Welcome to the official repository for the Z-Image（造相）project!
✨ Z-Image
Z-Image is a powerful and highly efficient image generation model with 6B parameters. It is currently has three variants:
🚀 Z-Image-Turbo – A distilled version of Z-Image that matches or exceeds leading competitors with only 8 NFEs (Number of Function Evaluations). It offers ⚡️sub-second inference latency⚡️ on enterprise-grade H800 GPUs and fits comfortably within 16G VRAM consumer devices. It excels in photorealistic image generation, bilingual text rendering (English & Chinese), and robust instruction adherence.
🧱 Z-Image-Base – The non-distilled foundation model. By releasing this checkpoint, we aim to unlock the full potential for community-driven fine-tuning and custom development.
✍️ Z-Image-Edit – A variant fine-tuned on Z-Image specifically for image editing tasks. It supports creative image-to-image generation with impressive instruction-following capabilities, allowing for precise edits based on natural language prompts.
Showcase of Z-Image on Photo-realistic image Generation
📖 Accurate Bilingual Text Rendering: Z-Image-Turbo excels at accurately rendering complex Chinese and English text.
Showcase of Z-Image on Bilingual Text Rendering
💡 Prompt Enhancing & Reasoning: Prompt Enhancer empowers the model with reasoning capabilities, enabling it to transcend surface-level descriptions and tap into underlying world knowledge.
reasoning.jpg
🧠 Creative Image Editing: Z-Image-Edit shows a strong understanding of bilingual editing instructions, enabling imaginative and flexible image transformations.
Showcase of Z-Image-Edit on Image Editing
🏗️ Model Architecture
We adopt a Scalable Single-Stream DiT (S3-DiT) architecture. In this setup, text, visual semantic tokens, and image VAE tokens are concatenated at the sequence level to serve as a unified input stream, maximizing parameter efficiency compared to dual-stream approaches.
Architecture of Z-Image and Z-Image-Edit
📈 Performance
According to the Elo-based Human Preference Evaluation (on AI Arena), Z-Image-Turbo shows highly competitive performance against other leading models, while achieving state-of-the-art results among open-source models.
1import torch
2from diffusers import ZImagePipeline,34# 1. Load the pipeline5# Use bfloat16 for optimal performance on supported GPUs6pipe = ZImagePipeline.from_pretrained(7"Tongyi-MAI/Z-Image-Turbo",8 torch_dtype=torch.bfloat16,9 low_cpu_mem_usage=False,10)11pipe.to("cuda")1213# [Optional] Attention Backend14# Diffusers uses SDPA by default. Switch to Flash Attention for better efficiency if supported:15# pipe.transformer.set_attention_backend("flash") # Enable Flash-Attention-216# pipe.transformer.set_attention_backend("_flash_3") # Enable Flash-Attention-31718# [Optional] Model Compilation19# Compiling the DiT model accelerates inference, but the first run will take longer to compile.20# pipe.transformer.compile()2122# [Optional] CPU Offloading23# Enable CPU offloading for memory-constrained devices.24# pipe.enable_model_cpu_offload()2526prompt ="Young Chinese woman in red Hanfu, intricate embroidery. Impeccable makeup, red floral forehead pattern. Elaborate high bun, golden phoenix headdress, red flowers, beads. Holds round folding fan with lady, trees, bird. Neon lightning-bolt lamp (⚡️), bright yellow glow, above extended left palm. Soft-lit outdoor night background, silhouetted tiered pagoda (西安大雁塔), blurred colorful distant lights."2728# 2. Generate Image29image = pipe(30 prompt=prompt,31 height=1024,32 width=1024,33 num_inference_steps=9,# This actually results in 8 DiT forwards34 guidance_scale=0.0,# Guidance should be 0 for the Turbo models35 generator=torch.Generator("cuda").manual_seed(42),36).images[0]3738image.save("example.png")
🔬 Decoupled-DMD: The Acceleration Magic Behind Z-Image
Decoupled-DMD is the core few-step distillation algorithm that empowers the 8-step Z-Image model.
Our core insight in Decoupled-DMD is that the success of existing DMD (Distributaion Matching Distillation) methods is the result of two independent, collaborating mechanisms:
CFG Augmentation (CA): The primary engine 🚀 driving the distillation process, a factor largely overlooked in previous work.
Distribution Matching (DM): Acts more as a regularizer ⚖️, ensuring the stability and quality of the generated output.
By recognizing and decoupling these two mechanisms, we were able to study and optimize them in isolation. This ultimately motivated us to develop an improved distillation process that significantly enhances the performance of few-step generation.
Building upon the strong foundation of Decoupled-DMD, our 8-step Z-Image model has already demonstrated exceptional capabilities. To achieve further improvements in terms of semantic alignment, aesthetic quality, and structural coherence—while producing images with richer high-frequency details—we present DMDR.
Our core insight behind DMDR is that Reinforcement Learning (RL) and Distribution Matching Distillation (DMD) can be synergistically integrated during the post-training of few-step models. We demonstrate that:
