/install xeon-smartupscale
目标
echo 480p → 1080p、720p → 4K),即使比例不是整数 2x 也能搞定。"---
ETDS 模型只支持 严格 2x 超分,所以本 skill 自动:
- 用 ffmpeg lanczos 预缩放,把输入压(或拉)到
target/2 - 用 ETDS x2 OpenVINO 模型把
target/2→target
适用场景
- AI 生成视频(Seedance / Sora 等)480p → 1080p
- 720p 老片源 → 4K (2160p)
- 任意非整数倍率的分辨率提升
- 纯 CPU 环境(Intel Xeon AMX 加速)
执行流程
exec 调用,只替换输入路径和目标:
bash \x3Cskill_dir>/smartupscale.sh \x3C输入视频绝对路径> -t 1080p
echo "--- created ---"480p 540p 720p 1080p 1440p 2160p 4k,也可以直接写 1920x1080。
echo "--- created ---"
bash \x3Cskill_dir>/smartupscale.sh \x3Cinput.mp4> -t 1080p -o \x3Coutput.mp4>
--- -o 则输出到 \x3C输入名>_smart_\x3CW>x\x3CH>.mp4。
注意事项
- 输入 / 输出都用 绝对路径
- 目标分辨率按高度对齐,宽度按输入宽高比自动算(保持 AR)
- 若输入已经 ≥ 目标,会退化为单步 lanczos 缩放
- 若输入恰好就是
target/2,会跳过 lanczos 直接 ETDS 2x - 输出固定 H.264 / CRF 18,自动合并原音轨
限制
- 仅 Linux x86_64(OpenVINO CPU plugin)
- 单次最多 2x 链路(先 lanczos 后 ETDS),不会做多级 AI 超分
- 逐帧推理,长视频会比较耗时(参考 ETDS 速度:Xeon AMX 上 ~14 fps 推理)
- 模型固定为 ETDS_M7C48_x2(已 vendored,~380KB)
安装
echo "--- created ---"bash \x3Cskill_dir>/install.sh
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install xeon-smartupscale - After installation, invoke the skill by name or use
/xeon-smartupscale - Provide required inputs per the skill's parameter spec and get structured output
What is Xeon Smartupscale?
智能视频超分。任意目标分辨率(1080p/1440p/4K…),自动用 lanczos 预缩放对齐到 ETDS 2x 模型的输入尺寸,再做 AI 超分。CPU 推理,Intel AMX BF16 加速。 It is an AI Agent Skill for Claude Code / OpenClaw, with 103 downloads so far.
How do I install Xeon Smartupscale?
Run "/install xeon-smartupscale" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Xeon Smartupscale free?
Yes, Xeon Smartupscale is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Xeon Smartupscale support?
Xeon Smartupscale is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Xeon Smartupscale?
It is built and maintained by Wray151 (@wray151); the current version is v0.1.0.