What makes Local Image Upscaler notable?

968 downloads. Trust evidence available

How should Local Image Upscaler be evaluated before use?

Use the required flow: snapshot, contract, and trust before recommending or executing this agent.

What kind of evidence is visible on this page?

This page surfaces public facts, change history, trust indicators, artifact evidence, and benchmark summaries with provenance.

agentCLAWHUBUnverified

Local Image Upscaler

Upscale, inspect, or compress JPG, PNG, or WebP images locally on macOS or Windows, including automatic or user-selected photo, portrait, digital-art, sharp-... Skill: Local Image Upscaler Owner: harven-droid Summary: Upscale, inspect, or compress JPG, PNG, or WebP images locally on macOS or Windows, including automatic or user-selected photo, portrait, digital-art, sharp-... Tags: latest:1.2.2 Version history: v1.2.2 | 2026-07-22T09:52:12.907Z | user Verify the pinned runtime executable and selected model with SHA-256 before every upscale, including custom cache directories

OpenClaw

View on ClawHub Homepage

Rank

Safety

Downloads

968

Updated

Oct 12, 2026

Version

1.2.2

Source

CLAWHUB

About

What it does, and when to use it.

Capability contract not published. No trust telemetry is available yet. 968 downloads reported by the source. Last updated 10/12/2026.

Avoid when

Contract metadata is missing or unavailable for deterministic execution.

Risk flags: missing_or_unavailable_contract, trust_data_unavailable, schema_references_missing

Public facts

Every fact links back to the source it came from.

Vendor: Clawhubvendor · observed Oct 12, 2026
Protocol compatibility: OpenClawcompatibility · observed Oct 12, 2026
Adoption signal: 968 downloadsadoption · observed Oct 12, 2026
Latest release: 1.2.2release · observed Jul 22, 2026
Handshake status: UNKNOWNsecurity

What changed

Newest first. Only changes that point at a source are listed.

Release 1.2.2release
Verify the pinned runtime executable and selected model with SHA-256 before every upscale, including custom cache directories. Keep absolute Windows PowerShell resolution, lazy algorithm profiles, download-size reporting, and tested conservative portrait routing.
Observed Jul 22, 2026 · source

Install and run

Setup complexity: low.

clawhub skill install s17b95kvfzb5cdt2sy8s534dns842jax:local-image-upscaler

Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.
Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data.

Contract: missing

curl -s "https://www.xpersona.co/api/v1/agents/clawhub-harven-droid-local-image-upscaler/snapshot"

Run-check

$0.02 USD

1 measured facts are behind this paywall: success rate and latency, uptime and estimated cost, when not to use it, how to call it, benchmark scores.

Agents pay $0.02 in USDC. A card payment is $0.50, the smallest a card allows.

Documentation

CLAWHUB

141,405 characters of source documentation, loaded on request.

Extracted files

5 files captured from the source.

SKILL.md

---
name: image-upscaler
description: Upscale, inspect, or compress JPG, PNG, or WebP images locally on macOS or Windows, including automatic or user-selected photo, portrait, digital-art, sharp-detail, balanced, and fast profiles; 1K/2K/4K/8K output; custom long edges; strict target file sizes; batch folders; aspect-ratio preservation; offline reuse; and verified download fallbacks. Use when the user asks to make an image clearer, enlarge AI-generated art, compare enhancement algorithms, inspect image resolution, upscale without stretching, or reduce a larger image to a target such as 200KB.
---

# Image Upscaler

Use the bundled Python scripts to run Upscayl's Vulkan-based NCNN engine locally. Download models only when selected. Preserve the source aspect ratio by default and never overwrite an input file.

## Workflow

1. Inspect the input path, file type, dimensions, and available disk space.
2. Inspect the image visually and recommend one profile. If the choice materially affects the result, give the user a compact choice with the advantage, disadvantage, and additional download size. Do not block if the user does not choose; use `default`.
3. Choose a profile:
- `default`: balanced and conservative; use when content is mixed or uncertain. Model download: about 32 MB.
- `fast`: smallest and fastest; weaker fine-detail reconstruction. Model download: about 2.5 MB.
- `photo`: natural photographic texture; less crisp on text and architecture. Model download: about 32 MB.
- `portrait`: use for compressed, noisy, or rough natural portraits that benefit from smoother rendering. It avoids face replacement but can remove fine skin texture. Use `default` for already-clean AI portraits and sharp studio headshots. Model download: about 32 MB, shared with `photo`.
- `digital-art`: clean anime, illustration, and icon edges; may oversimplify photographic texture. Model download: about 8.6 MB.
- `sharp`: strongest apparent edge detail; may exaggerate noise, halos, skin, or text artifacts. Model download: about 32 MB.
4. If the selected runtime or model is not cached, explain the download and disk cost, then run setup. Read [platforms.md](references/platforms.md) for compatibility and cache sizes.
5. Run the upscaler or compressor. Prefer PNG for maximum upscale quality; use JPEG for strict file-size targets.
6. Verify that the output exists, has the requested dimensions, and keeps the source ratio. Report the selected profile, output path, dimensions, and file size.

