Run Z-Image Turbo on Free Google Colab with a T4
A Juejin AI article author reports running Z-Image Turbo INT8 in ComfyUI on the free tier of Google Colab, using an NVIDIA Tesla T4 with about 15GB of memory at 768×768 resolution. The documented setup also covers workflow installation, Google Drive backups, HTTP API access, and optional phone access through a temporary HTTPS tunnel; one recorded generation took about 10 seconds.
This personal walkthrough shows that a free Colab T4 can run a quantized open-source image model, providing a reproducible starting point for experimenting with ComfyUI and API-based image generation.
最近 Google AI Ultra 的土区订阅优惠挺火,也让我重新关注起了 Google Colab。
原本只是想看看会员送的 GPU 算力到底能做什么,没想到折腾了一圈,居然用免费的 T4 显卡成功跑通了开源 AI 生图模型。
不用买显卡,不用配置本地 CUDA 环境,甚至还能通过手机远程生成图片。
这篇文章记录完整过程,包括踩过的坑和解决办法。
一、先说最终效果
这次使用的配置:
#Z-Image Turbo INT8#Google Colab#NVIDIA Tesla T4#ComfyUI#Comfy-Org#Hugging Face#Google Drive#Cloudflare Tunnel