank" rel="ugc" href="https://huggingface.co/city96/Qwen-Image-gguf/tree/main">here</a> which will be lighter for your graphics card model. Here's a list of the different models needed to use it:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/1843568/qwenimagefp8e4m3fn">Qwen model</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/f5aiteam/CLIP/resolve/main/qwen_2.5_vl_7b_fp8_scaled.safetensors?download=true">Clip</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/1912333/qwen-image-vae">Qwen image VAE</a></li><li>Qwen Lightning Lora <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Qwen-Image-Lightning/tree/main">here</a></li><li>Qwen's Lora <a target="_blank" rel="ugc" href="https://civarchive.com/models/1662740?modelVersionId=2106185">here</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/1551668?modelVersionId=2270374">here</a> (greatly helps to achieve photorealistic renderings)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/1837476/graamxljuicy-merge-photorealistic-sdxl-model-for-realistic-sfw-and-nsfw-hd-photos">My SDXL merge model </a>(you can use another one but this one is mine so it's better^^)</li><li><a target="_blank" rel="ugc" href="https://drive.google.com/file/d/1lpT-1PqJAe8aG4fn9oo1JiBMQTZt4v3u/view?usp=sharing">Detection models </a>(Folder also present in the archive provided with the workflow, put them in /models/ultralytics/bbox)</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy/DMD2/blob/main/dmd2_sdxl_4step_lora_fp16.safetensors">DMD2 Lora for SDXL</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/xinsir/controlnet-tile-sdxl-1.0/blob/main/diffusion_pytorch_model.safetensors">Xinsir Tiled Controlnet </a>(rename it as "Xinsir_controlnet-tile-sdxl-1.0.safetensors")</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/232746/real-humans">Lora for Face detailer</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/494912">Lora for Hands detailer</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/328932/detailed-nipples-xl">Lora for Nipple detailer </a>(this one is mine too ^^)</li></ul>All nodes used should be able to be downloaded via the Manager but I have not tried with a fresh install, I will provide details if necessary.You can change the aspect ratio for images of various formats, and a landscape true/false button is used to calculate tile sizes for upscaling. I recommend a base image resolution between 1.2 and 1.4 megapixels, depending on your preferences, the output image will be 2x larger. You can also mute/bypass the Grain filter at the end of the chain if you wish.Leave a review or tip me if you wish, it will be greatly appreciated, I give you here all my experience and my know-how on ComfyUi in order to obtain the best possible images with Qwen and SDXL. I hope you have fun with it :)

Files with SHA256: 815e9862c57257f8cd31c6596702a57b74b7d1705b6685511e94f039e9f2eeba