Possible Regression in Cursor Image Editing / Img2Img Pipeline

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Over the past several days, Cursor’s image editing behaviour appears to have changed significantly, particularly when using multiple image references and requesting localised edits while preserving composition.

Previously, the system handled:

composition preservation

environment-only modifications

subtle additions to existing artwork

reference-guided edits

…with relatively high fidelity.

Current behaviour now frequently results in:

repeated 400 request failures

ignored composition/layout

full scene regeneration instead of controlled edits

loss of original character positioning

unrelated image outputs loosely derived from prompt keywords

Observed Failure Pattern

The issue appears most reproducible when:

using two reference images

requesting preservation of the original composition

asking for environmental additions only

specifying “maintain characters intact”

performing subtle img2img edits rather than full generations

Example prompt:

“Using the first image following the same art style guide can you add alien creatures referenced in the second image to the surrounding jungle environment in the first image hiding in cover with subtle hints to their presence? Maintain the first image created with the characters intact, only add additional creatures to the surrounding environment.”

Suspected Behaviour

The pipeline appears to:

attempt image editing

fail repeatedly (400)

abandon img2img preservation

fall back to full image regeneration

The resulting image often:

changes composition entirely

relocates characters

alters scene structure

ignores the uploaded base image

This behaviour feels substantially different from previous Cursor builds where composition retention was far more reliable.

Additional Notes

This issue appears across multiple models, not only Gemini 3.1 Pro.

The degradation specifically affects workflows involving:

concept art

sketch refinement

composition-preserving edits

iterative environment detailing

At present the workflow reliability has dropped enough that alternative platforms are becoming more viable for image editing tasks.

Would appreciate clarification on whether:

image editing behaviour changed recently

moderation/routing systems were updated

img2img preservation was altered internally

reference-image handling is currently degraded

Steps to Reproduce

Open Cursor image generation/editing.
Upload a base artwork image containing an established composition and character placement.
Upload a second reference image containing creature/environment reference material.
Use a prompt requesting:
preservation of the original image composition
preservation of existing characters
subtle environmental additions only
matching of the original art style
Example prompt used:

“Using the first image following the same art style guide can you add alien creatures referenced in the second image to the surrounding jungle environment in the first image hiding in cover with subtle hints to their presence? Maintain the first image created with the characters intact, only add additional creatures to the surrounding environment.”

Generate the image.
Observe:
repeated 400 request failures
eventual fallback generation
composition/layout changes
poor preservation of uploaded source image
full image regeneration instead of localised editing

Expected Behavior

The system should perform a controlled image edit using the uploaded base image as the primary composition reference.

Expected output behaviour:

preserve the original image composition and camera framing
preserve existing character placement and anatomy
maintain the original art style
apply only the requested environmental modifications
add subtle creature presence to surrounding jungle areas without restructuring the scene

The generation should behave as an image-editing/img2img operation rather than a full scene regeneration.

The system should also complete the request without repeated 400 failures or fallback behaviour that ignores the uploaded source image.

Operating System

MacOS

Version Information

Version: 3.2.16 (Universal)
VSCode Version: 1.105.1
Commit: 3e548838cf824b70851dd3ef27d0c6aae371b3f0
Date: 2026-04-28T21:07:47.682Z (4 days ago)
Layout: glass
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin x64 24.6.0

For AI issues: which model did you use?

Gemini 3.1 Pro

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Hey, thanks for the detailed report. There’s an important detail about how image generation works in Cursor, and it explains almost everything you’re seeing.

Image generation in Cursor is text-to-image, not img2img or an image editor. The reference images you attach are sent to the model as style and content guidance, but they’re not treated as a base image that gets a composition-preserving edit on top. So:

  • Composition, character placement, and framing aren’t preserved at a pixel level. A new image is generated each time.
  • “Maintain characters intact, only add creatures to environment” is localized edit or inpainting behavior, and it isn’t supported right now. This isn’t a regression, that pipeline was never there.
  • If it looked like the composition stayed the same before, it was most likely just lucky generations with a similar prompt and seed.

About the repeated 400 errors, some of them are refusals from the upstream provider due to prompt content filtering. Prompts with alien creatures, hidden beings, weapons, and similar themes can trigger this inconsistently from request to request.

What you can try for better stability:

  • Describe the final scene in one go, instead of “edit the first image, add…”. For example: “jungle scene in the style of […], with characters X and Y in poses …, alien creatures hidden in foliage”.
  • Rephrase parts that might read as threats or violence in a more neutral way.
  • For real composition-preserving editing, external img2img tools are still the better option for now, then you can use the result in Cursor.

Also, this looks like the same case you opened here: Image editing regression + repeated 400 errors + inconsistent content filtering five days ago. Let’s continue in that thread, and I’ll close this one as a duplicate.