feat: image input on chatgpt.ask.ask and gemini.ask.ask (0.43.0) (#34)

Pass `image_url` and the model looks at the picture and answers about it —
JPEG, PNG, GIF, WEBP, BMP, up to 5 MB. An image ask is slower than a text
one: the model uploads the picture before it starts answering, so allow
90-150 s rather than the text path's 20-70 s.

Both ask responses gain `images: MediaItem[]` — the pictures the answer
itself displayed, favicon-sized assets excluded.

Neither endpoint will GENERATE an image. Anonymous chatgpt.com and
gemini.google.com both gate image generation behind a login (their own
logged-out pages say so), so `images` is only ever what the answer showed,
never something the model drew. The doc comments say this outright rather
than leaving a caller to discover it.

`image_url` is optional and the base client already drops undefined query
params, so an existing call is byte-identical on the wire.

Requested in SB-001078.


Claude-Session: https://claude.ai/code/session_01JdMkgmhJ1o6yc7MT3BuCvW

Co-authored-by: kasparasizi1 <132673909+kasparasizi1@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

Author: 3 people

CHANGELOG.md

@@ -5,6 +5,23 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.43.0] - 2026-09-04
+
+### Added
+
+- **Image input on `chatgpt.ask.ask` and `gemini.ask.ask`** — pass `image_url`
+  and the model looks at the picture and answers about it (JPEG, PNG, GIF,
+  WEBP, BMP; up to 5 MB). An image ask is slower than a text one — allow
+  90-150 s. Requested in SB-001078.
+- **`images` on both ask responses** (`MediaItem[]`) — the pictures the answer
+  itself displayed, with favicon-sized assets excluded.
+
+### Note
+
+Neither endpoint will GENERATE an image. Anonymous `chatgpt.com` and
+`gemini.google.com` both gate image generation behind a login, so `images` is
+only ever what the answer showed, never something the model drew.
+
 ## [0.42.0] - 2026-09-04
 
 ### Added

package.json

@@ -1,6 +1,6 @@
 {
   "name": "scrapebadger",
-  "version": "0.42.0",
+  "version": "0.43.0",
   "description": "Official Node.js SDK for ScrapeBadger - Async web scraping APIs for Twitter, Google, Vinted, Reddit, and more",
   "type": "module",
   "main": "./dist/index.js",

src/chatgpt/ask.ts

@@ -40,6 +40,9 @@ export class AskClient {
    * @param params.prompt - The prompt to send (max 4096 characters).
    * @param params.country - ISO-3166 alpha-2 egress country (default: "US").
    * @param params.web_search - "auto", "force", or "off" (default: "auto").
+   * @param params.image_url - Public image URL to attach; ChatGPT answers about
+   *   the picture. It will NOT generate one — anonymous chatgpt.com requires a login
+   *   for that. Allow 90-150s for an image ask.
    * @returns The answer, its citations, and the full retrieved search set.
    *
    * @example
@@ -61,6 +64,7 @@ export class AskClient {
         prompt: params.prompt,
         country: params.country,
         web_search: params.web_search,
+        image_url: params.image_url,
       },
     });
   }

src/chatgpt/types.ts

@@ -20,6 +20,15 @@ export interface ChatGPTAskParams {
   country?: string;
   /** Whether ChatGPT should browse (default: "auto") */
   web_search?: ChatGPTWebSearchMode;
+  /**
+   * Public http(s) URL of an image to attach. ChatGPT looks at the picture and
+   * answers about it (JPEG/PNG/GIF/WEBP/BMP, up to 5 MB). An image ask takes
+   * noticeably longer — allow 90-150s.
+   *
+   * ChatGPT will NOT generate an image: anonymous chatgpt.com gates that behind a
+   * login.
+   */
+  image_url?: string;
 }
 
 /** Parameters for the brand-visibility endpoint. */
@@ -106,6 +115,14 @@ export interface ChatGPTSearchResult {
 // =============================================================================
 
 /** A ChatGPT answer with its sources. */
+/** An image the answer itself displayed. */
+export interface MediaItem {
+  /** Image URL. */
+  url: string;
+  /** Alt text, when one was supplied. */
+  title?: string | null;
+}
+
 export interface ChatGPTAskResponse {
   /** The prompt that was sent */
   prompt: string;
@@ -119,6 +136,11 @@ export interface ChatGPTAskResponse {
   search_results: ChatGPTSearchResult[];
   /** Distinct domains across the sources */
   source_domains: string[];
+  /**
+   * Images the answer displayed. Never a generated image — see
+   * `image_url` on the params.
+   */
+  images: MediaItem[];
   /** Whether ChatGPT ACTUALLY browsed the web */
   /** True when the render budget expired mid-answer; `answer` is partial. */
   truncated: boolean;

src/gemini/ask.ts

@@ -38,6 +38,9 @@ export class AskClient {
    * @param params.prompt - The prompt to send (max 4096 characters).
    * @param params.country - ISO-3166 alpha-2 egress country (default: "US").
    * @param params.web_search - "auto", "force", or "off" (default: "auto").
+   * @param params.image_url - Public image URL to attach; Gemini answers about
+   *   the picture. It will NOT generate one — anonymous gemini.google.com requires a login
+   *   for that. Allow 90-150s for an image ask.
    * @returns The answer, its citations, and the full retrieved search set.
    *
    * @example
@@ -59,6 +62,7 @@ export class AskClient {
         prompt: params.prompt,
         country: params.country,
         web_search: params.web_search,
+        image_url: params.image_url,
       },
     });
   }

src/gemini/types.ts

@@ -20,6 +20,15 @@ export interface GeminiAskParams {
   country?: string;
   /** Whether Gemini should ground with web search (default: "auto") */
   web_search?: GeminiWebSearchMode;
+  /**
+   * Public http(s) URL of an image to attach. Gemini looks at the picture and
+   * answers about it (JPEG/PNG/GIF/WEBP/BMP, up to 5 MB). An image ask takes
+   * noticeably longer — allow 90-150s.
+   *
+   * Gemini will NOT generate an image: anonymous gemini.google.com gates that behind a
+   * login.
+   */
+  image_url?: string;
 }
 
 /** Parameters for the brand-visibility endpoint. */
@@ -90,6 +99,14 @@ export interface GeminiSearchResult {
 // =============================================================================
 
 /** A Gemini answer with its sources. */
+/** An image the answer itself displayed. */
+export interface MediaItem {
+  /** Image URL. */
+  url: string;
+  /** Alt text, when one was supplied. */
+  title?: string | null;
+}
+
 export interface GeminiAskResponse {
   /** The prompt that was sent */
   prompt: string;
@@ -103,6 +120,11 @@ export interface GeminiAskResponse {
   search_results: GeminiSearchResult[];
   /** Distinct domains across the sources */
   source_domains: string[];
+  /**
+   * Images the answer displayed. Never a generated image — see
+   * `image_url` on the params.
+   */
+  images: MediaItem[];
   /** True when the render budget expired mid-answer; `answer` is partial. */
   truncated: boolean;
   /** Whether Gemini ACTUALLY grounded the answer with a web search */

src/internal/version.ts

@@ -6,4 +6,4 @@
  *
  * @internal
  */
-export const SDK_VERSION = "0.42.0";
+export const SDK_VERSION = "0.43.0";