<h2 id="how-to-use-this-workflow">How to use this workflow</h2><h3 id="navigation">Navigation</h3><p>Entire workflow is divided into 3 sections:</p><ol><li><p>Info</p></li><li><p>Main</p></li><li><p>Subgraph</p></li></ol><p>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.</p><p>It looks best on 1440p monitor. If you have a monitor of different size, you can modify zoom level in <strong>bookmark</strong> nodes to fit your needs.</p><h3 id="models">Models</h3><p>You can choose Turbo models here, both checkpoints (AIO) and diffusion. All necessary settings are already in a <strong>Subgraph</strong>.</p><p>Pay special attention to <strong>Model Patch Torch Settings</strong> node. If your graphic card does not support pytorch, simply bypass or remove the node.</p><h3 id="loras">LoRAs</h3><p>I find it very convenient to use LoRA Manager to <em>inject</em> desired LoRA into workflow. Once it is done you can choose trigger words (if there are several ones) in the <strong>Subgraph</strong>. No need to manually type trigger words in your prompt. They will be concatenated automatically.</p><p>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.</p><h3 id="prompt">Prompt</h3><p>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.</p><p>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: <code>ollama pull gemma3:12b</code></p><p>### Image save location</p><p>The up-scaled image is saved in the following location: <code>ComfyUI/output/Z-Image Turbo/[CURRENT DATE]/</code>. All generation metadata is embedded into the final image. No problem with CivitAI reading metadata. All thanks to <strong>SaveImageWithMetaData</strong> node.</p><h3 id="notification-sounds-(optional)">Notification sounds (optional)</h3><p>Place my custom <a target="_blank" rel="ugc" href="https://drive.google.com/drive/folders/1efJS9O5slbArdEV7SFw6x_9kGPGtL5tF?usp=sharing">notification sounds</a> in <code>ComfyUI/custom_nodes/comfyui-custom-scripts/web/js/assets</code></p><p>---</p><p>The rest of the workflow is self-explanatory, I hope.</p><p>Enjoy!</p>

Z-Image Turbo powered by Ollama

<ul><li><p>both checkpoint (AIO) and diffusion models supported</p></li><li><p>image is saved with all metadata, so CivitAI will have no problem reading it</p></li></ul>

Files with SHA256: b3d1256d099bd1cf1115160b7e38698cf900a051eebbb6a35794af1e46304cfb