WeReCooking/Z-Anime-CPU

Z-Anime Image Generation (CPU)

CPU Inference - Generate anime images with Z-Anime 6B via stable-diffusion.cpp GGUF. ~30 min at 512x512 on free CPU.

Features

• —Image Generation - text to image via GGUF quantized Z-Anime 6B (S3-DiT)
• —Distill 4-Step - fast inference with distilled model (4 steps, CFG 1.0)
• —CLI Support - command-line interface for local generation
• —MCP Server - AI assistant integration (Claude, Cursor, VS Code)

Image Generation

1. 1.Enter a prompt describing the anime image
2. 2.Adjust resolution (512x512 recommended on CPU)
3. 3.Adjust CFG (1.0 best, max 1.5)
4. 4.Click Generate

Timing: ~30 min at 512x512 on free CPU. GPU is much faster (~2 min).

Models

Settings

• —Steps: 4 (distilled)
• —CFG: 1.0 (max 1.5, do not exceed)
• —Sampler: euler_ancestral
• —Resolution: 512x512 (recommended on CPU)

API

Python Client

from gradio_client import Client

client = Client("WeReCooking/Z-Anime-CPU")
result = client.predict(
    prompt="anime girl with silver hair, fantasy armor, dramatic lighting",
    negative_prompt="worst quality, low quality, blurry, bad anatomy",
    resolution="512x512",
    cfg=1.0,
    seed=-1,
    api_name="/generate"
)
print(result)  # (image, status_message)

MCP (Model Context Protocol)

This Space supports MCP for AI assistants (Claude Desktop, Cursor, VS Code).

MCP Config:

{
  "mcpServers": {
    "z-anime": {"url": "https://werecooking-z-anime-cpu.hf.space/gradio_api/mcp/"}
  }
}

CLI Usage

# Basic generation
python app.py "anime girl with silver hair, fantasy armor" -o output.png

# With seed and CFG
python app.py "1girl, maid outfit, cafe background" --seed 42 --cfg 1.0 -o maid.png

# Custom resolution and negative prompt
python app.py "landscape, cherry blossoms, shrine" -r 768x512 -n "blurry, low quality" -o landscape.png

# Start Gradio UI instead
python app.py --gradio

Credits

• —Z-Anime by SeeSee21
• —Z-Image by Alibaba Tongyi Lab
• —stable-diffusion.cpp by leejet
• —Qwen3-4B Text Encoder GGUF by worstplayer