[BUG] Non-vision models fail even when user only sends text - OpenCode auto-reads clipboard images #26668

Description

Bug Description

When using a non-vision model (e.g., DeepSeek V4 Flash), users cannot send plain text messages because OpenCode automatically reads images from the clipboard and tries to send them to the model, even when the user hasn't attached any images.

Steps to Reproduce

1. Configure DeepSeek V4 Flash model (non-vision)
2. Copy an image to clipboard (screenshot, etc.)
3. Type a text-only message in OpenCode (no image attached)
4. Submit the message
5. Error: "Cannot read 'clipboard' (this model does not support image input)."

Expected Behavior

• OpenCode should NOT automatically read clipboard images when user only sends text
• Or provide a config option to disable auto clipboard image reading

Actual Behavior

• Even with no image attached to the message, OpenCode reads clipboard and fails
• Other models (e.g., minimax-m2.5-free) work fine because they might handle this differently

Evidence

Error message shown:
ERROR: Cannot read "clipboard" (this model does not support image input). Inform the user.
Note: The user never attached any image - it was purely a text message.

Environment

• OpenCode: 1.14.46
• OS: macOS
• Model: deepseek-v4/deepseek-v4-flash
• Provider: DeepSeek API

Related Issues

• Feature Request: Automatic Image Vision Processing for Non-Vision Models #21004: Automatic Image Vision Processing for Non-Vision Models
• Clipboard images routed directly to model instead of MCP when model lacks vision support #18437: Clipboard images routed directly to model instead of MCP when model lacks vision support

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

Operating System

No response

Terminal

No response