openai/codex · warning · io::Error

generated image exceeds the executor file size limit

Error message

generated image exceeds the executor file size limit

What it means

Executor branch of save_image_generation_result (no save_root): the trimmed base64 image string exceeds MAX_EXECUTOR_GENERATED_IMAGE_BASE64_BYTES before decoding, returning io::ErrorKind::InvalidData. The caller catches it, logs 'failed to save generated image', and returns None — the tool call still succeeds; only workspace persistence is skipped.

Source

Thrown at codex-rs/ext/image-generation/src/tool.rs:307

                    .write_file(
                        &PathUri::from_abs_path(&path),
                        bytes,
                        Default::default(),
                        /*sandbox*/ None,
                    )
                    .await?;
                Ok(path)
            }
            .await;
            (output_dir, save_result)
        }
        None => {
            let environment = environment?;
            let output_dir = environment.cwd.join("generated_images");
            let save_result: io::Result<AbsolutePathBuf> = async {
                let result = result.trim();
                if result.len() > MAX_EXECUTOR_GENERATED_IMAGE_BASE64_BYTES {
                    return Err(io::Error::new(
                        io::ErrorKind::InvalidData,
                        "generated image exceeds the executor file size limit",
                    ));
                }
                let bytes = BASE64_STANDARD
                    .decode(result.as_bytes())
                    .map_err(|error| io::Error::new(io::ErrorKind::InvalidData, error))?;
                if bytes.len() > MAX_EXECUTOR_GENERATED_IMAGE_BYTES {
                    return Err(io::Error::new(
                        io::ErrorKind::InvalidData,
                        "generated image exceeds the executor file size limit",
                    ));
                }

                let artifact_path =
                    image_generation_artifact_path(&environment.cwd, session_id, call_id);
                let path = output_dir.join(artifact_path.as_path().file_name().unwrap_or_default());
                let sandbox = Some(&environment.file_system_sandbox_context);

View on GitHub (pinned to 339751715c)

Solutions

1. Request smaller dimensions or lower quality so the base64 payload fits the cap.
2. Configure a host-side save_root so artifacts go through the LOCAL_FS path with its 32 MiB decoded cap instead.
3. Treat the warning as informational if the in-tool base64 result is sufficient for the model.

// Host side: request sizes that fit the executor cap, or supply a save_root
request.size = Some("1024x1024");
request.quality = Some(ImageQuality::Medium);

fn fits_executor_base64_limit(result: &str) -> bool {
    result.trim().len() <= MAX_EXECUTOR_GENERATED_IMAGE_BASE64_BYTES
}

When it happens

Trigger: An image generation/edit tool result delivered through the executor (environment.cwd-based path) whose base64 payload exceeds the executor size cap (about 4/3 of the 32 MiB decoded limit).

Common situations: High-resolution or lossless PNG output from the image model; quality set to high with a large canvas; executor contexts with no host-provided save root.

Related errors

• generated image destination already exists
• generated image directory is not a real directory
• pid-managed app-server shutdown is unsupported on this platf
• pid-managed updater shutdown is unsupported on this platform
• failed to read start time for pid-managed app server {pid}

AI-assisted analysis of openai/codex@339751715c (2026-08-25). Data as JSON: /api/errors/4363c4a6187c02b0. Report an issue: GitHub.