Z-Image Turbo LoRA · Text to Image

Z-Image Turbo is the 8-step distilled model that made 6B parameters feel like plenty: fast, Apache 2.0, uncensored, and small enough that it became the local photorealism default. This node puts Turbo online with a loras socket, so you can stack up to three of your own LoRA files per request, ~1MP in nine aspect ratios, JPEG/PNG/WebP, $0.012 an image.

Who this is for

LoRA users without a great GPU, mostly. A style LoRA you downloaded, a subject LoRA you trained, a character someone published - all of them normally mean a local workflow, a 6B checkpoint, and a minute per image on modest hardware. Here you wire a SpicyAPI LoRA node holding a direct .safetensors URL into this node's loras socket and get a picture in a second or two of server time. Twelve images for the price of a coffee, and no disk space consumed.

The second, sneakier use: dataset and LoRA evaluation at volume. Prompt sweeps and seed sweeps are how you find out whether a LoRA is over-baked, and doing that locally is slow. Doing it here is cheap.

The inputs

• prompt - subject, composition, lighting, in sentences. If the LoRA has a trigger word, put it first; forgetting the trigger is the single most common "my LoRA does nothing" failure.
• aspect_ratio - 1:1 by default, nine ratios available.
• output_format - jpeg (default), png or webp.
• seed - fixed by default, so the same seed and settings reproduce the same image.
• Optional loras - fed by a SpicyAPI LoRA node: a direct download link to a .safetensors file plus a strength. Chain several nodes to stack, up to three per request. The link has to open without a login; a Civitai model page won't, a raw file URL will. On this model the LoRA strength range goes up to 4, versus 0–1 on the editing endpoints, so check which node you're on before wondering why 1.0 looks weak.

Outputs: images (a normal IMAGE, so Save Image or anything downstream) and task id (STRING, the SpicyAPI task id for logs or a support request).

Install

cd ComfyUI/custom_nodes
git clone https://github.com/spicyapi-ai/comfyui-spicyapi

Manager → search SpicyAPI. Restart ComfyUI, then Settings > SpicyAPI, paste a key from spicyapi.ai/console/keys, and restart once more so the node list is rebuilt from your account's catalogue. SPICY_API_KEY for headless installs. No model downloads, no pip installs - an empty dependency list, using only what ComfyUI already ships (Python 3.10+, V3 node API, tested with ComfyUI 0.39). There's also a ready-made lora template in ComfyUI's template browser under comfyui-spicyapi, preloaded with an Apache-2.0 pixel-art LoRA for Z-Image Turbo, which is the fastest way to confirm the whole LoRA path works before you paste in your own link.

Watch out for

Stacking. Z-Image handles one LoRA well and gets unstable with several, which is a known local finding, not a hosted quirk - the three-LoRA limit is a ceiling, not an invitation. If you do stack, keep strengths low. Also resist asking for more resolution than the model gives: Turbo's output is about a megapixel natively, so pick a ratio rather than expecting a 4MP file.

Money-wise: set Max cost per run in the SpicyAPI settings (the pack quotes each request before submitting and stops anything over your ceiling), and leave the seed on fixed while you compare prompts - a re-queue then reuses the cached image instead of charging again. Errors you'll meet: No SpicyAPI key is set, 40310 (unverified email), 40004 (no deployment serves that combination), or a LoRA that fails to load because its URL wants a login.