Copilot Can't Read Images

[IgorWnek]

Select Topic Area

Question

Copilot Feature Area

VS Code

Body

I'd like to ask why Copilot in VS Code cannot read image files in any other way than using the Add Context (cmd + /) and then Image From Clipboard?

It's not possible to read files other then text. I must admit - it's disappointing. In my flow I wanted to use Playwright MCP to have instant feedback about the frontend being produced by Copilot, so it would be able to see a screen shot and apply fixes right away.
That's possible using other agents, but unfortunately not here. Using Image From Clipboard is not an option because simply it's too slow.

To me it's really disappointing. We have multimodal LLMs here, like Claude Sonnet 4, but we can't use their full capabilities.

I treated this "idea" as a breakthrough for my work to make things a lot faster but I can't.

So here comes the question - why can't Copilot read images like any other files in the project?

[FireFistisDead]

@IgorWnek Right now, Copilot in VS Code is limited to reading text-based files from the workspace. It doesn’t treat image files as first-class context, which is why the only way to add them is through Add Context → Image from Clipboard.

The reasons are mainly:

1. Technical limitations – The Copilot VS Code extension and backend were initially designed for text/code completion. Extending that pipeline to handle binary image assets directly from the filesystem would require additional infrastructure and APIs.

2. Privacy & security – Automatically allowing Copilot to send binary assets (like screenshots, UI images, or private design files) from a workspace to the cloud raises stricter compliance and security concerns compared to text/code. The clipboard-based approach ensures the user explicitly controls what image gets shared.

3. Roadmap prioritization – While multimodal models (like GPT-4o, Claude Sonnet, etc.) can process images, Copilot’s current integration in VS Code hasn’t caught up yet. GitHub/Microsoft have prioritized code-focused use cases first, so image context is still experimental and constrained.

---

In short:

Copilot can’t read image files in the project directly because its integration today is designed for textual contexts only, and image support hasn’t been generalized beyond the clipboard due to security, workflow, and infrastructure considerations.

That said, your feedback is valid — many developers working on frontend, design, and testing workflows would benefit from being able to reference project images directly. It’s likely to come down to GitHub expanding Copilot’s multimodal features in future releases.

maybe this will clear your doubt and have a nice day😄

[trusktr]

We need this so Copilot can look at snapshot images and understand what to fix or change visually.

[a-man-called-q]

in VS Code it can read image.. but in intellij it goes filtered..

[Anthonioxz]

https://github.com/xdevplatform/twitter-api-typescript-sdk/actions

[Pedroj2112-dev]

Currently, Copilot in VS Code only supports adding images via Add Context → Image from Clipboard. There’s no way to let Copilot directly access image files from the project itself.
This limitation makes it difficult to use workflows where screenshots or other visual assets are essential. For example, I was hoping to integrate Playwright MCP so that Copilot could review screenshots of the frontend and immediately suggest fixes. Other multimodal agents already allow this kind of flow, but with Copilot it isn’t possible. The clipboard approach is too slow and not practical for fast iteration.

[FireFistisDead]

@IgorWnek you can just say co pilot once that you can access this folder for images and apply fixes acording to it and keep this flow for whole project this kind of promp can be a quick solution for this

[IgorWnek]

Sorry, I think I wrote that you can't do it. And this is the whole reason of this post.

[FireFistisDead]

@IgorWnek can you tell me which model are you currently using?

[IgorWnek]

@FireFistisDead I'm really advanced when it comes to this stuff so, as I've pointed in my question, I'm aware of the fact that we can use Image From Clipboard feature. But that's not the solution to my use-case.

I stated in my post that I've been using Sonnet 4, which has multimodal capabilities. It's just GitHub's backend or VS Code internal API which doesn't allow images to be passed to a model.

As far as I know, there's no solution for that. But I wanted to raise this topic because this is more and more popular to use Playwright MCP as part of the flow, so that agent is being able to quickly validate what's been done. And as @Pedroj2112-dev said, and I said it too in my post, other agents allow this kind of interaction with multimodal-capable LLMs.

So the real question here is if anyone from the GH/VSC crew could say whether this is going to be shipped to the Copilot at all?

[IgorWnek]

Yup, correctly. These are all the things we know already. Thanks for sharing @raunak0400. 👍

However, it doesn't answer the question:

So the real question here is if anyone from the GH/VSC crew could say whether this is going to be shipped to the Copilot at all?

[IgorWnek]

I am aware of all things you just mentioned. Thanks for your response. 👍

[IgorWnek]

I clearly stated that it doesn't answer the question. Thanks

[IgorWnek]

Sorry, what's the connection of that with the topic?

[justin-titanium]

Well, this used to work. I was always pasting images. Just stopped working one day.