ComfyUI Tutorial for Beginners: Text-to-Image & Image-to-Image Workflows (2026)

In Module 1 we learned how to move around Comfy. Now we put it to work. This tutorial builds a text-to-image workflow from a blank canvas checkpoint, CLIP text encode, KSampler, empty latent, VAE decode connecting every node so you understand what each one is actually doing and why the data flows the way it does. Then we make one small change and turn the exact same graph into an image-to-image workflow. We walk through every KSampler setting: seed, steps, CFG, sampler, and scheduler. Then we spend real time on denoise, the single value that decides how much of your source image survives. We dial it from 0.9 down to 0.0 so you can see exactly what each step changes. You'll also learn how to stop chasing one perfect image and start generating takes — locking seeds, preview vs. save, and why every image in your asset library carries the workflow that made it, so you can reopen any result as a live graph. The last section is troubleshooting. Errors aren't a punishment, they're a hint. We cover the six failures that break Comfy most often and how to read a red node like a sentence. Run the workflow: https://links.comfy.org/4gDmB3b