How to use this workflow

Navigation

Entire workflow is divided into 3 sections:

1. Info

2. Main

3. Subgraph

To navigate between sections use numbers [1-3]. I don't expect you would waste too much of your time for info screen, so stick to the main window after you learn how to use this.

It looks best on 1440p monitor. If you have a monitor of different size, you can modify zoom level in bookmark nodes to fit your needs.

Models

You can choose Turbo models here, both checkpoints (AIO) and diffusion. All necessary settings are already in a Subgraph.

Pay special attention to Model Patch Torch Settings node. If your graphic card does not support pytorch, simply bypass or remove the node.

LoRAs

I find it very convenient to use LoRA Manager to inject desired LoRA into workflow. Once it is done you can choose trigger words (if there are several ones) in the Subgraph. No need to manually type trigger words in your prompt. They will be concatenated automatically.

Unfortunately LoRAs work well only with original Z-Image Turbo model. If you choose derivatives, you might experience inconsistencies, even deformations. I've played with different samplers and schedulers, but results were not very good. You will find a small table on main screen.

Prompt

In this version there is only one prompt field. You decide if the workflow uses exactly your prompt 1️⃣ or let Ollama enhance it for you 2️⃣.

Ollama must be installed and server must be running of course. I use gemma3:12b. It's a bit slower than gemma3n, but produces better prompts. You can get this model by running this command: ollama pull gemma3:12b

### Image save location

The up-scaled image is saved in the following location: ComfyUI/output/Z-Image Turbo/[CURRENT DATE]/. All generation metadata is embedded into the final image. No problem with CivitAI reading metadata. All thanks to SaveImageWithMetaData node.

Notification sounds (optional)

Place my custom notification sounds in ComfyUI/custom_nodes/comfyui-custom-scripts/web/js/assets

---

The rest of the workflow is self-explanatory, I hope.

Enjoy!