[enhancement]: Add managed file inputs for custom nodes

[Gohankaiju]

Is there an existing issue for this?

• I have searched the existing issues

Contact Details

No response

What should this feature add?

I would like custom nodes to be able to accept user-uploaded document/data files from the browser, without requiring the file to already exist on the InvokeAI server.

A concrete use case is a custom node that accepts a PDF, Markdown file, text file, CSV, JSON, or YAML file and processes it in Python. For example, a node may upload a PDF to an external document-processing service, parse a CSV, or load structured data from JSON/YAML.

Today this is difficult in hosted InvokeAI deployments. A string path only works when the file is already visible to the server, which is not true when a user accesses a remote InvokeAI instance from their own browser. The existing upload flow is image-specific and tied to ImageField, image DTOs, thumbnails, and image upload APIs.

A lightweight first version could focus only on node inputs:

• Add a managed FileField or equivalent invocation input type
• Let users upload/select/remove a file directly from the node input UI
• Store the uploaded file in a server-managed location
• Pass a stable file reference to the invocation instead of a local path or file contents
• Expose a safe server-side API such as context.files.get_path(file_id) or similar
• Validate uploads with extension/MIME allowlists and size limits

Alternatives

The current alternatives do not solve the hosted custom-node use case:

• Passing a string path only works when the file already exists on the server.
• Reusing ImageField is not appropriate because it is image-specific.
• Encoding files into text/base64 values would bloat workflow and queue data.

Additional Content

No response