Starting 24 hours ago, when using the image generation tool, in most cases, I can only generate images at a fixed resolution of 1376 x 768. Previously, I could request the tool to generate images with any aspect ratio.
Steps to Reproduce
Using the “Generate a night sky image with a 1:1 aspect ratio.” prompt.
Expected Behavior
Returns a night sky image with a 1:1 aspect ratio.
I did a bit more digging and I’d like to follow up with some more information if that would help in the aid of debugging/fixing this issue…
Cursor 2.4 changelog states that “Google Nano Banana Pro” is used for image generation: https://ai.google.dev/gemini-api/docs/image-generation#generate-4k-images
According to Google docs, this is Gemini 3 Pro: https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image-preview
Per Google docs, Gemini images generate 1K resolutions by default: https://ai.google.dev/gemini-api/docs/image-generation#generate-4k-images
Using a 16:9 ratio for 1K resolution images, the output is 1376x768: https://ai.google.dev/gemini-api/docs/image-generation#3-pro-image-preview
To me, this suggests that the image_size value being passed from the Cursor backend –> Gemini is always set to 1K. My guess is also that the 16:9 ratio is either explicitly set, or used as a default by omission. These two combined result in output images always being set to 1376x768.
Hope this is helpful! Thank you
P.S. Apologies for using backticks to format the links! I am considered a “new user” and so can’t actually post the links, but I wanted to include references
@sscally, great research, thanks. Your analysis matches what we’re seeing on our side. The image generation tool really isn’t passing size or aspect ratio parameters to the provider, so the result is the Gemini default (1K, 16:9 = 1376x768).
I shared your analysis with the team, it’ll definitely help with prioritization. No ETA yet, but having a clear root cause makes this much easier to fix.
Hey @Alexander_Kotev, thanks for confirming. This is a known bug and we’re tracking it on our side. We understand the root cause, the backend isn’t passing the size and aspect ratio parameters to the provider, so you get the default Gemini output 1K at 16:9, which is 1376x768. We don’t have an ETA for a fix yet, but we’ll post an update here as soon as we have one.
Hey @Redstone1024,
Just checking in - are you still seeing image generation locked to 1376x768? We’ve shipped a number of updates since this was reported. If a recent update resolved it, feel free to let me know. Otherwise, any additional details would help us dig deeper.
cc: @sscally@Ayaan_Rulhania@Alexander_Kotev
@mohitjain Can confirm I’ve been able to generate images with other dimensions (like 512x512 for example).
However, doing a quick test this morning a simple “generate me a 512x512 image of a cat” produced a 1536x1024 image. That’s not 1376x768 like the original bug, but still wasn’t what I was looking for! (I understand these things are non-deterministic but I figured I’d bring that up)
EDIT: I was wrong earlier. I’m only getting 1536x1024 images. The 512x512 images I was referring to earlier were due to post-processing automation I have set up for my project. So it seems like there are still some bugs.
Thanks for the detailed follow-up (and the edit correcting the 512x512 observation - that’s really helpful to know).
So the resolution shifted from 1376x768 to 1536x1024, but custom dimensions still aren’t being respected. That confirms the core issue is still present on current versions. I’ve re-escalated this with the team.
When generating images in Cursor (via Agent or Composer), explicitly requesting a specific size such as 750x1500px completely ignored. Regardless of how the prompt is phrased, the generated image is always 1536 × 1024 pixels.
This makes it impossible to generate UI assets, icons, thumbnails, or any image that requires precise dimensions.
Impact:
This blocks all workflows that depend on exact image sizes. A quick fix or support for explicit width x heightparameters would be greatly appreciated.
Steps to Reproduce
Open Cursor Agent / Composer
Prompt: Generate a 200x200px solid blue square
Download or inspect the generated image
Expected Behavior
The image should match the requested dimensions (e.g., 750x1500 px).
Actual Behavior:
The image is always generated at 1536 × 1024 px, ignoring the requested size entirely.
Thanks for the clarification! I really hope custom sizes can be supported soon. The main reason I use Cursor’s built-in tool is to avoid switching to external image generators—I’d love to keep my entire workflow within Cursor. Fingers crossed for an update!
Critical issue for me. As long as this is not fixed, I need to have the images generated using Gemini Pro directly. I canceled my Cursor Pro+ subscription because now the costs are generated on Gemini side. And I need to integrate every picture manually which is very cumbersome. Please prioritize this fix.
I also encountered the same issue. The integration between Codex and GPT-Image-2 has greatly improved convenience, not only for coding. I hope Cursor can fix this bug as soon as possible.
Yes, I strongly agree, and this is just a minor issue. Using Cursor itself should have solved it in a very short time, yet this problem has existed for several months. I really can’t imagine what new things you are developing.
Thanks for the continued reports - and for confirming it’s still happening on the latest version with custom sizes ignored (1536×1024 regardless of what’s requested).
This remains a known issue we’re tracking, and the extra confirmations from @light_english, @JBDev, and @Jackyfan genuinely help it get prioritized. The cause is what @deanrie described earlier in the thread: the requested size isn’t being passed through to the image provider, so you end up with a default output. I don’t have a timeline to share yet, but I’ll post here as soon as that changes.
I know that’s frustrating when exact dimensions are central to your workflow, so I won’t pretend there’s a clean workaround — there isn’t a reliable in-product way to force a specific size today. I’ll keep this thread updated.
