EvoLink Z-Image Turbo (Official)

Here's the irony: Z-Image Turbo is the model the local community loves precisely because it runs anywhere - 6B parameters, ~30 seconds on a 3060, no GPU sweat. So why would anyone call it over an API? Because not everyone has the 3060. This node is the official EvoLink wrapper for Z-Image Turbo's hosted tier, and it exists for the same reason all API wrappers exist: no local GPU, or a workflow that's already in the cloud and wants one consistent path. Same model, different doorstep.

What it does

Text-to-image, fast and cheap. Z-Image Turbo is the distilled variant - 8 or so steps, CFG-guided once, built for speed - and on EvoLink's side that speed shows up as a low per-image cost and a quick turnaround. It's the budget tier of the EvoLink image lineup, and the README frames it for batch and "real-time-ish" scenarios. If your workflow is iterating on a dozen ideas, this is the image node that won't make you watch a progress bar all afternoon.

Mechanically it's the pack standard: submit to /images/generations, poll, download before the 24-hour result links expire, return an IMAGE tensor ready for Save Image. Nothing runs locally.

The inputs that matter

• prompt - required, Chinese-friendly.
• size - auto (model default) or a ratio: 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, 1:2, 2:1. No pixel-string option here - it's ratios only, which is fine because that's mostly what you'd pick anyway.
• seed - 0 = random; a fixed seed reproduces the result. Worth using if you're iterating and want a stable reference point.
• nsfw_check - a boolean NSFW content check, off by default. Z-Image locally is famously uncensored, so this switch is the cloud version's way of giving you the choice; flip it on if you're on a platform that demands it.

Outputs

The standard five: image (downloaded, → Save Image), task_id (job-log receipt), status_info, result_urls (24h links), response_json. Same shape as every EvoLink image node, so downstream wiring transfers.

Install and gotchas

Part of the comfyui-evolink pack. Manager → search EvoLink → Install → fully restart ComfyUI (close the console window; a browser refresh isn't enough). Or:

cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink

Only requests as a dependency; no model files, no GPU. API key from evolink.ai/dashboard/keys, fill once, blank afterward so it doesn't get shared inside workflow JSON.

The honest caveat: if you have a half-decent local GPU, running Z-Image Turbo locally is free-after-electricity and this node is a convenience purchase, not a necessity - which is a different trade than the Seedream/Veo/Sora nodes in this pack, where local isn't an option at all. Wrong node in your search results usually means the pack wasn't fully restarted after install; "finished but no image" means the image output isn't wired to a Save/Preview node (rerunning is cached and free). 401 bad key, 402 no credit, 429 rate limit, free failed tasks - the usual.