v1.0.5: README 警告——image_in 仅限 provider 真支持图片的模型，纯文本模型声明会导致 400

README.en.md

@@ -111,6 +111,31 @@ The MCP server starts automatically with the session.
 | Multimodal model | Paste with Alt+V directly — native vision, plugin not involved |
 | Non-multimodal + image path | Type `@screenshot.png` or paste the path; the model calls `read_image` |
 | Non-multimodal + just screenshotted/copied | Do **not** Alt+V paste (the CLI rejects pasting on non-multimodal models with `Current model does not support image input`); just ask "analyze this screenshot" — the model calls `read_clipboard_image` to read the system clipboard |
+| Non-multimodal + paste was rejected | Tell the model "my paste was blocked" — it will switch to `read_clipboard_image` automatically; no need to save the file |
+
+#### Want Alt+V pasting? (declare `image_in` on the model)
+
+Kimi Code's frontend blocks pasting on models that lack image support. Add
+`image_in` to the model in `~/.kimi-code/config.toml`:
+
+```toml
+[models."opencode-go/deepseek-v4-flash"]
+capabilities = [ "thinking", "tool_use", "image_in" ]   # append image_in
+```
+
+> ⚠️ **Important**: `image_in` only lets the **frontend accept** the paste — it does
+> not mean the provider can actually receive images.
+>
+> - Declare it **only when the provider truly supports image input** (e.g.
+>   MiniMax-M3, k3, gpt-5.6-luna, grok-4.5) — those models see natively and the
+>   plugin stays idle
+> - **Never declare it on text-only providers** (e.g. deepseek-v4-flash, GLM text
+>   variants): the paste gets accepted, then the request fails at the provider with
+>   `400 unknown variant image_url, expected text`. The frontend block is a guard.
+>
+> Correct usage for text-only models: `@image-path` (`read_image`) or screenshot and
+> ask directly (`read_clipboard_image`); if a paste is rejected, tell the model
+> "the paste was blocked" and it will read the clipboard instead. **No commands needed.**
 
 ## Activation and triggering
 

README.md

@@ -95,7 +95,12 @@ KimiCode 前端默认会拦截「不支持图片输入」模型的粘贴。在 `
 capabilities = [ "thinking", "tool_use", "image_in" ]   # 追加 image_in
 ```
 
-> 注意：声明 `image_in` 只是让前端放行粘贴。纯文本模型依然**看不懂**图片内容——这正是插件的用武之地：模型收到图片但看不见内容时，会按 SYSTEM.md 规则自动调 `read_clipboard_image` 从剪贴板读图。**全程零命令、零前缀**。
+> ⚠️ **重要警告**：`image_in` 只是让**前端放行**，不代表 provider 真能收图。
+>
+> - **仅当 provider 本身支持图片输入**（如 MiniMax-M3、k3、gpt-5.6-luna、grok-4.5 等）时声明才有意义——这些模型粘贴后原生看图，插件闲置
+> - **纯文本 provider**（如 deepseek-v4-flash、glm 文本版）**不要声明**——声明后粘贴会被放行，但请求发给 provider 时图片 part 会被拒（400 `unknown variant image_url, expected text`）。此时前端拦截反而是保护
+>
+> 纯文本模型的正确用法：`@图片路径`（read_image）或截图后直接提问（read_clipboard_image 读剪贴板）；粘贴被拦截就直接告诉模型「粘贴被拦截了」，它会自动改读剪贴板。**全程零命令**。
 
 ## 激活与触发
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "kimi-eyes",
-  "version": "1.0.4",
+  "version": "1.0.5",
   "description": "Give non-multimodal models in Kimi Code the ability to analyze images and screenshots via a user-configured VLM (OpenAI-compatible or Anthropic protocol).",
   "keywords": [
     "kimi",