AI on the Lot 2026 - Z-Image Turbo - t2i + Upscale
The second workflow released for AI on the Lot 2026, made available for ComfyUI 101 workshop attendees to open and follow along if they choose. Same text-to-image setup using Z-Image Turbo, now with an upscale step added for higher resolution output.
简介
AI on the Lot 2026 - Z-Image Turbo - t2i + Upscale is a streamlined ComfyUI workflow that takes you from a text prompt to a polished, high-resolution image in one pass. It centers on a custom text-to-image node (Z-Image Turbo; UUID f2fdebf6-dfaf-43b6-9eb2-7f70613cfdc1) to generate the initial image and then routes the output through an integrated upscale stage for added detail and clarity. The SaveImage node captures your outputs to disk, while the ImageCompare node produces a side-by-side composite so you can visually judge how much quality the upscale added—and whether any artifacts were introduced.
你需要提供
- 正在读取工作流
在 ComfyUI 中运行。 此工作流无法在本页面运行。你可以在 Comfy Cloud 中打开（已预先加载所需内容），或下载后在自己的 ComfyUI 中使用。
在 Cloud 中试用常见问答
The workflow does not bundle a specific model. The Z-Image Turbo node uses whichever checkpoint you load in your ComfyUI environment (as exposed by that node). Point it to your preferred base model and VAE, and the rest of the graph will work unchanged.
Use the upscale factor or strength controls attached to the Z-Image Turbo/upscale stage. Start at 2x and increase only if needed. If edges look crunchy, reduce the upscale factor, lower the sharpening/strength (if available), or slightly increase generation steps for a cleaner base before upscaling.
ImageCompare creates a side-by-side composite of the base (pre-upscale) image and the upscaled image. Use it to check detail gains and spot artifacts. If you don’t see the comparison, confirm both image outputs are connected to ImageCompare’s inputs before running.
Set and record a fixed seed in the Z-Image Turbo node and keep your prompt, steps, guidance, and upscale settings unchanged. SaveImage will write files with consistent naming; keeping the same seed and parameters will yield repeatable outputs with the same model and node settings.