Image Compress

v0.3.2其他

opencode-image-compress

Automatically compresses images before sending to AI providers to avoid size limits

半年未更新

#anthropic #compress #compression #image #images #opencode

GitHub 仓库 npm 页面

GitHub 星标

月装机量

100

近 7 天 11

综合评分SCORE

24.1

生态多维模型

快速安装与配置

opencode.json

写入当前项目的 opencode.json，只对这个仓库生效。

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-image-compress@0.3.2"]
}

写入 ~/.config/opencode/opencode.json，对所有项目生效。

~/.config/opencode/opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-image-compress@0.3.2"]
}

opencode 启动时会通过内嵌运行时自动加载 npm 依赖并缓存至本地目录，无需手动在全局环境执行安装。

README.md

查看原始文件→

OpenCode plugin that automatically compresses images before sending them to AI providers, preventing "image exceeds maximum" errors.

Problem

High-resolution images (especially @2x Retina screenshots) often exceed provider size limits, making it impossible to use them in conversations:

Without the plugin: "image exceeds 5 MB maximum" error when attaching a large image

Solution

With the plugin installed, images are automatically compressed before being sent. The same image works seamlessly:

With the plugin: the same large image is compressed and described by the AI without errors

Installation

Add to your opencode.json (global or per-project):

{
	"plugin": ["opencode-image-compress@latest"]
}

That's it. OpenCode will automatically install the plugin on next launch.

How It Works

The plugin compresses images in two scenarios:

Images pasted in chat - via the experimental.chat.messages.transform hook
Images read by the agent - via the tool.execute.after hook when the Read tool returns image attachments

In both cases, it detects the provider and model, and compresses images that exceed provider-specific limits.

Provider Limits

Verified as of January 2026. Source URLs are in src/providers.ts.

Provider	Limit	Target (70%)	Source
Anthropic	5 MB	3.5 MB	docs
Amazon Bedrock	3.75 MB	2.6 MB	docs
OpenAI	20 MB	14 MB	docs
Azure OpenAI	20 MB	14 MB	docs
Google (AI Studio)	100 MB	70 MB	blog
Google Vertex AI	7 MB	4.9 MB	docs
Google Vertex Anthropic	5 MB	3.5 MB	(follows Anthropic limits)
Groq	4 MB	2.8 MB	docs
Fireworks AI	10 MB	7 MB	docs
Perplexity	50 MB	35 MB	docs
xAI (Grok)	20 MB	14 MB	docs
DeepSeek	10 MB	7 MB	(community consensus)
Mistral	10 MB	7 MB	docs
Together AI	20 MB	14 MB	(not documented)
Other/Unknown	5 MB	3.5 MB	(conservative default)

Proxy Providers

Providers like github-copilot, opencode, github-models, and openrouter proxy to upstream APIs. The plugin resolves the actual provider from the model ID (e.g., claude-sonnet-4-5 → Anthropic → 5 MB, gpt-4o → OpenAI → 20 MB) to apply the correct limit.

Compression Strategy

Skip if already under target (70% of provider limit)
Scale down if dimensions exceed 2048px
Reduce quality progressively (JPEG/WebP/AVIF)
PNG: increase compression level, then shrink dimensions
Fallback: aggressive resize to 1024px

Supported Formats

JPEG, PNG, WebP, AVIF, GIF (converted to PNG), other formats (converted to JPEG).

Logging

The plugin logs compression activity to OpenCode's app log system (service: "image-compress"). View logs in the TUI or via the API.

Toast Notifications

Set OPENCODE_IMAGE_COMPRESS_PLUGIN_SHOW_TOAST=1 to show TUI toast notifications when images are compressed or when compression fails.

Toast notification showing "Compressed 2 images: 47.54 MB → 3.58 MB (92% smaller)"

Development

git clone https://github.com/kynnyhsap/opencode-image-compress.git
cd opencode-image-compress
bun install

# Run OpenCode with the plugin loaded from source
bun dev

# Run all checks (format, lint, typecheck, unit tests)
bun check

# Individual commands
bun fmt
bun lint
bun typecheck
bun test:unit
E2E_TESTS=1 bun test:e2e

License

MIT