Desktop AI Engine

Deep Learning Hardware-Level 功能

Upscayl does not guess pixels. It reconstructs textures using trained AI on your GPU — explore the workflow and the technical core behind our open source image engine.

01

简单四步工作流

拖入 PNG、JPG、JPEG 或 WebP,按侧边栏引导操作。选模型、设输出文件夹,一键放大。批量模式可排队整个文件夹。

  • 拖放或即时选择图像
  • Batch Upscayl 文件夹处理开关
  • Double Upscayl 处理难源选项
Desktop workspace screenshot highlighting the four-step import and upscale workflow
02

前后对比

预览 results with a built-in slider before you save. Watch a low resolution source transform into a sharp upscale, from 300×168 to 1200×672 in this example, with model control on the left panel.

  • 预览窗格中的交互式对比滑块
  • General Photo 与专用模型下拉菜单
  • 运行前显示目标分辨率
Interactive before-and-after preview pane with draggable comparison divider
03

本地 GPU 处理

一切在硬件上通过 Vulkan 加速运行。处理叠加层在离线 AI 重建时提供信息,可随时停止。

  • 实时进度与清晰状态消息
  • 模型安装后 100% 离线
  • 无上传、无云端队列延迟
Progress overlay while Vulkan-accelerated enhancement runs entirely offline
Under the Hood

Neural Core & Pipeline

Specialized models, native GPU paths, and production-grade batch tooling — all running locally on your machine.

高级模型架构

在专为不同任务设计的深度神经网络间切换。低多边形游戏素材或旧家庭照片,Upscayl 动态匹配数学模型。

Real-ESRGANRemacriUltraSharpDigital_Art

Vulkan API 本地加速

No remote latency or subscription queues. Upscayl utilizes the local Vulkan API to communicate directly with your AMD, Nvidia, or Intel GPU, maximizing throughput securely.

100% 离线硬件原生无云费用

企业级批量导出

Perfect for texture packs, archival scanning, or comic workflows. Point at any directory, select your scale factor, and let the background queue process hundreds of files concurrently.

文件夹输入自动命名队列并行

灵活输出格式

Export into high-fidelity wrappers without intermediate compression loss. Save and compare using PNG, JPG, or modern WebP depending on your requirements.

无损 PNGWebP 就绪自定义缩放因子
免费下载