Graphic Design · GPC-44

ComfyUI and Open-Source Image Generation

ComfyUI and Open-Source Image Generation is a hands-on course in creating images with FLUX and Stable Diffusion models through ComfyUI, from building workflows node by node and controlling results with ControlNet and LoRA to batch generation through the API. It suits designers, marketing teams and developers, and you leave with a working workflow for your own projects.

Updated 1 Oct 2026

From 6,750 THB / person 7,500 −10% excl. VAT 7% · group rates available

Corporate training quote Join the waiting list

PDFDownload the course outline

Duration12 hours · 2 days
FormatOnsite / live online
Next roundOn request
CertificateIncluded

Course overview

Ready-made AI image tools are easy to use, but when results have to match real work, such as product images that must keep their exact shape, a campaign set that needs one consistent style, or large volumes of images produced automatically, teams soon hit platform limits, per-image costs and concerns about sending data outside. Open-weight models such as FLUX and Stable Diffusion, run through ComfyUI, offer the flexibility and control over every step of image generation that these jobs need.

This course takes learners from the basics of ComfyUI to building workflows for real work. It starts with choosing models and hardware and installing ComfyUI on a machine with a graphics card or on Google Colab, then covers the key nodes, FLUX.2 and SDXL, image-to-image, inpainting and outpainting, structural control with ControlNet and style control with LoRA. Learners go on to custom nodes, subgraphs, preparing data to train a LoRA, upscaling, batch generation and calling ComfyUI through its API with Python, while learning about model licences, copyright and images of real people. The course closes with each learner building a workflow for their own work. (2 days, 6 hours per day, 12 hours in total, Intermediate level.)

What you’ll gain

Choose models and hardware for the job and install ComfyUI locally or on Google Colab
Understand the core ComfyUI nodes and build a text-to-image workflow yourself
Use FLUX.2 and SDXL and tune the sampler, steps and seed to get the result you want
Use image-to-image, inpainting and outpainting to edit specific parts of an image
Control composition with ControlNet and style with LoRA and reference images
Install custom nodes safely and organise workflows for reuse
Upscale images, generate in batches and call ComfyUI through its API with Python
Check model licences and commercial-use terms before using the images

Who this course is for

Graphic designers and creatives who want finer control over AI images than ready-made tools allow
Marketing and e-commerce teams that need large numbers of product or campaign images
Developers who want to connect image generation to their systems or apps through an API
People who already use Midjourney or web-based AI image tools and want to run models on their own machine

Prerequisites

Some experience with any AI image tool and basic prompt writing
Confident computer use, including handling large files and folders, and basic Python
A machine with an NVIDIA graphics card of 8 GB or more, or an Apple Silicon Mac with 16 GB of memory or more; otherwise Google Colab or a cloud GPU paid for by the learner
At least 50 GB of free local or cloud storage for model files

Curriculum

Course Details

The course runs for 2 days, 6 hours per day (12 hours in total, 09:00-16:00), with hands-on workflow building in ComfyUI throughout. Intermediate level. Learners should have used an AI image tool before and be able to write basic Python. ComfyUI and the models are free to download, but each model has its own licence. Learners use their own machine with a graphics card, or Google Colab or a cloud GPU at their own cost. A paid Colab plan is recommended because the free tier has GPU and usage restrictions. Learners take home the course handbook, workflow files for every lab, Python scripts for the API and a summary table of models and their licences.

Day 1 ComfyUI Fundamentals and Image Control

Section 1: Lab: Choosing Models and Hardware and Installing ComfyUI

How diffusion models work, and how the FLUX and Stable Diffusion families differ
How much graphics memory each model needs, and quantised options for lower-spec machines
Running locally, on Google Colab or on a cloud GPU: speed and cost compared
Lab: install ComfyUI locally or launch it on Colab and organise the model folders

Section 2: Lab: Nodes and Your First Workflow

The ComfyUI interface: nodes, links, the queue and built-in workflow templates
Core nodes: Load Model, Text Encode, KSampler, VAE Decode and Save Image
Settings that shape the image: sampler, scheduler, steps, CFG and seed
Save workflows as JSON and reopen them from the data embedded in image files
Lab: build a text-to-image workflow node by node

Section 3: Lab: FLUX.2 and SDXL

The files FLUX.2 needs: the model, the text encoder and the VAE
Choosing a model variant and precision that fit your graphics card
Natural-language prompts for FLUX compared with keyword prompts for SDXL
Using reference images with models that support editing
Lab: compare the speed and quality of both models from the same prompt

