Stable Diffusion
& ComfyUI Pro —
Full control
The most comprehensive course for Stable Diffusion and ComfyUI. From local installation to LoRA Training, advanced ControlNet and Flux. You'll leave with 8 impressive visual projects and a unique personal style.
Full free access · no registration · no credit card
- check_circle Immediate access to all 5 modules
- check_circle 8 sample Workflows to download
- check_circle Lifetime updates
- check_circle A private Discord for creators
What you learn in the course
Six areas of knowledge that will turn you from a beginner into a studio-level image creator.
Full local installation
Installing Stable Diffusion and ComfyUI on Windows, Mac and Linux — including NVIDIA GPU and Apple MPS setup.
SDXL and Flux — full control
The differences between SDXL Turbo, SDXL Base, Flux Schnell and Flux Dev. When to use each and optimal settings.
ControlNet — control over pose and composition
Pose, Depth, Canny, Reference and IP-Adapter — total control over every visual aspect of the image.
LoRA Training — your style
Training a personal style, a consistent character and specific faces. Kohya SS, Google Colab, Hyperparameter tuning.
Advanced ComfyUI and API
Custom Nodes, Workflow Sharing, ComfyUI API integrated with Python and Batch Processing of hundreds of images.
img2img, Inpainting and Outpainting
Editing existing images at a pro level — changing specific details, expanding the frame, combining new elements.
The full curriculum
Each module ends with a practical project you can put straight into your portfolio.
01
Module 1 · 2 hours
The work environment
- play_circle Installation on Windows with an NVIDIA GPU
- play_circle Installation on Mac — MPS and CPU
- play_circle Google Colab Setup — for those without a GPU
- play_circle ComfyUI — basic structure, Nodes and Connections
- rocket_launch Project: a full environment setup and a first image
02
Module 2 · 3 hours
SDXL and Flux Mastery
- play_circle SDXL Turbo vs SDXL Base vs Flux Schnell vs Flux Dev — what when
- play_circle Prompt Anatomy: Subject, Style, Lighting, Quality Tags
- play_circle Negative Prompts — what to always include and why
- play_circle CFG Scale, Steps, Sampler — what each parameter does
- play_circle Seed Management and Reproducibility
- rocket_launch Project: 20 images in a consistent style
03
Module 3 · 3 hours
ControlNet — full control over the image
- play_circle OpenPose — precise pose control
- play_circle Depth Map — three-dimensional composition
- play_circle Canny Edge — precise composition stabilization
- play_circle Reference and IP-Adapter — keeping a character consistent
- rocket_launch Project: automatic Product Photography
04
Module 4 · 4 hours
LoRA Training — train your own style
- play_circle Dataset Preparation — how many images, what resolution
- play_circle Kohya SS Training on Google Colab
- play_circle Hyperparameters: LR, Epochs, Network Dim
- play_circle Character LoRA, Style LoRA, Face LoRA
- play_circle Testing and Fine-tuning the LoRA
- rocket_launch Project: a LoRA for specific faces
05
Module 5 · 3 hours
Advanced ComfyUI and API
- play_circle Custom Nodes: ComfyUI-Manager, Impact Pack
- play_circle Complex Workflows: an Upscaling Pipeline, Video Frames
- play_circle ComfyUI API — sending requests from Python
- play_circle Batch Processing — automatically processing hundreds of images
- rocket_launch Project: an Automated Image Pipeline
Sample lesson
Here is a real example of a lesson from the course — so you know exactly what awaits you.
ControlNet OpenPose — precise pose control
What is OpenPose and why is it a game-changer?
OpenPose is a model that detects the position of the body joints in an image — head, shoulders, elbows, hands, knees, ankles — and represents them as a colored skeleton. ControlNet learns to use this skeleton as a condition for diffusion, so that your AI model produces an image with exactly the same pose, regardless of the prompt.
Before OpenPose, every attempt to get a specific pose relied on a textual description — and often led to unexpected results. With OpenPose, it is enough to take a reference image, extract a skeleton from it, and feed it into ComfyUI. The result? The same pose across all the characters you produce.
Setting up ControlNet OpenPose in ComfyUI — step by step
-
1
Loading the ControlNet Model
Download the file control_v11p_sd15_openpose.pth from the ControlNet repository on HuggingFace. Save it in the ComfyUI/models/controlnet/folder. In ComfyUI add a Node of type Load ControlNet Model and select the file.
-
2
Preparing the Pose Image
Add a Node of type DWPose Estimator (from ComfyUI-ControlNet-Aux). Connect a reference image to it via Load Image. The Estimator will generate a visual skeleton that serves as input to the diffusion.
-
3
Configuring Apply ControlNet
Add a Node Apply ControlNet. Connect: conditioning from the CLIP Text Encode, control_net from the model you loaded, andimage from the DWPose. Set strength = 0.8.
-
4
Connecting to the KSampler and saving
Connect the output of Apply ControlNet topositive conditioning in the KSampler. Run with steps=25, cfg=7.0, sampler=dpm_2_a. Connect VAE Decode and Save Image to finish.
The Workflow in graphical form
a simplified workflow — the full version with a downloadable .json file is inside the course
Weight values — what suits what
Common mistakes and how to avoid them
Using ControlNet SDXL on an SD 1.5 model (and vice versa)
Every SD version requires a matching ControlNet. An SDXL OpenPose model will not work on SD 1.5. Always check that the names match — SDXL models contain "xl" in their name.
A reference image at a different resolution from the diffusion output
If the Reference image is 512×512 and the KSampler output is 1024×1024, the skeleton will scale differently than you expected. Always Resize the reference image to the same resolution.
Too high a Weight with an uncommon pose
Very unusual poses (like an athlete mid-air) with a weight of 1.0 may produce incorrect anatomy. Lower it to 0.7 and check the result.
Free access — no payment
The course is open to everyone — no registration, no credit card. Just click and start.
workspace_premium Start the courseHave questions?
Do you need a powerful GPU to learn?
An NVIDIA card with 8GB+ VRAM is recommended for the most comfortable experience — an RTX 3060 and up does an excellent job. But we take care of everyone: a full Google Colab guide is also provided that runs everything in the cloud for free, so you can learn even from a completely old computer with no GPU.
What is the difference between Midjourney and Stable Diffusion? Why learn SD?
Midjourney is easier to use and produces impressive results quickly — great for browsing and inspiration. Stable Diffusion gives you full control: you choose every parameter, train models on your style, and run everything locally with no API cost. The course shows exactly when to choose each tool and why.
Will we also learn Flux?
Absolutely. Module 2 is entirely dedicated to SDXL and Flux — including a detailed comparison between Flux Schnell (fast, for prototypes) and Flux Dev (higher quality, for final projects). You will understand when each is preferable and how to configure each in ComfyUI.
How much disk space do you need?
A minimum of 20GB for a basic work environment with one model. 50GB+ is recommended if you want to keep several models, ControlNet models and LoRAs. In Colab there is no per-session disk limit — everything is in the cloud.
Ready to start?
8 projects and 15+ hours of
knowledge that changes things
workspace_premium
Start the course — free
Full free access · no registration · no credit card