Image Data Lost When Pasting from Clipboard in Roo-Code #4750
Description
App Version
v3.20.3
API Provider
Google Gemini
Model Used
gemini-2.5-pro-preview-06-05
🔁 Steps to Reproduce
Hi Team,
I've encountered a critical issue with how Roo-Code and KILO Code handle images pasted from the clipboard. It appears that when a user pastes an image, the image data is not being correctly packaged and sent to the LLM.
Description:
For comparison, this functionality works as expected in other clients like Cline and Cherry Studio. When I paste an image from my clipboard in those applications, the LLM successfully receives and describes the image content. However, performing the same action in Roo-Code or KILO Code results in the LLM behaving as if it only received an empty message, completely unaware that an image was part of the prompt. LLM misleads users and provides incorrect descriptions upon receiving a message
Steps to Reproduce:
- Copy any image to the system clipboard.
- Paste the image into the chat input box in Roo-Code or KILO Code.
- Send the message (requesting the LLM to describe the image).
- Observe the LLM's response (it will acknowledge receiving an image but provide an incorrect description).
Verification:
To confirm my suspicion, I inspected the network request payload sent by Roo-Code and found that it contained no image-related data. This confirms that the image is being lost on the client side before being transmitted.
Related Issues:
I believe this issue is strongly related to, and likely shares the same root cause as, the problems discussed in the following links:
- feat: add critical image analysis rule to enhance accuracy #4722
- Agent hallucinates image content #4697
This is a severe bug as it breaks a core multi-modal feature and makes image-based interaction completely non-functional. I kindly request that this issue be given high priority.
Thank you for your time and attention to this matter.
💥 Outcome Summary
Image Data Lost in Roo-Code