Section 4: Lab: Image-to-Image, Inpainting and Outpainting

Image-to-image and the denoise value that decides how much the image changes
Draw masks in the Mask Editor and edit only part of an image
Outpaint to extend an image and change its aspect ratio for different media
Lab: replace the background of a product photo while keeping the product intact

Section 5: Lab: Controlling Composition with ControlNet

What Canny, Depth and Pose ControlNets are used for
Preprocessors that turn a source image into lines, depth or poses
Tune strength and the active range to balance accuracy and natural results
Combine several ControlNets in one workflow
Lab: create several styles from the same composition

Day 2 Advanced Workflows and Putting Them to Work

Section 6: Lab: Style Control with LoRA and Reference Images

What a LoRA is, how to load it, set its weight and stack several
Finding LoRAs on model hubs and reading their licences and terms first
Using reference images to keep style, colour and composition consistent across a set
Lab: a campaign set of four images in one style

Section 7: Lab: Custom Nodes, the Manager and Subgraphs

Install and update custom nodes through ComfyUI Manager
The risk of custom nodes running code on your machine, and how to choose trustworthy sources
Make workflows readable with groups and subgraphs
Lab: turn your workflow into a reusable set of nodes

Section 8: Preparing Data and Training Your Own LoRA

Choosing training images: quantity, variety and quality
Writing a caption for each image and defining a trigger word for a style or product
Images of real people require the consent of the person before training
Training on a cloud GPU and the settings to understand before you start
Lab: prepare a product dataset and test a LoRA that has already been trained

Section 9: Lab: Upscaling, Batches and the ComfyUI API

Upscale with upscaling models, and tiled upscaling for large images
Generate image sets from a list of prompts or values that change per image
Export a workflow in API format and send jobs to ComfyUI with Python
Lab: a script that creates product image variations from a CSV file

Section 10: Workshop: Licences, Copyright and a Real-World Workflow

Model licences differ: some allow commercial use and some do not
The copyright status of AI images is still uncertain, so never assume an image is copyright-free
Do not imitate real people or other brands without permission
Workshop: design a workflow for your own work, from source image to finished image
Present the results, share workflows and agree how the team will use them

Schedule & training options

For individuals — public rounds

No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770

Join the waiting list Ask about rounds

For organisations — in-house / private

Tailor the content to your team’s tools and projects
Your dates, at your office or live online
Quotation with tax ID for procurement

Corporate training quote

Instructors

Samit KoyomInstructor

Frequently asked questions

What is ComfyUI, and how does it differ from web-based AI image tools?

ComfyUI is open-source software for running image models such as FLUX and Stable Diffusion, where each step, such as loading a model, entering a prompt, choosing a sampler and saving the image, is a node you connect into a workflow. Unlike web tools such as Midjourney, which hide these steps, it lets you adjust every setting, add ControlNet or LoRA, save workflows for reuse and run them on your own machine or a cloud GPU.

What is the difference between ControlNet and LoRA, and what does each control?

ControlNet controls the structure of an image from a reference, such as edges, a human pose or depth, so the composition follows your layout. LoRA is a small add-on file that adjusts the base model to produce a particular style, character or product consistently. In general, use ControlNet to control shape and viewpoint, use LoRA to control style or identity, and combine both in one workflow when needed.

What are the steps for editing part of an image with inpainting in ComfyUI?

Load the original image into the workflow, paint a mask over only the area you want to change, then write a prompt describing what should appear there and set the denoise value: low values keep more of the original detail, high values change more. Generate several times with different seeds and pick the most natural result. To extend an image beyond its original frame, use outpainting, which works on the same principle.

Who is the ComfyUI and Open-Source Image Generation course for, and what should I know beforehand?

It suits graphic designers who want finer control over AI images, marketing and e-commerce teams producing images at volume, developers connecting image generation to systems through an API, and Midjourney users who want to run models themselves. It is an intermediate course, so you should have used an AI image tool and written prompts before, be comfortable with computers and be able to write basic Python.

What will I be able to do after the ComfyUI and Open-Source Image Generation course?

You will be able to install ComfyUI and build your own text-to-image workflows with FLUX.2 and SDXL, edit parts of images with inpainting and outpainting, control structure with ControlNet and style with LoRA, and upscale, batch-generate and call ComfyUI through its API with Python, checking model licences before commercial use. Participants receive a Certificate of Completion with its own number that can be verified online.

What computer specification do I need for the ComfyUI and Open-Source Image Generation course?

You need a computer with an NVIDIA graphics card of 8 GB or more, or an Apple Silicon Mac with at least 16 GB of memory, plus at least 50 GB of free space for model files. Without one, you can use Google Colab or a cloud GPU at your own expense; paid Colab is recommended because the free tier has GPU limits. ComfyUI and the models are free to download, but each model has its own licence.