About

Hello, I am looking to get an expert to improve our ComfyUI image generation pipeline. I run a business that uses ComfyUI to create images of a child in fantasy settings based on an image of the child provided by the customer. Notes about our current workflow: Pipeline is roughly: base image generation - upscale - face swap We use stable diffusion currently (SDXL) The image style is an animated, cartoon/disney aesthetic This is hosted on RunPod We have two main goals with this workflow improvement process: 1. Most importantly, we want to automate the image production pipeline. As it stands, the operator needs to interface with ComfyUI to initiate the image generation, then sort out the usable pictures (we generate roughly 4x what we need because it does not always give good images), run them through the upscaler workflow, then run them through the face swap workflow. This is quite labor intensive and we want to be able to have a single workflow that automatically runs each image through the final step to output a folder of images. Eventually, this workflow will be setup to be called via API, so it should not require user interaction. The main thing stopping us from just merging the workflows is we don't want to waste GPU time upscaling/face swapping the bad images, which brings me to goal #2... 2. We would like to improve the image quality of the outputs. Notably things like prompt adherence, and avoiding issues such as extra fingers, AI text, and other negative artifacts. This side of things seems like a requirement in order to streamline the generation process.. The new workflow must generate images with a cohesive cartoon style and must have an accurate face swap. When you apply, please give me some background on your experience in this niche and a rough idea of how you plan to accomplish this task. Thanks!