Skip to main content
Endpoint: POST https://fal.run/fal-ai/z-image/turbo/lora Endpoint ID: fal-ai/z-image/turbo/lora

Try it in the Playground

Run this model interactively with your own prompts.

​
Quick Start

import fal_client

def on_queue_update(update):
    if isinstance(update, fal_client.InProgress):
        for log in update.logs:
           print(log["message"])

result = fal_client.subscribe(
    "fal-ai/z-image/turbo/lora",
    arguments={
        "prompt": "A hyper-realistic, close-up portrait of a tribal elder from the Omo Valley, painted with intricate white chalk patterns and adorned with a headdress made of dried flowers, seed pods, and rusted bottle caps. The focus is razor-sharp on the texture of the skin, showing every pore, wrinkle, and scar that tells a story of survival. The background is a blurred, smoky hut interior, with the warm glow of a cooking fire reflecting in the subject's dark, soulful eyes. Shot on a Leica M6 with Kodak Portra 400 film grain aesthetic."
    },
    with_logs=True,
    on_queue_update=on_queue_update,
)
print(result)
Tongyi-MAI’s Z-Image Turbo with LoRA support delivers 6B-parameter text-to-image generation at $0.0085 per megapixel with 8-step inference. Apply up to 3 custom LoRA weights at inference time without retraining, enabling style consistency and brand adaptation at approximately 118 generations per dollar on fal. Built for: Custom style application | Brand-consistent generation | Character consistency workflows | Rapid design iteration

​
LoRA Flexibility Without Training Overhead

Z-Image Turbo LoRA adds inference-time style customization to the base model’s speed advantages. Apply pre-trained LoRA adapters directly through the API, combining multiple style influences in a single generation call without touching the underlying model weights. What this means for you:
  • Apply up to 3 LoRA weights simultaneously: Combine custom styles, character adapters, and brand guidelines in a single generation request through the loras parameter
  • Batch generation at scale: Generate up to 4 images per request with configurable inference steps (1-8 range), optimizing the speed-quality tradeoff for your use case
  • Acceleration options: Choose between “none”, “regular”, or “high” acceleration modes to balance generation speed against output fidelity
  • Production-ready safety: Built-in safety checker (enabled by default) filters NSFW content automatically, with optional prompt expansion for enhanced detail at +$0.0025 per request
  • Flexible output formats: Generate images in JPEG, PNG, or WebP with landscape (4:3), portrait (3:4), or square (1:1) aspect ratios

​
Performance That Scales

Z-Image Turbo LoRA adds minimal overhead to base model pricing while enabling custom style application at inference time.
MetricResultContext
Cost per Megapixel$0.0085~118 generations per $1.00 on fal
Inference Steps1-8 configurableDefault 8 steps balances speed and quality
Batch CapabilityUp to 4 imagesSingle API call generates multiple variants
LoRA SupportUp to 3 concurrent weightsApply custom styles without base model retraining
Base Model Cost$0.005/MPLoRA adds $0.0035/MP for style flexibility

​
Technical Specifications

SpecDetails
ArchitectureZ-Image Turbo (6B parameters) with LoRA inference
Input FormatsText prompts, optional LoRA weights (up to 3)
Output FormatsJPEG, PNG, WebP
Image SizesLandscape 4:3, Portrait 3:4, Square 1:1
Acceleration ModesNone, Regular, High
TrainingZ-Image Trainer for custom LoRA creation
LicenseCommercial use permitted
API Documentation | Quickstart Guide

​
How It Stacks Up

Z-Image Turbo – The base Z-Image Turbo endpoint runs at 0.005/MPformaximumcostefficiencywhenstylecustomizationisn′trequired.TheLoRAvariantadds0.005/MP for maximum cost efficiency when style customization isn't required. The LoRA variant adds 0.0035/MP overhead to enable custom style application, character consistency, and brand adaptation at inference time. AuraFlow – Z-Image Turbo LoRA prioritizes cost efficiency at $0.0085/MP with 8-step inference and runtime style customization. AuraFlow emphasizes open-source flexibility and longer inference paths for applications requiring maximum creative control and community-driven development. FLUX.2 [dev] LoRA – Z-Image Turbo LoRA delivers comparable style customization at 0.0085/MPversusFLUX.2[dev]LoRA′s0.0085/MP versus FLUX.2 [dev] LoRA's 0.021/MP, making it 2.5x more cost-efficient for high-volume workflows. FLUX.2 [dev] LoRA offers higher resolution outputs and more sophisticated prompt interpretation for applications where output quality justifies the premium.

