ComfyUI Course

ComfyUI Course - Learn ComfyUI From Scratch | Full 5 Hour Course

All the basics in one long video: install ComfyUI, run and build your first workflows, then use image to image, LoRAs, ControlNet, and Z-Image Turbo with FP8 and GGUF models. Watch this one first.

Good to know: About 5 hours. A card with 6 to 8 GB of VRAM is enough for this episode.

ComfyUI Nunchaku Tutorial: Install, Models, and Workflows Explained

Install the Nunchaku node with the Easy Install script, choose int4 or fp4 models for your card, and run Z-Image, Qwen Image and Flux 1 Dev faster with less VRAM.

Suggestion: you can skip this one. Nunchaku has some problems now, and the INT8 models do the same job: faster images with less VRAM. INT8 works in ComfyUI itself, with no extra node to install. See Episode 25.

AI Image Editing in ComfyUI: Flux 2 Klein

Set up Flux 2 Klein (4B and 9B), build an image workflow from scratch, then change it into an edit workflow. Also covers inpainting with a mask and editing with two images.

Good to know: Both models ran on 6 GB of VRAM. Episode 9 shows the faster 9B KV version.

FLUX.2 Klein 9B KV: Speed and Image Consistency in ComfyUI

Flux 2 Klein 9B KV makes edits from reference images faster. Learn background changes, custom ratios, inpainting, outpainting, edits with two to four images, and guiding the result with 3D shapes from Blender.

Good to know: Update ComfyUI first: if the KV node is red, your ComfyUI is not updated.

Ernie Model in ComfyUI - Worth It? + New Nodes Guide

Learn the difference between ComfyUI, Easy Install and Pixaroma nodes, use the new Note and Resolution nodes, and test Ernie Image Turbo on food, text, anime, paintings and real people.

Good to know: Uses a GGUF model: Q8 on 24 GB of VRAM, Q6 or Q4 for less.

ComfyUI Anima Base & Microsoft Lens + New Pause Image Node

Generate anime images with Anima Base and let Gemma 4 write its prompts, try Microsoft's Lens model, and use the new Pause Image node to check an image before you upscale it.

Good to know: Lens needs more VRAM than Anima.

ComfyUI Krea 2 Workflows for Better Text to Image Results

Generate images from text with Krea 2: a simple workflow, LoRA styles, a prompt enhancer, a low VRAM version, and an extra pass with a second KSampler for more detail and 2K images.

Good to know: Runs well on 12 GB of VRAM, text to image only. Episode 25 has a faster INT8 version.

ComfyUI INT8 Performance Boost! Boogu, Krea2 & Z-Image

What INT8 models are and how they can be faster with less VRAM, tested with Boogu, Flux Klein, Krea 2 and Z Image Turbo, plus the new Save Image Pixaroma node and Seed updates.

Good to know: INT8 is not always faster: it depends on the model and your GPU. Needs ComfyUI 0.27 or newer.

ComfyUI Krea 2 Edit + MiniMax H3 Prompts Generated Locally

Edit images with Krea 2 and its Edit LoRA: new outfits and backgrounds with the same person, or a character in a new scene. Then write MiniMax H3 prompts on your own PC.

Good to know: Needs two extra custom nodes, Krea 2 Enhancer and Krea 2 Identity.

ComfyUI Qwen Image 2.1 Tutorial | Generate, Edit, Inpaint & More

Generate and edit images with Qwen Image 2.1 in more than 20 workflows: text to image, transparent PNG, inpaint, remove backgrounds, character sheets, outpaint, and edits with up to six reference images.

Good to know: Non-commercial license. It is best at editing; for text to image, Krea 2 is still the better pick.

Free AI Model for Design & Text in ComfyUI | Ming Image 0.1

Ming Image 0.1 is made for design and text. Generate posters, thumbnails, logos, app screens and social posts, then edit images, like changing the words in a design.

Good to know: Better for illustrations and design than for realistic images. MIT license, so commercial use is allowed.

Flux 3 Klein in ComfyUI | Generate, Edit & Combine Images

Test Flux 3 Klein, the new Black Forest Labs model, in 19 free workflows: generate an image, edit one image, and combine several images into one. Plus where it struggles and what helps.

Good to know: Needs ComfyUI 0.41 or newer. The Flux 3 files on Hugging Face need a free approval: log in and click Agree.

