A Critical Daily-Use Bug

[student996007]

When initiating a new chat in a Conversation, if the selected model is sourced from: Model Settings > Create Engine > OpenAI Specification > Other API URL and API Key (e.g., through an API proxy/relay like NewAPI), a significant issue occurs.

Specifically, when using models such as Gemini Pro 2.5 or Claude 4.5 (and potentially others), whether attempting to paste content (Ctrl+V) or upload a PNG file, the system consistently displays the error message: "This file format is not supported by this model, please try another file."

However, these models inherently support vision capabilities (and Gemini, for instance, is even known to support video (.mp4) input). Crucially, these same models function normally with visual input when accessed via "OpenAI compatible" API calls in other AI clients.

( On another note, there's a characteristic that isn't a bug but feels like one due to its very poor user experience: when copying and pasting PNG images into the chat input field using Ctrl+V, ① only one PNG can be pasted at a time, and attempting to paste two or more PNGs simultaneously will result in a complete failure of the paste operation; ② furthermore, after successfully pasting a single PNG, subsequent Ctrl+V attempts to paste more images become unresponsive, as if the function is locked, forcing users to rely solely on the "Add photo and files" option within the "+" menu for any further uploads. )

[nbonamy]

The problem is that on a custom engine, it is not possible to know if the model is a vision model or not. Sending images to a non-vision model can incur large costs in terms of tokens. I guess the solution would be to add an option in the Custom Engine screen to allow sending images and videos but this will be global for the entire engine and not a per model setting