Vulnerability Database

362,557

Total vulnerabilities in the database

vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations — vllm

Interpretation Conflict

Summary

Issue 1: EXIF orientation not normalized → The image orientation processed by the model differs from how humans view it, introducing interpretation bias.

Issue 2: PNG tRNS not explicitly flattened before converting to RGB → After conversion, transparent/semi-transparent pixels are rendered unexpectedly, making otherwise subtle overlay elements visible and distorting the input content. (This attack is similar to AlphaDog: RGBA handling is already correct in vLLM, but since tRNS permits RGB images, the correct processing path isn’t taken.)

Issue 3 : Pillow only loads the first frame when loading APNG or GIF files.

Root Cause

  • Rotation: After opening an image, ImageOps.exif_transpose is not called to normalize EXIF orientation.
  • Transparency: Only RGBA→RGB is flattened with a background; PNGs carrying tRNS in P/L/RGB + tRNS and other non-RGBA modes take the image.convert("RGB") path, which implicitly discards/remaps transparency semantics.

Affected Code

https://github.com/vllm-project/vllm/blob/16b37f3119918c1e5a39f303e0d0892c65c07a90/vllm/multimodal/image.py#L77-L84

https://github.com/vllm-project/vllm/blob/16b37f3119918c1e5a39f303e0d0892c65c07a90/vllm/multimodal/image.py#L37-L43

https://github.com/vllm-project/vllm/blob/16b37f3119918c1e5a39f303e0d0892c65c07a90/vllm/multimodal/image.py#L26-L34 > Current state: ImageOps.exif_transpose is not used. (Although the rescale_image_size function (https://github.com/vllm-project/vllm/blob/main/vllm/multimodal/image.py#L14) exists and includes a transpose parameter, I’ve found that it doesn’t seem to be called anywhere outside the test directory.）

> Call order: _convert_image_mode runs first; if the conditions are met, convert_image_mode is called. > > Issue: Only the “RGBA → RGB” path is explicitly flattened. P, L, or RGB with tRNS all fall back to image.convert("RGB"). For PNGs that include tRNS, convert("RGB") directly produces 24-bit RGB, leading to: > > * P mode: The transparent index becomes an actual RGB color (often black, white, or an undefined background), so transparency is lost. > * L/LA and RGB + tRNS: convert("RGB") doesn’t composite against a chosen background first, so elements that relied on transparency to be hidden or softened become solid.

Impact & Scope

  • Impact: Pixels the model sees can diverge from operator expectations (due to orientation or transparency handling), potentially altering downstream reasoning.
  • Scope: The image I/O and mode-conversion paths in vllm/multimodal/image.py. The existing RGBA→RGB flattening is correct; the issues center on missing EXIF normalization and non-RGBA tRNS not being explicitly composited.

Case

EXIF： http://qiniu.funxingzuo.top/exif_orient_180.jpg tRNS: http://qiniu.funxingzuo.top/hello.png

Fix

A fix for this vulnerability was merged here: https://github.com/vllm-project/vllm/pull/44974

  • Published: Jun 17, 2026
  • Updated: Jun 18, 2026
  • GHSA: GHSA-8jr5-v98p-w75m
  • Severity: Medium
  • Exploit:
  • CISA KEV:

CVSS v3:

  • Severity: Low
  • Score: 4.8
  • AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:L/A:L

CWEs:

Software From Fixed in
Python icon vllm 0.11.0 0.23.0.x

Frequently Asked Questions

A security vulnerability is a weakness in software, hardware, or configuration that can be exploited to compromise confidentiality, integrity, or availability. Many vulnerabilities are tracked as CVEs (Common Vulnerabilities and Exposures), which provide a standardized identifier so teams can coordinate patching, mitigation, and risk assessment across tools and vendors.

CVSS (Common Vulnerability Scoring System) estimates technical severity, but it doesn't automatically equal business risk. Prioritize using context like internet exposure, affected asset criticality, known exploitation (proof-of-concept or in-the-wild), and whether compensating controls exist. A "Medium" CVSS on an exposed, production system can be more urgent than a "Critical" on an isolated, non-production host.

A vulnerability is the underlying weakness. An exploit is the method or code used to take advantage of it. A zero-day is a vulnerability that is unknown to the vendor or has no publicly available fix when attackers begin using it. In practice, risk increases sharply when exploitation becomes reliable or widespread.

Recurring findings usually come from incomplete Asset Discovery, inconsistent patch management, inherited images, and configuration drift. In modern environments, you also need to watch the software supply chain: dependencies, containers, build pipelines, and third-party services can reintroduce the same weakness even after you patch a single host. Unknown or unmanaged assets (often called Shadow IT) are a common reason the same issues resurface.

Use a simple, repeatable triage model: focus first on externally exposed assets, high-value systems (identity, VPN, email, production), vulnerabilities with known exploits, and issues that enable remote code execution or privilege escalation. Then enforce patch SLAs and track progress using consistent metrics so remediation is steady, not reactive.

SynScan combines attack surface monitoring and continuous security auditing to keep your inventory current, flag high-impact vulnerabilities early, and help you turn raw findings into a practical remediation plan.