How to Upscale Images in ComfyUI

Seven ways to make your images bigger and sharper: a simple resize, upscale models, Z-Image and Fluxmania with denoise, Seed VR 2, and the paid Nano Banana Pro and Recraft.

Good to know: Nano Banana Pro and Recraft are paid API nodes. The Fluxmania workflow needs Nunchaku (Episode 2).

ComfyUI for Image Manipulation: Remove BG, Combine Images, Adjust Colors

Use nodes instead of Photoshop: remove backgrounds, even for a whole folder, combine images like layers, add filters and color changes, then connect it all with Flux Klein edits in one run.

Good to know: The last workflow builds on the Flux Klein edit workflow from Episode 4.

ComfyUI Inpainting + Batch Load Images from a Folder

Edit only one part of an image with Inpaint Crop and Inpaint Stitch, replace a background using a mask, and run a whole folder of images through one workflow with Load Images From Folder.

Good to know: Builds on the Flux Klein edit workflow from Episode 4 and Remove Background from Episode 18.

Train Krea 2 LoRA Online, Use It Locally in ComfyUI

Train a Krea 2 LoRA online with fal AI for a style or a character, convert it with Krea Lora Converter so it works in ComfyUI, and test its strength with XY Plot.

Good to know: Training uses fal AI, a paid online service (about 3 dollars per LoRA when it was recorded).

How to Outpaint in ComfyUI + New Control Panel, Run Log & Text Join

Outpaint an image in one or more directions with Flux Klein 9B and keep the original image intact. Plus the new LoRA Loader, Text Join, Run Log and Control Panel nodes.

50 Free ComfyUI Automation Workflows

50 free workflows: point a workflow at a folder, press Run, and it does the whole folder. Resize, watermark, edit with Flux Klein, make many images in one Run, and caption images with Qwen3-VL.

Good to know: Every workflow has its own chapter, so you can jump to the one you need.

ComfyUI Qwen VL3: Creating Prompts from Images and Text

Use Qwen VL3 to turn an image, a text, or both into a detailed prompt, then connect it to an image workflow like Z-Image so you do not have to copy and paste.

Good to know: A big prompt model and a big image model in one workflow may not fit in your VRAM.

Qwen 3.5 in ComfyUI + Align Tool & Pixaroma Nodes Updates

Generate prompts from text or from an image with Qwen 3.5 and send them to a Flux Klein workflow, plus the new Align tool and updates to Preview Image, Resolution, Image Crop and Note.

Good to know: Qwen 3.5 and Flux Klein in one workflow need VRAM for two models. Episode 18 does the same with Gemma 4.

Gemma 4 + New ComfyUI Nodes That Make Prompting Easy!

Let Gemma 4 write prompts from text or images, transcribe audio and describe videos, then use the new Switch, Prompt Stack, Prompt Pack, Prompt Multi, Prompt Reader, Text Overlay and Remove Background nodes.

Good to know: Gemma 4 needs about 8 GB of VRAM.

ComfyUI XY Plot & Find/Replace Nodes + GitHub Suspended My Account

Change words in a prompt before it is used with Find and Replace, and test many settings at once in a grid with XY Plot, plus the move from GitHub to GitLab.

Good to know: The nodes start after the news about the move from GitHub to GitLab.

Pause LLM Text and Create a Reusable Prompt Library in ComfyUI

Build a reusable prompt library in the Prompt Pixaroma node with tags, random styles and lists. Then pause the text from an LLM, edit it, and let the workflow continue.

ComfyUI Video Models: InfiniteTalk + Wan 2.2 + SCAIL + LTX-2

Image to video with Wan 2.2, character animation from a reference video with Wan Animate and SCAIL, talking avatars with InfiniteTalk, LTX-2 with your own audio, and video upscale.

Good to know: Video models are heavy: tested on 24 GB of VRAM. Wan 2.2 with small GGUF models can run on 8 GB.

ComfyUI Scail 2 Video: WiFi Nodes, New Groups & Loops

Make a character image follow the motion of a reference video with Scail 2, and keep big workflows clean with the new Set and Get WiFi nodes, Pixaroma groups, Combine and loops.

Good to know: Scail 2 needs about 16 GB of VRAM and the KJ Nodes.

ComfyUI MiniMax H3: Best Video Generation Workflows

Make MiniMax H3 videos from text, from a first frame, a last frame or both, and from several reference images, plus a character that sings or speaks to your audio.

