Z-Image Turbo
by Tongyi-MAI
Three LoRAs at once, at roughly a hundred generations per dollar.
What it is
Z-Image Turbo is "a super fast text-to-image model of 6B parameters" from Tongyi-MAI, running eight-step inference. Its documentation puts the arithmetic plainly: roughly 118 generations per dollar. It is by a wide margin the cheapest way in this studio to put your own trained style on an image.
The reason to care is the LoRA support. You can apply up to three LoRA weights at once, combining a custom style, a character adapter and a set of brand guidelines in a single request , without retraining anything. Most models let you attach one. This one lets you stack.
It takes up to four images per request, in JPEG, PNG or WebP, at whichever aspect ratio you need.
What it is good at
A style, a character and a brand look applied together. That combination normally means training a new model; here it is three attachments.
Around a hundred generations per dollar at the provider. This is the model you use when the answer is "try it and see".
Up to four images per request, so a single run gives you options rather than one answer.
Where it struggles
- Inference steps are limited to a range of one to eight. That is what makes it fast, and it is also its ceiling , this is not the model for the final, most detailed render.
- A safety checker is on by default and filters certain content automatically.
- It is a 6B-parameter model. For photographic fidelity or complex scenes, the larger models in the studio pull ahead noticeably.
How to prompt it
Keep it short and lean on the LoRA. The point of this model is that your trained style carries most of the look, so the prompt only has to say what is in the picture, not what it should feel like. Long atmospheric prompts are wasted here.
An example that shows what that means
A three-quarter portrait of a woman in a linen shirt against a plain warm-grey backdrop, soft frontal light, shallow depth of field, shoulders square to camera.
What we offer here
Straight from the catalogue, so this list never goes stale. The price is the provider’s published rate with our markup on top, and it is on the button before you run anything.
Questions
How many LoRAs can I use?
Up to three at once. That is the main reason to pick this model , a style, a character and a brand look can be applied in a single request.
Why is it so cheap?
It is a 6B-parameter model running eight-step inference. The trade is fidelity: fast and inexpensive, but not the one for a final high-detail render.
How many images per run?
Up to four, in JPEG, PNG or WebP. The price scales with the count, and the Studio shows the total before you run.