Z-Image FAQs

What is Z-Image?

Z-Image is a powerful AI model offering photorealistic image generation, accurate rendering of both Chinese and English text, and robust adherence to bilingual instructions. It achieves performance comparable to or exceeding leading competitors with only 8 steps, making it a highly efficient and capable tool for diverse image creation needs.

What makes Z-Image's architecture special?

Z-Image utilizes a Scalable Single-Stream DiT (S3-DiT) architecture. This innovative design unifies various conditional inputs, such as text and image embeddings, with noisy image latents into a single sequence. This single input stream, concatenating text, visual semantic tokens, and image VAE tokens, maximizes parameter efficiency compared to traditional dual-stream approaches.

How fast is Z-Image?

Z-Image offers impressive speed, with sub-second inference latency on enterprise-grade H800 GPUs. On NVIDIA A10 GPUs, most generations are completed within a maximum of 2 seconds using only 9 steps. For consumer-grade GPUs like the RTX 3090/4090, generation typically takes 2-3 seconds, while mid-range cards average 4-5 seconds.

Can Z-Image render bilingual text accurately?

Yes, Z-Image excels at accurately rendering both Chinese and English text. It maintains facial realism and overall aesthetic composition while doing so, demonstrating strong compositional skills and a keen sense of typography. This capability extends even to challenging scenarios involving small font sizes.

What is the Prompt Enhancer (PE)?

The Prompt Enhancer (PE) is a key feature within Z-Image that employs a structured reasoning chain to inject logic and common sense into the image generation process. This enables the model to effectively handle complex tasks, such as solving visual puzzles like the 'chicken-and-rabbit problem' or visualizing abstract concepts like classical Chinese poetry. Furthermore, the PE can infer user intent even from ambiguous instructions, ensuring a logically coherent and relevant output.

How does Z-Image perform against competitors?

According to the Elo-based Human Preference Evaluation conducted on the Alibaba AI Arena, Z-Image demonstrates highly competitive performance when compared to other leading models in the field. Notably, it achieves state-of-the-art results among open-source models, highlighting its superior quality and efficiency within the publicly available AI landscape.

What kind of creative editing can be done with Z-Image?

Z-Image-Edit offers creative image editing capabilities with a strong understanding of bilingual instructions, enabling imaginative and flexible image transformations. Users can seamlessly modify images without external tools, leveraging built-in features for a streamlined editing workflow and high-quality results.

How can users optimize results when using Z-Image?

To achieve the best Z-Image results, users should specify bilingual text requirements clearly, describe lighting, shadows, and textures for photorealistic quality, and utilize the Prompt Enhancer for complex creative tasks. Taking advantage of the fast 8-step generation for rapid iteration and leveraging its compositional skills for poster design also helps.

Z-Image is an AI image editor and generator designed for photorealistic image creation, precise bilingual text rendering, and robust editing, operating with advanced S3-DiT architecture. It prioritizes speed, generating high-quality images in few steps.

• Access the Z-Image platform, identifying the "Text to Image" or "Image Editor" sections for your task.
• Input your descriptive prompt into the designated text field, specifying desired imagery, lighting, and any bilingual text requirements.
• Utilize the integrated Prompt Enhancer (PE) for complex reasoning tasks or to refine ambiguous instructions, ensuring accurate interpretation.
• Initiate the image generation process; Z-Image will produce results in approximately 8 steps, often within 2-5 seconds on consumer GPUs.
• Review the generated image, then apply Z-Image-Edit for further creative transformations or adjustments using natural language instructions.