Gemma4 throu ollama -> vision capabilities

[Glog78]

Description

I try to make gemma4 analyse screenshots ( for example error identification )

i used a two way check ( same model ) -> here is what vim ai does

... additionally it would be fantastic if a image from the clipboard could be auto attached to the context ...

Plugins

none

OpenCode version

1.15.7

Steps to reproduce

take a local gemma4 model
try to analyse a picture

Screenshot and/or share link

No response

Operating System

Arch

Terminal

ghostty

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [Bug] Images silently discarded with local Ollama vision models + TUI lacks drag-and-drop #26780: Images silently discarded with local Ollama vision models + TUI lacks drag-and-drop (very similar: Ollama + vision model + image not processed)

Additionally, for the clipboard image attachment feature request mentioned at the end of your issue:

• Image Clipboard Paste Not Working in OpenCode #10154: Image Clipboard Paste Not Working in OpenCode (closed)

Please check if those issues already cover your problem. If so, consider adding a comment there with your findings instead.

[Glog78]

found a solution with all this issues -> solution was to add

"modalities": {
"input": ["text", "image"],
"output": ["text"]
}