Info
This workflow is made specifically for Z-Image Turbo AIO models in mind. There is only one node you interact with, all the background tasks are in the Subgraph.
I tried to make it as simple as possible, the only resources you are going to download are models and LoRAs of your choice.
If you encounter any missing nodes, see the info page how to get them, or simply install them from ComfyUI Manager.
Description
What's new in this version
Universal Switch node to simplify execution flow
FAQ
Comments (4)
I don't understand. Does this require running both OLM software and COMFYUI simultaneously? OLM uses a quantized QWEN3-8B VL model to read and analyze optimization hints (this model is loaded in VRAM and is always available) + ZIT (the Z model is always loaded in VRAM) to generate the image? Does that mean at least a 4090-24GB graphics card is needed? Or is the entire process done within COMFYUI? (COMFY automatically uses dynamic loading/unloading technology)
All is managed with ComfyUI nodes. Ollama should run on your system, but it does not use VRAM until the workflow requires it. Then the Ollama model (in my case gemma3:12b) is unloaded and the workflow continues.
First run takes ca. 3 minutes on my machine (NVIDIA RTX 5060 with 16GB VRAM), but all subsequent renderings take 40-50 seconds to complete, including upscale.
Unless you choose different Z-Image model it will continue to render faster. Thanks to CacheDiT Accelerator, the rendering time is cut almost in half (not the upscale).
If you do not want Ollama to be running on your system, just use option 1️⃣ and let your own prompt do the job. You should probably bypass Ollama nodes in Subgraph too. But, really, Ollama does not eat your resources. It simply resides in the background and is only used when you decide to use it (option 2️⃣).
Just try it.
@Zoltar358 Thank you very much for your reply.
@wyxzddsjj919 No problem. Glad to help. Let me know if it works for you, I am really curious.