Z Image Turbo ControlNet: Photoreal Image-to-Image with Depth & Pose Control

tongyi-mai/z-image/turbo/controlnet/lora

The most powerful version of Z Image Turbo. Combine ControlNet (Canny, Depth, Pose) for structure locking with custom LoRAs for style transfer in a single high-speed workflow.

Prompt *

Image *

URL of the input image used for ControlNet-based generation.

LoRAs *

LoRA #1

LoRA Path

URL, HuggingFace repo ID (owner/repo) to lora weights.

LoRA Scale

Scale of the LoRA model.

List of LoRAs to apply (maximum 3).

Aspect Ratio (W:H)

Control Scale

Controls how strongly the ControlNet conditions affect the output.

Control Start

Specifies the start point of the ControlNet conditioning during generation.

Control End

Specifies the end point of the ControlNet conditioning during generation.

Preprocess

Defines what preprocessing (if any) will be applied to the ControlNet input image.

Number of Inference Steps

Specifies the number of inference steps during generation.

Seed

Magic Prompt

Enables automatic prompt expansion to improve results; increases cost by 0.0025 credits per request.

Output Format

Specifies the output image format.

Idle

The rate is $0.01 per image.

Introduction to Z Image Turbo ControlNet LoRA

Z Image Turbo ControlNet LoRA is the ultimate precision tool for creators who need both structural control and stylistic freedom. It fuses the lightning-fast generation of Z Image Turbo with ControlNet's geometric guidance (Canny, Depth, Pose) and the limitless customization of LoRAs. This playground is designed for advanced workflows: upload a reference image to lock the composition, load custom LoRAs to define the art style, and generate high-fidelity results that strictly follow your layout while adopting your desired aesthetic—all in seconds.

Z Image Turbo ControlNet On X: Insights And Updates

Key Capabilities

Precise Structure Control: Use ControlNet to lock pose, depth, or edges from your input image. Perfect for keeping a character's posture or a room's layout identical while changing everything else.
Style Injection via LoRA: Load up to 3 custom LoRA models (via URL/Path) to apply specific art styles or character details on top of your controlled structure.
Advanced Conditioning: Choose from canny (edge detection), depth (3D distance), or pose (human keypoints) preprocessors to tell the model exactly what to respect from your reference image.
Fine-Grained Influence: Independently adjust Control Scale (how much the reference image matters) and LoRA Scale (how much the style matters) for the perfect balance.

How to use Z Image Turbo ControlNet LoRA

Upload Reference: Upload an image to the Image slot. This will serve as the structural guide.
Select Preprocess:

- Canny: For keeping strict outlines and details.

- Depth: For architectural renders or maintaining 3D volume.

- Pose: For changing a character's outfit or background while keeping their position fixed.

Load LoRAs: Add your desired style or character LoRAs in the LoRAs list.
Tune Control: Adjust Control Scale (Default 0.9). Lower it if you want the model to be more creative; raise it for strict adherence.

Pro Tips

Control Step Timing: Use Control End (Default 0.4) to let the structure be defined early, but allow the LoRA style to take over the details in the later steps.
Aspect Ratio: Ensure your Aspect Ratio matches your input image's shape to avoid stretching or cropping.
Magic Prompt: Enable Magic Prompt if you want the model to add more detail to your scene automatically, useful when your manual prompt is simple.

Related Tools

For simple text-to-image with styles (no structure control needed), use Z Image Turbo LoRA.
For pure text-to-image speed, use the base model: Z Image Turbo Text to Image.

Related Models

qwen-image/qwen-image-2512

Create realistic visuals from prompts with precise multilingual text control and balanced layouts.

qwen-image/edit

Edit images with AI for precise text and visuals.

qwen-edit-2509/lora/relight

Advanced relighting and multi-image fusion tool with fast ControlNet support for detailed, consistent design results.

qwen-image/qwen-image-edit-2511/lora

LoRA-based visual editing model offering structure-aware asset transformation for creative pros

qwen-edit-2509/lora/inscene

Advanced AI editing merges scenes and styles with precise structure control for designers.

dreamina-4-0/text-to-image

Next-gen AI visual tool merging text-driven image creation with precision editing.

Frequently Asked Questions

Can I use Z Image Turbo ControlNet for commercial text-to-image projects on RunComfy?

Yes, Z Image Turbo ControlNet is distributed under the Apache 2.0 open-source license, which generally allows commercial use. However, using it on RunComfy does not override or bypass the model’s original license terms. If you plan to deploy Z Image Turbo ControlNet commercially for text-to-image generation at scale, review the official license from the model creators to ensure proper compliance.

Are there technical limitations when using Z Image Turbo ControlNet for text-to-image generation?

Yes. Z Image Turbo ControlNet currently supports maximum output resolutions of up to about 1536×1536 pixels. The prompt input is limited to approximately 200–250 tokens, and users can apply up to 4 simultaneous reference conditions through the Fun ControlNet Union (Canny, HED, Depth, Pose, or MLSD). These constraints balance quality, speed, and GPU resource efficiency.

What are the main strengths of Z Image Turbo ControlNet compared to earlier text-to-image models?

Z Image Turbo ControlNet stands out for its bilingual English and Chinese comprehension, improved prompt fidelity, and text rendering within generated images. Using a 6-billion-parameter Single-Stream DiT structure, it performs image generation in just 8 inference steps, delivering fast, high-quality text-to-image outcomes while being more VRAM-efficient than larger competitors like SDXL or Flux.

Can I fine-tune or customize Z Image Turbo ControlNet in my text-to-image pipeline?

Yes, Z Image Turbo ControlNet supports the Base and Edit variants for fine-tuning and image editing tasks. Developers can adapt these for domain-specific text-to-image generation while still benefiting from the core DiT efficiency. Note that fine-tuned derivatives must also respect the original Apache 2.0 licensing conditions.

What happens after my free trial of Z Image Turbo ControlNet ends on RunComfy?

After your free trial credits (usd) are used, you’ll need to purchase additional usd to continue generating text-to-image outputs with Z Image Turbo ControlNet. Pricing is listed under the 'Generation' section of your account. You can monitor usage and costs directly within your RunComfy dashboard.

What kind of support is available for Z Image Turbo ControlNet users on RunComfy?

If you encounter issues using Z Image Turbo ControlNet for text-to-image generation, you can reach RunComfy’s support team via hi@runcomfy.com. They assist with API integration, usage limits, and troubleshooting performance or licensing-related questions.

RunComfy

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

Z Image Turbo ControlNet: Photoreal Image-to-Image with Depth & Pose Control | RunComfy