​
Capabilities

  • Text prompt input
  • Configurable resolution
  • Adjustable inference steps
  • Reproducible generation (seed)
  • Synchronous mode
  • Batch generation
  • Safety checker
  • LoRA support

​
API Reference

​
Input Schema

​
prompt
string
required
The prompt to generate an image from.
​
image_size
ImageSize | Enum
default:"landscape_4_3"
The size of the generated image. Default value: landscape_4_3Possible values: square_hd, square, portrait_4_3, portrait_16_9, landscape_4_3, landscape_16_9
​
num_inference_steps
integer
default:"8"
The number of inference steps to perform. Default value: 8Range: 1 to 8
​
seed
integer
The same seed and the same prompt given to the same version of the model will output the same image every time.
​
sync_mode
boolean
default:"false"
If True, the media will be returned as a data URI and the output data won’t be available in the request history.
​
num_images
integer
default:"1"
The number of images to generate. Default value: 1Range: 1 to 4
​
enable_safety_checker
boolean
default:"true"
If set to true, the safety checker will be enabled. Default value: true
​
output_format
OutputFormatEnum
default:"png"
The format of the generated image. Default value: "png"Possible values: jpeg, png, webp
​
acceleration
AccelerationEnum
default:"regular"
The acceleration level to use. Default value: "regular"Possible values: none, regular, high
​
enable_prompt_expansion
boolean
default:"false"
Whether to enable prompt expansion. Note: this will increase the price by 0.0025 credits per request.
​
loras
list<LoRAInput>
default:""
List of LoRA weights to apply (maximum 3).

​
Output Schema

​
images
list<ImageFile>
required
The generated image files info.
​
timings
Timings
required
The timings of the generation process.
​
seed
integer
required
Seed of the generated Image. It will be the same value of the one passed in the input or the randomly generated that was used in case none was passed.
​
has_nsfw_concepts
list<boolean>
required
Whether the generated images contain NSFW concepts.
​
prompt
string
required
The prompt used for generating the image.

​
Input Example

{
  "prompt": "A hyper-realistic, close-up portrait of a tribal elder from the Omo Valley, painted with intricate white chalk patterns and adorned with a headdress made of dried flowers, seed pods, and rusted bottle caps. The focus is razor-sharp on the texture of the skin, showing every pore, wrinkle, and scar that tells a story of survival. The background is a blurred, smoky hut interior, with the warm glow of a cooking fire reflecting in the subject's dark, soulful eyes. Shot on a Leica M6 with Kodak Portra 400 film grain aesthetic.",
  "image_size": "landscape_4_3",
  "num_inference_steps": 8,
  "sync_mode": false,
  "num_images": 1,
  "enable_safety_checker": true,
  "output_format": "png",
  "acceleration": "regular",
  "enable_prompt_expansion": false,
  "loras": []
}

​
Output Example

{
  "images": [
    {
      "content_type": "image/png",
      "height": 768,
      "url": "https://storage.googleapis.com/falserverless/example_outputs/z-image-turbo-output.png",
      "width": 1024
    }
  ],
  "prompt": ""
}

​
Limitations

  • image_size restricted to: square_hd, square, portrait_4_3, portrait_16_9, landscape_4_3, landscape_16_9
  • num_inference_steps range: 1 to 8
  • num_images range: 1 to 4
  • output_format restricted to: jpeg, png, webp
  • acceleration restricted to: none, regular, high
  • Content moderation via safety checker