Forem: Jean

Getting Started with Qwen Lightning v1.1 in ComfyUI for 8GB VRAM

Jean — Fri, 15 Aug 2025 17:37:50 +0000

Artificial intelligence image generation has never been easier or faster than with Qwen Lightning v1.1 in ComfyUI. Whether you’re a beginner or an experienced AI artist, this guide walks you through every step of installation, configuration, and optimization, ensuring a smooth workflow and stunning results.

🎥 Watch the full step-by-step video tutorial here: https://youtu.be/7p23f-gDp4I

Step 1: Download the Necessary Models

GGUF Model: Download the Qwen Lightning v1.1 GGUF model for 8GB VRAM.

Text Encoder: Download the FP8 scaled text encoder.

VAE: Download the associated VAE file.

These components are essential for generating high-quality, detailed AI images.

Step 2: Organize Your Models

Place the downloaded files in the appropriate ComfyUI folders:

GGUF → ComfyUI/models/unet

Text Encoder → ComfyUI/models/text_encoders

VAE → ComfyUI/models/vae

LoRA → ComfyUI/models/loras

Correct placement ensures ComfyUI can access all resources seamlessly.

Step 3: Update ComfyUI and Load the Workflow

Open the ComfyUI Manager and click Update All to ensure the UI, nodes, and templates are up-to-date.

Download the example workflow JSON from Hugging Face.

In ComfyUI, navigate to Workflow → Open and select the JSON file.

Assign loaders: UNet = GGUF, VAE = VAE, CLIP = FP8 scaled.

Step 4: Install and Configure Qwen Lightning v1.1 LoRA

Place the LoRA file in the loras folder.

Load it using the LoRA Loader node.

Adjust the strength (0.7–0.9 to start) to influence the image generation while maintaining clarity and style.

Step 5: Optimize for Speed and Quality

Enable Sage Attention and FP16 Accumulation to reduce VRAM usage and speed up generation.

Adjust CFG between 4–4.5 to balance quality and rendering speed.

With these settings, Qwen Lightning v1.1 produces sharp, vibrant, and polished images compared to v1.0 or LightX2V.

Step 6: Generate Images

Test your workflow with sample prompts. Adjust LoRA strength and CFG as needed. Qwen Lightning v1.1 delivers clean edges, vivid colors, and excellent text clarity in complex scenes.

This workflow is perfect for AI artists, developers, and content creators seeking high-quality results with minimal setup. Whether creating cyberpunk environments, fantasy worlds, or realistic portraits, Qwen Lightning v1.1 in ComfyUI is a reliable, professional-grade solution.

Run Wan 2.2 in ComfyUI with Just 8GB VRAM — Full Image-to-Video AI Workflow

Jean — Sun, 03 Aug 2025 02:08:58 +0000

This tutorial walks through how I installed and ran the Wan 2.2 image-to-video AI model in ComfyUI on an 8GB VRAM machine. It covers downloading the 5B GGUF model, placing Lightx2v LoRA, AutoVAE, and UMT5 encoder into the correct folders, and importing the official Hugging Face workflow. I also demonstrate how I used Gemini CLI to fix the missing sageattention module and generate a cinematic wolf animation from a single image.

🎬 Watch the full video walkthrough: https://youtu.be/7hUO6KhUsvQ

I Built Games & Apps with Qwen3-Coder Without Writing Any Code (Full Tutorial) 🤯

Jean — Sun, 27 Jul 2025 22:55:35 +0000

Hey Dev Community!

The pace of AI in development is staggering. There's a new state-of-the-art model every few weeks, but I wanted to go beyond the benchmarks and see what it's actually like to build with one. I spent a day with the new Qwen3-Coder and used it to generate five fully functional applications from scratch.

This post is a summary of that journey, with the full video walkthrough embedded below. My goal was to test its "agentic coding" capabilities and show a practical, easy way for anyone to start experimenting.

https://youtu.be/l65aOfy4NgQ

What is Qwen3-Coder?
In short, it's a new open-source coding model from Alibaba's Qwen team. Its key feature is being "agentic," meaning it's designed to do more than just complete code—it can understand a high-level goal, plan the steps, and generate the complete application logic.

The "Easy Button": Alibaba Cloud Model Studio
Instead of a complex local installation, I used the official Alibaba Cloud Model Studio. It's a web-based playground that gives you:

✅ Free Access: You can use the full suite of Qwen3-Coder models.

💰 A Massive Free Token Grant: New users get over 50 million tokens, which is more than enough to build some seriously cool projects.

The video includes a full, step-by-step tutorial on how to get signed up and claim the tokens, which takes you from having no account to being ready to code in about 10 minutes.

The 5 Projects I Built with a Single Prompt Each 🚀
Here’s what I prompted Qwen3-Coder to build. The video shows the entire process of generating the code and testing each one in CodePen.

3D Car Racing Game: A simple but functional racing game with keyboard controls and a UI.

3D Plane Fighter Game: An air combat game where you can fly a plane and shoot at enemy targets.

AI Image Generator Website: A complete front-end for a DALL-E style image generator, with a prompt area, settings, and an inspiration gallery.

3D Zombie Dragon FPS: A first-person shooter where you have to survive against waves of enemies.

Web-Based Music Sequencer: A multi-track sequencer with playable piano keys and different instrument tracks.

My Main Takeaway
The model's ability to generate the full stack (HTML, CSS, and JS) for these interactive apps was mind-blowing. The music sequencer and the website front-end worked nearly perfectly on the first try.

The 3D games required more specific prompting to get the controls right (e.g., specifying WASD movement, mouse look), which was a great lesson in prompt engineering for agentic tasks. It's not just about what you ask, but how you ask it.

I'm genuinely excited about what this means for rapid prototyping and solo developers. I'd love to hear your thoughts!

How I Fixed VACE Plugin Installation Errors in ComfyUI Using Gemini CLI

Jean — Fri, 11 Jul 2025 19:04:00 +0000

If you’ve been struggling to install the VACE AI video generator plugin in ComfyUI, this guide shows how I resolved the issue using Gemini CLI — a handy command-line tool for managing plugins.

The Problem
When trying to install VACE, the plugin failed to register correctly in ComfyUI, causing errors and preventing video generation. Common troubleshooting methods like restarting ComfyUI or reinstalling dependencies didn’t work.

Video Walkthrough
For a step-by-step visual guide, check out my full video walkthrough here:
📺 https://youtu.be/BmGJZSRFLJw

Final Thoughts
If you’re stuck with VACE install errors in ComfyUI, using Gemini CLI like this can save you time and frustration. Feel free to share your experiences or ask questions in the comments!

How to Run Gemini CLI Using NPX and Configure MCP Tools for Advanced AI Workflows

Jean — Tue, 01 Jul 2025 07:23:01 +0000

This post covers how to run Gemini CLI using npx without needing a global install. I walk through:

Running Gemini CLI on-demand with npx

Creating and fixing .gemini/settings.json

Securely adding your API key

Setting up MCP tools like taskmaster-ai and context7

Common errors and how to fix them (like JSON formatting and encoding issues)

If you’re exploring AI automation from the terminal and want flexibility without a full install, this guide is for you.

📺 Watch the full video walkthrough:
https://youtu.be/HpasFDFxsmk?si=78D76_6re8iszbAG