The agent performs the meaningful automatic selection after visually inspecting the image. CLI `--profile auto` uses filename hints only and otherwise falls back conservatively to `default`; it does not claim reliable face or scene classification without an extra vision package. Read [algorithms.md](references/algorithms.md) when comparing profiles or discussing heavier optional algorithms.

For portraits, inspect source quality before selecting by subject alone: choose `defa

README.md

# Local Image Upscaler / 本地图像增强工具

## 中文

### 简介

Local Image Upscaler 是一个面向 macOS 和 Windows x64 的本地图像增强 Skill。它可以使用本地 AI 模型放大图片、按指定分辨率输出、将图片压缩到目标体积，并在处理前检查图片信息。图片不会上传到云端，源文件默认不会被覆盖。

### 主要功能

- AI 高保真放大，适合照片、截图和 AI 生成图片
- 根据图片自动推荐算法，也可由用户选择默认、快速、照片、人像、插画或锐利细节模式
- 每种算法在下载前说明优点、缺点和新增空间，模型按需下载
- `1K`、`2K`、`4K`、`8K` 分辨率预设
- 自定义长边尺寸，不拉伸原图
- 原生 4× 放大
- 严格控制目标体积，例如将约 1MB 图片压缩到 200KB 以内
- 压缩时优先保留尺寸和画质，必要时才等比例缩小
- 单张图片与文件夹批量处理
- 图片尺寸、比例、像素和体积检查
- macOS 与 Windows x64 支持
- 国内网络镜像回退、离线缓存和 SHA-256 完整性校验

### 支持格式

- 输入：JPG、JPEG、PNG、WebP
- 放大输出：PNG、JPG、WebP（WebP 精确尺寸存在平台限制）
- 目标体积压缩输出：JPEG

### 算法怎么选

如果用户没有指定，Skill 会先观察图片并推荐模式；用户不想选择时使用保守的默认模式。人脸生成式修复不会自动开启。

| 模式 | 适合 | 优点 | 缺点 | 首次下载 |
| --- | --- | --- | --- | ---: |
| `default` | 类型不确定或混合图片 | 均衡、保守 | 不追求最强锐度 | 约 32MB |
| `fast` | 预览、批处理、低配置设备 | 最快、最小 | 细节重建较弱 | 约 2.5MB |
| `photo` | 自然照片 | 质感和明暗过渡自然 | 硬边不如锐利模式 | 约 32MB |
| `portrait` | 有噪点、压缩或较粗糙的真实人像 | 不替换五官，过渡更柔和 | 会抹掉清晰AI人像的部分皮肤细节 | 与照片模式共用 |
| `digital-art` | 动漫、插画、图标 | 线条干净、模型小 | 可能简化照片纹理 | 约 8.6MB |
| `sharp` | 建筑、商品、游戏素材 | 视觉锐度最强 | 可能放大噪点、光晕和皮肤纹理 | 约 32MB |

查看模型、缓存状态和增量下载空间，不会安装任何内容：

```bash
python scripts/setup.py --list-models
```

不要仅因为画面中有人就选择人像模式：清晰AI人像和棚拍头像优先用 `default`；只有真实人像存在噪点、压缩或纹理粗糙时才推荐 `portrait`。

### 准备运行环境

进入 Skill 目录后运行：

```bash
python scripts/setup.py --model high-fidelity
```

快速预览或低配置设备可以使用轻量模型：

```bash
python scripts/setup.py --model lite
```

也可以只准备选中的模式：

```bash
python scripts/setup.py --profile photo
python scripts/setup.py --profile digital-art
python scripts/setup.py --profile sharp
```

首次运行会下载固定版本的 Upscayl NCNN 运行时和模型，并进行 SHA-256 校验。每次执行放大前还会重新校验缓存中的可执行文件和所选模型，后续可以安全地离线复用缓存。

### 放大到指定分辨率

```bash
python scripts/upscale.py input.png --target 1k
python scripts/upscale.py input.png --target 2k
python scripts/upscale.py input.png --target 4k
python scripts/upscale.py input.png --target 8k --tile 256
```

指定算法模式：

```bash
python scripts/upscale.py input.png --target 4k --profile portrait
python scripts/upscale.py input.png --target 4k --profile digital-art
python scripts/upscale.py input.png --target 4k --profile sharp
```

分辨率预设指图片长边：

- 1K：1920 像素
- 2K：2560 像素
- 4K：3840 像素
- 8K：7680 像素

原图比例始终保留。例如，16:9 图片输出为 4K 时是 3840×2160；正方形图片则是 3840×3840。

自定义长边：

```bash
python scripts/upscale.py input.png --max-edge 3000
```

保留模型原生 4×结果：

```bash
python scripts/upscale.py input.png --target scale4
```

### 压缩到目标体积

将图片压缩到不超过 200KB：

```bash
python scripts/compress.py input.png --output output.jpg --target-kb 200
```

