Z Image Turbo

Z-Image Turbo is the distilled version of Alibaba's Z-Image - the model that took the local photorealism crown from Flux - and distillation means it trades a little refinement for a lot of speed. It's the node you reach for when you want Z-Image's strong text encoder and photo-real output but need volume: iteration, batch drafts, moodboards. It ships in the BizyAirPlus pack in the Self-Hosted category, served as an API call rather than a local weights install.

Here's the giveaway that it's a distilled model: the node has no steps and no guidance_scale inputs. The Base edition of this same node exposes both (28 steps, CFG 3.0); Turbo doesn't, because distilled models are tuned to run at a fixed, low step count and don't respond to CFG the way full models do. The input list reads like a trimmed Base: prompt, negative_prompt, width/height (256–2048, default 1024), and seed (-1 = random).

How it works

The one required input the Base node doesn't have is batch_size - "number of units generated," so one call can return several images. That plus the missing sampling dials is the whole design story: this is a fire-and-forget generator. Outputs are image, request_id, and urls. Like its Base sibling, it's grouped Self-Hosted, and the api_config input (per-request base_url/api_key override) is how you point it at the endpoint you actually run - the README doesn't document that server setup, so it's on you to know what you're connecting to.

The inputs that matter

• prompt - the model's text encoder is its calling card. Long, specific prompts with typography and recognizable content are where it shines.
• batch_size - number of images per call. This is the volume knob that Turbo exists to make cheap.
• width / height - 256–2048; the defaults are fine for drafts.

Plus seed, skip_error, and api_config.

Installation

One-time, for the whole pack:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Or ComfyUI Manager, search BizyAirPlus. Restart, add your key, and check the endpoint via api_config if you're self-hosting.

Common issues

• "Why can't I set steps?" - that's the distillation, not a missing feature. Turbo runs fixed-step; the Base node is the one with sampling dials.
• Quality ceiling surprises you - distilled models lose some refinement versus Base by design. If a shot needs to be perfect, that's the Base node's job.
• Endpoint unreachable - Self-Hosted means you own the server. Confirm api_config's base_url.
• Cloud failures - key and network first, then the ComfyUI console for the detailed error.