Concurrent Submit | Zhenzhen_Z_Image_Turbo

The speed tier of the pack, in the one place the category label actually fits: zhenzhen/Qwen. Zhenzhen_Z_Image_Turbo wraps Alibaba's Qwen Z-Image Turbo - a distilled, small-step image model (the num_inference_steps default is 8, which is the whole story) - and this Submit node runs it through the 30-slot image pool.

Z-Image Turbo is the kind of model the distillation KB is about: you take a big diffusion model and train it to generate in a fraction of the steps. Eight steps is draft-tier speed. The Qwen category placement isn't cosmetic - this node also breaks the pack's usual config pattern in a way worth knowing.

The one thing that's different about this node

Every other image node in the family takes an api_config connection from the settings node. This one takes a bare apikey string field instead. There's no api_config input in the schema at all. If you're used to the Seedance2 settings pattern, the Z-Image node is where the pack's older style survives: paste your key into the apikey widget (or it falls back to a stored config). It also exposes real sampling knobs the other nodes hide:

• guidance_scale - 0–20 (default 0, which means "let the model decide").
• num_inference_steps - 1–50, default 8. This is the turbo dial; more steps = slower but steadier.
• output_quality - 0–100, default 80, for the JPEG path.
• seed - a real INT seed up to 2^31-1, not just cache control. This one can actually be sent to the API.

Other inputs that matter

• prompt - the text. Multiline.
• model - h-image-turbo. One option; it's the model name, not a menu.
• size - eight presets from 512x512 up to 1536x1024 / 1024x1536, plus Custom for the custom_size string (1280x720 format).
• output_format - jpeg, png, or webp.
• skip_error - on, the node returns a blank image instead of raising on failure.

Output is a task (COMFLY_IMAGE_FUTURE) → Concurrent Collect Images (30) → IMAGE tensors + status.

Install

ComfyUI Manager → search "Comfyui-zhenzhen" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

No model downloads. You need a Zhenzhen key - for this node, that goes in the apikey field.

Where people get burned

• Wrong config path. People wire the Seedance2 settings node into this expecting api_config - there's no such input. Put the key in the apikey widget.
• Eight steps is a draft, not a finish. The turbo tier trades fidelity for speed; if the results look rough, raise num_inference_steps before you blame the API. It's one of the few places this pack lets you actually turn the sampler dial.
• Custom size format is strict. widthxheight, no spaces, 1280x720 - and sizes outside 64–2048 get rejected up front.
• The 30-slot pool still means 30 billed calls. Turbo is cheap, but cheap × 30 slots × n is still a number.

The pitch for this node is simple: it's the fastest, cheapest image path in the pack, in a batch. If you need a flood of quick concept tiles or thumbnail drafts, this is the one to queue 30 of and walk away.