Feature Request: WebUI inline image artifacts in assistant messages

[loopy321]

Feature Description

Add support for assistant-side inline image artifacts in the llama-server WebUI.

Current WebUI supports user-side image/PDF attachments, including parsing PDFs as images for vision models. However, assistant responses are text-only. For document-vision workflows, especially PDFs, it would be useful for the assistant/WebUI to render image artifacts inline in the chat.

Example workflow:

1. User uploads a PDF with pdfAsImage enabled.
2. User asks: “Show me diagram 1”
3. Vision model identifies the relevant page/region.
4. WebUI crops the rendered PDF page region.
5. Assistant response renders the cropped image inline.

Motivation

This enables visual document workflows such as:

• extracting visual data from PDFs
• showing figures/tables from scientific papers
• showing chart/table snippets from reports
• collecting visual snippets into chat tables

The model should not redraw or hallucinate the image. The WebUI should crop the original PDF-rendered pixels and display them as an inline artifact.

Possible Implementation

Add an assistant artifact type, e.g.:

type ChatImageArtifact = {
  id: string;
  type: "image";
  label: string;
  mime: "image/png";
  dataUrl?: string;
  sourceFileId?: string;
  page?: number;
  bbox?: [number, number, number, number];
};