Customize your input with more control.
Customize your input with more control.
Type # to reference inputs.
Hint: Drag and drop image files from your computer, images from web pages, paste from clipboard (Ctrl/Cmd+V), or provide a URL. Accepted file types: jpg, jpeg, png, webp, gif, avif, heic, heif
Customize your input with more control.
Your request will cost $0.01 per megapixel.
Tongyi-MAI's Z-Image Turbo delivers controlled image-to-image generation at $0.01 per megapixel through a 6-billion parameter architecture. Trading raw size for inference speed, this model processes edge, depth, and pose-guided transformations in seconds while supporting up to 3 custom LoRA weights simultaneously. Built for developers who need predictable costs and sub-10-second iterations on reference-based image workflows.
Built for: Style Transfer Pipelines | Pose-Guided Character Design | Depth-Aware Scene Editing
At $0.01 per megapixel, Z-Image Turbo runs 4-10x more cost-effectively than full-scale diffusion models while maintaining controllable output through structured conditioning inputs.
| Metric | Result | Context |
|---|---|---|
| Inference Steps | 1-8 steps | Default 8 steps; reduce to 4-6 for draft iterations |
| Inference Speed | Sub-10 seconds | 6B parameter model optimized for turbo inference |
| Cost per Megapixel | $0.01 | 100 generations per $1.00 at 1MP resolution |
| Max Resolution | Auto-scaling | Dynamically adjusts to input image dimensions |
| Related Endpoints | Image-to-Image, Image-to-Image LoRA | Standard variants without ControlNet preprocessing |
Z-Image Turbo combines ControlNet conditioning with LoRA customization in a single inference call, eliminating the multi-stage pipelines typical of traditional image-to-image workflows. Rather than chaining separate models for preprocessing, style application, and refinement, you feed a reference image and select from four conditioning modes: canny edge, depth mapping, pose detection, or none, with control strength adjustable across the inference timeline.
What this means for you:
preprocess parameter, no separate detection models requiredcontrol_start and control_end (0.0-1.0 range) to determine when ControlNet guidance activates during the 1-8 step generation process| Spec | Details |
|---|---|
| Architecture | Z-Image Turbo |
| Input Formats | Image URL (jpg, jpeg, png, webp, gif, avif) + text prompt |
| Output Formats | JPEG, PNG, WebP |
| ControlNet Modes | None, Canny Edge, Depth Map, Pose Detection |
| License | Commercial use permitted |
API Documentation | Quickstart Guide | Pricing
Z-Image Turbo Image-to-Image ($0.01/MP) – Z-Image Turbo ControlNet LoRA adds preprocessing automation and multi-LoRA support at identical pricing, trading simplicity for advanced control workflows. The standard image-to-image endpoint remains ideal for straightforward style transfers where manual preprocessing is acceptable.
FASHN Virtual Try-On V1.5 – Z-Image Turbo ControlNet LoRA provides general-purpose image transformation with pose and depth control for diverse creative workflows. FASHN specializes in garment-to-model fitting with body-aware warping for e-commerce product visualization.