Good to know: Big sizes are slow: 5 seconds in Full HD took about 10 minutes. Faster ways come later: the Speed LoRA (Episode 32) and FastH3 (Episode 35).

ComfyUI MiniMax H3 Speed LoRA + VRAM Monitor Nodes

Make MiniMax H3 videos faster with the Speed LoRA, from 20 steps down to 8 or even 4, and keep an eye on your VRAM with the Monitor Pixaroma node.

Good to know: 8 steps is the best balance of speed and quality. The model setup is in Episode 29.

MiniMax FastH3 in ComfyUI | Text to Video and Image to Video

Make videos faster with MiniMax FastH3, in 8 steps: text to video, first frame to video, and first and last frame to video, with long prompts from a free Gemini Gem.

Good to know: Start ComfyUI normally, without Kitchen Attention: the attention nodes are inside the workflows.

Free AI voice in ComfyUI, Qwen3-TTS Clone Voice and Custom Voice Design

Make free AI voices with Qwen3-TTS: use ready-made speakers, design a new voice from a text prompt, clone a voice from a short audio sample, save voices, and make dialogues with several voices.

How to Use Fish Audio S2 Text to Speech in ComfyUI

Install Fish Audio S2 text to speech, fix the common errors, add tags for emotion and pauses, clone a voice from a short sample with its transcript, and make dialogues with several speakers.

Good to know: Needs around 20 GB of VRAM. The license is for research and non-commercial use.

Make Images React to Music in ComfyUI + ACE-Step AI Music

Preview an image before you save it, make an image move to music with Audio React and save it as an MP4, then generate free music from text with ACE-Step XL Turbo.

Good to know: Audio React moves the image with code, not AI, so the motion is simple 2D.

Stable Audio 3 in ComfyUI: Create AI Music and Sound Effects

Generate instrumental music and sound effects from text with Stable Audio 3 Medium, let Gemma 4 write better audio prompts, even from an image, plus node color and Load Image updates.

Good to know: Stable Audio 3 makes instrumentals only, no lyrics.

MiniMax Music 3 + Local AI Prompt Generator

Turn a short idea into a detailed prompt with AI Prompt Pixaroma, for Krea 2 and Z Image Turbo. Then make songs with MiniMax Music 3, with caption and lyrics from a local AI.

Good to know: The song length is a maximum, so a song can end early. Try a fixed seed.

Image to 3D Made Easy: Trellis 2 + Pixal3D in ComfyUI

Turn one image into a 3D model with Trellis 2 or Pixal3D, with colors or textures, and compare them. Then generate a clean image from a 3D model with Flux Klein.

Good to know: Trellis 2 is the one to start with. The texture workflows are very slow on a small card.

Pixal3D Multiview in ComfyUI | Load & Save 3D Nodes

Get more accurate 3D models with Pixal3D Multiview by giving it the front, back and side views. Plus the new Load 3D and Save 3D Pixaroma nodes, and clean renders with Flux Klein.

Good to know: Continues Episode 33.

I Went Full Mad Scientist in ComfyUI - Pixaroma Nodes

The first Pixaroma Nodes: how to install them, then use 3D Builder, Image Composer, Paint and Image Crop to prepare the image for an edit model, and Image Compare to check the result.

Good to know: This is the first version of the nodes. Episode 13 shows them rebuilt, with more options.

Vibe Code Your First ComfyUI Custom Node Step by Step

Make your own ComfyUI custom node with AI, no coding needed: how a node is made, where the files go, a simple node with Gemini, then one with a preview window using Claude.

Good to know: Uses online AI services. Gemini has a free plan with limits.

ComfyUI Pixaroma Nodes Update 2: Better Composer, 3D Builder, Paint

Pixaroma Nodes update: the rebuilt Image Composer with placeholders and background removal, Paint with tablet pressure, and the 3D Builder with your own GLB or OBJ models, all used in Flux Klein edit workflows.

Good to know: Uses the Flux 2 Klein 9B edit workflow from Episode 4.

ComfyUI Pixaroma Nodes: New Load Image, Notify & Utility Nodes

The new Load Image Pixaroma node with its filename output and resize options, Copy and Open buttons, save folders by date, a switch for width and height, and small Number and Text nodes.

Good to know: Audio React needs about 16 GB of free system RAM for Full HD.