ComfyUI Tutorial Series Ep 72: Z-Image Turbo Workflows, ControlNet Essentials & LoRA Training
In this episode, you’ll learn how to set up and use the Z-Image Turbo model in ComfyUI, compare BF16 vs FP8 performance, build ControlNet-based consistency workflows, generate prompts with Qwen VL, and train effective LoRAs for characters and styles.
You’ll see full workflow demonstrations, speed comparisons, prompt techniques, and practical tips for avoiding common issues like oversmoothing, unwanted artifacts, or inaccurate styles. This tutorial is ideal for users who want realistic renders, consistent characters, and optimized settings for image-to-video pipelines.
What You’ll Learn
How to install, configure, and choose between Z-Image BF16, FP8, and All-in-One models
Best settings for steps, schedulers, CFG, and resolution
How image dimensions affect clarity and style
Using Qwen VL to auto-generate prompts from reference images
ControlNet basics with Canny, Depth, and DW pose preprocessors
How to train a LoRA using FAL: steps, trigger words, settings, and troubleshooting
Get the workflows and instructions from discord for free
Step 1: use this link to join if you didn't join already / discord
Step 2: go to this post to get the workflows / discord
More about Z-Image Model
https://huggingface.co/Tongyi-MAI/Z-I...
https://huggingface.co/SeeSee21/Z-Ima...
ControlNet
https://huggingface.co/alibaba-pai/Z-...
Lora Training on Fal
https://fal.ai/models/fal-ai/z-image-...
Check Other Episodes
• ComfyUI Tutorial Series: Mastering Stable ...
Unlock exclusive perks by joining our channel:
/ @pixaroma
Grab your Pixaroma merch from here: / @pixaroma
If this tutorial helped you, leave a like, comment, and subscribe for more ComfyUI workflow tutorials.
#comfyui #zimage #comfyuitutorial
00:00 Introduction to Z-Image
01:30 Z-Image Turbo Setup, Settings, Tips
04:59 Comparing Z-image models BF16 vs FP16 vs AIO
07:15 Higher Resolution Generation, Upscale
11:13 Z-Image Turbo + Qwen VL - Get better prompts
14:07 Z Image Turbo ControlNet
15:35 Training Z-Image Lora, Testing Lora