Issue uploading image using gemma4:26B in Opencode TUI

Description

Trying to upload an image file (.jpg) but the response was the path given instead of the image.
The following is a copy/paste from the chat session

====================================
It appears the system is passing me the file path as text rather than uploading the actual image data. This often happens if there is a communication error between the upload interface and the model's input processing.
Since the path is on a mounted volume (/Volumes/Photo/...), the upload process might be failing to access or read the file from that external drive. Here are a few things to try:
1 Move the file locally: Copy the image from the external drive to your local Downloads or Desktop folder, then try uploading it from there.
2 Test with a different image: Try uploading a different .jpg from a different location to see if the issue is specific to that file or drive.
3 Check file integrity: Ensure the file isn't corrupted and can be opened normally by your computer's photo viewer.
4 Refresh the session: Sometimes a quick refresh of the browser or application can reset the upload mechanism.

@@@@ Here we tried suggeestion #1

Since the error remains the same even with a different file and volume, it confirms the issue isn't the specific photo or the drive.
The system is still only providing me with the file path as a string of text rather than the image data itself. This suggests a failure in the upload pipeline where the image is not being correctly processed and attached to the message before being sent to me.

@@@@

The failure of suggestion #1 confirms that the issue is not related to the file's location or the external drive.
The error message is now explicitly stating: "this model does not support image input". This indicates a fundamental technical mismatch: even though I am instructed that I have image capabilities, the specific environment or API instance you are currently using to interact with me appears to be restricted to text-only.
It's likely that when you "upload" a file, the system is attempting to pass it through a text-reading pipeline (like the read tool) which fails because it cannot process image bytes, and then returns that error message.

@@@@ Here I restarted Opencode

Since restarting did not work, it confirms this is a system-level issue where the interface is attempting to pass images as text files rather than multimodal data. This is a technical bug in how Opencode handles file uploads for this model.
Since I cannot fix this integration myself, please report this issue to the developers at https://github.com/anomalyco/opencode/issues so they can address it.

@@@@ Here I had asked about using a different mode and if it would make any difference

That really depends on whether the issue is a limitation of the model or a bug in the Opencode interface.
There are two ways to look at this:

1. If it is an Opencode integration issue (Most Likely)
   If the "upload" button in Opencode is programmed to always treat files as text-based tasks (meaning it tries to use a read tool to grab the file path rather than attaching the actual image data), then yes, it will likely affect all models.
   In this scenario, even if you switch to a model that is exceptionally good at vision (like a Qwen-VL variant), Opencode might still be trying to "read" the image as if it were a text document, causing the same failure.
2. If it is a model-specific integration issue
   If the code that handles the communication is working fine for some models but is specifically misconfigured for Gemma 4 (e.g., it's not enabling the "vision" flag for this specific model), then a different model might work.
   To test this, you should check two things:
   • Verify Model Capability: Ensure that the Gemma 3 or Qwen models you have are actually the vision-capable versions (e.g., Qwen-VL). If you switch to a standard text-only model, you will definitely get the same error.
   • Test with a Different Model: If you have access to a model like Claude 3.5 Sonnet or GPT-4o through the tool, try uploading there. If those work, the issue is specific to the Gemma 4 configuration in Opencode.
   Summary: If the "upload" action is fundamentally broken in the app, switching models won't help. If the app's "conversation" with Gemma 4 is what's broken, a different model might solve it.

=====================================================================
That ended the chat with its suggestions/summary/conclusion.
I hope I have provided sufficient information.
Running on Mac Studio Ultra, Opencode and Ollama

Plugins

No response

OpenCode version

1.18.34

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

Sequoia 15.7.9

Terminal

Mac built in Terminal to start Ollama and the TUI from Opencode desktop for the chat