Node Based Image Pipelines Explained
Build, read and debug node graph workflows for reproducible image generation
A taste of a lesson
I added a LoRA loader and now my images look burnt and oversaturated. The graph runs fine otherwise.
Burnt output after adding an adapter usually means one of three things. First, strength too high: try model and text encoder strengths around 0.6 to 0.8 instead of 1.0 or more. Second, family mismatch: an adapter trained for a different base model often produces exactly this. Check its model card. Third, it stacks with a high guidance value; lower CFG a little. Test one at a time with a fixed seed and a preview after the sampler. Which base model was the LoRA trained on?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Build a minimal text to image graph and explain each node
- Extend graphs with image inputs, control, adapters and inpainting
- Add a two stage upscaling pass
- Debug graphs methodically and share them reproducibly
Lesson plan
- 1 The minimal graph Build and understand a basic text to image graph node by node. Start
- 2 Images in: image to image and inpainting Feed existing images and masks into the graph. Start
- 3 Control and adapters Add structure control and LoRA adapters correctly. Start
- 4 Upscaling stages Add a second pass that increases resolution and detail. Start
- 5 Debugging graphs Find and fix errors quickly in complex workflows. Start
- 6 Reproducible, shareable workflows Save, organise and share graphs safely. Start
Try asking
About this tutor
An advanced tutor for people moving from one box generators to node based tools, where you wire together model loading, text encoding, sampling, decoding and post processing as a visible graph. You will learn what each standard node does, how data flows through the graph, how to build a basic text to image graph and extend it with image inputs, control, adapters, inpainting and upscaling stages, and how to debug and share workflows reproducibly. The concepts apply across node based tools; ComfyUI is a well known open example, but we focus on ideas that transfer.
Reviews
4.5
2 ratingsSample
- Mikael H.Sample
Typed sockets explanation finally made the graph click. I rebuilt my messy workflow from scratch and understand every node now.
- Rui C.Sample
Strong debugging lesson. The metadata leak warning was something I had never considered. Some parts went fast for me.
About the teacher
The technical side of image AI: local models, node pipelines, adapters and settings
9 tutors 331 lessons taught Sample
I teach the engineering side of image generation to people who want control rather than a single text box. I started as a hobbyist running open models on my own machine and later built image pipelines for small studios, so I know where the frustrations are: memory errors, inconsistent batches, settings nobody explains. I teach from first principles, then from...
See Quentin's profile and tutorsMore like this
Other tutors on the same or nearby topics.