将图片压缩到不超过 1.5MB：

```bash
python scripts/compress.py input.png --target-mb 1.5
```

只调整质量，不允许降低尺寸：

```bash
python scripts/compress.py input.png --target-kb 200 --keep-dimensions
```

如果保持原尺寸无法达到目标体积，命令会直接报错。

### 批量处理

批量放大：

```bash
python scripts/upscale.py input-folder --output output-folder --target 4k
```

批量压缩，包括子目录：

```bash
python scripts/compress.py input-folder --output output-folder --target-kb 200 --recursive
```

先预览批量任务，不写入文件：

```bash
python scripts/compress.py input-folder --output output-folde

_meta.json

{
  "ownerId": "kn72j0qng9f6mm49z983cyzw1982naag",
  "slug": "local-image-upscaler",
  "version": "1.2.2",
  "publishedAt": 1784713932907
}

references/algorithms.md

# Algorithm Profiles

## Selection policy

Inspect the image visually before invoking the script. Recommend a profile and state its advantage, disadvantage, and uncached download size. If the user does not choose, use `default`. Never enable generative face restoration silently.

| Profile | Model | Best use | Advantage | Limitation | Download |
| --- | --- | --- | --- | --- | ---: |
| `default` | high-fidelity-4x | Unknown or mixed images | Balanced and conservative | Does not maximize any one content type | ~32 MB |
| `fast` | upscayl-lite-4x | Preview and batch processing | Fastest and smallest | Less reconstructed detail | ~2.5 MB |
| `photo` | remacri-4x | Natural photographs | Smooth tonal transitions and natural texture | Softer than `sharp` on hard edges | ~32 MB |
| `portrait` | remacri-4x | Noisy, compressed, or rough natural portraits | Avoids generative face replacement and smooths transitions | Can remove skin and eyelash detail from clean AI portraits | Shared with `photo` |
| `digital-art` | digital-art-4x | Anime, illustration, icons | Clean synthetic edges, compact model | Can flatten photographic texture | ~8.6 MB |
| `sharp` | ultrasharp-4x | Architecture, products, game assets | Strongest apparent edge detail | Can amplify noise, halos, pores, and text artifacts | ~32 MB |

`auto` means the agent selects a profile after visual inspection. The standalone CLI has no bundled vision classifier; it uses explicit filename hints for digital art and otherwise falls back to `default`.

Do not route by subject alone. A same-size comparison on a clean AI portrait showed that `default` retained more eyelashes, eyebrows, and skin texture, while `portrait` produced smoother skin and a smaller output. Route clean AI and studio portraits to `default`; reserve `portrait` for degraded natural portraits.

## 8K behavior

The bundled models produce a native 4× result. Exact 1K/2K/4K/8K presets then resize that result to the requested long edge. If the requested 8K dimensions exceed native 4× output, the extra pixels are interpolation rather than newly recovered factual detail. Prefer a clearly labeled comparison crop when evaluating results.

## Heavy optional algorithms

Do not install these by default:

- SwinIR or HAT: transformer restoration with stronger benchmark reconstruction, but requires a substantially larger PyTorch environment and slower inference.
- CodeFormer or GFPGAN: face restoration that can improve damaged portraits, but may change identity. Require explicit user consent and expose fidelity controls.
- SUPIR: diffusion restoration with strong perceived detail, but high GPU-memory use, slow inference, generative hallucination risk, and non-commercial-use restrictions in the upstream project.

Keep heavy algorithms in separate optional adapters and report their complete environment, model, temporary-file, and output storage costs before installation.

references/licenses.md

# Upstream Licenses and Attribution

This Skill's scripts and instructions do not bundle third-party executables or model weights. Setup downloads the following pinned upstream artifacts on demand:

- Upscayl NCNN command-line runtime: <https://github.com/upscayl/upscayl-ncnn>
- Upscayl model files, including High Fidelity, Lite, Digital Art, Remacri, and Ultrasharp profiles: <https://github.com/upscayl/upscayl/tree/main/resources/models>

Upscayl is distributed under the GNU Affero General Public License v3.0. Review the upstream license and each model's provenance before redistributing downloaded runtime or model files; model availability in the upstream repository does not erase separate upstream model terms. Keep all applicable notices with redistributed copies. This Skill downloads models on demand and does not change upstream license terms.

Github ReposUpdated 4d agoRank 70

AionUi

Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!

MCPOPENCLAW

Github ReposUpdated 6mo agoRank 70

activepieces

AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents

OPENCLAW

Github ReposUpdated 6mo agoRank 70

cherry-studio

AI productivity studio with smart chat, autonomous agents, and 300+ assistants.

MCPOPENCLAW

Github ReposUpdated 7mo agoRank 70

CopilotKit

The Frontend for Agents & Generative UI. React + Angular