chore(📚): small documentation fixes

apps/docs/content/api/index.mdx

@@ -36,6 +36,10 @@ This section documents only the additions that are specific to React Native, the
 
 - [`Canvas`](/api/canvas): the React component that gives you a WebGPU rendering surface.
 - [`GPUDeviceProvider`](/api/gpu-device-provider): provides a `GPUDevice` to your component tree.
-- [`installWebGPU`](/api/install-webgpu): installs the WebGPU globals on a worklet runtime.
 - [`warnIfNotHardwareAccelerated`](/api/warn-if-not-hardware-accelerated): warns when the device falls back to software rendering.
-- [Native Extensions](/api/gpu-device-extensions): React Native specific helpers added to `GPUDevice`.
+- [`installWebGPU`](/api/install-webgpu): installs the WebGPU globals on a worklet runtime.
+- [Native Extensions](/api/gpu-device-extensions): native buffer and frame import, `createImageBitmap`, high bit depth formats, and other additions to the WebGPU globals.
+- [Video Player](/api/video-player): decodes a video into frames the GPU samples without a copy.
+- [`drawElementImageToTexture`](/api/draw-element-image-to-texture): copies the pixels of a React Native view into a `GPUTexture`.
+
+Also covered in the guides: [`GPUOffscreenCanvas`](/docs/getting-started/offscreen-canvas) for headless rendering, and [`importDevice`](/docs/getting-started/native-api#importing-an-external-device) / [`adoptTexture`](/docs/integrations/react-native-skia#zero-copy-texture-sharing) for sharing devices and textures with other libraries.

apps/docs/content/api/warn-if-not-hardware-accelerated.mdx

@@ -7,42 +7,15 @@ Checks `adapter.info.isFallbackAdapter` and logs a console warning. Common on An
 
 ## Usage
 
-Call once when the adapter becomes available, typically in a `useEffect`:
+[`useDevice`](/docs/getting-started/react-api#usedevice) and [`GPUDeviceProvider`](/api/gpu-device-provider) already call it for you. Call it yourself when you request the adapter directly:
 
 ```tsx twoslash
-import { useEffect } from "react";
-import { useDevice, warnIfNotHardwareAccelerated } from "react-native-webgpu";
-
-export function Scene() {
-  const { adapter, device } = useDevice();
-
-  useEffect(() => {
-    if (adapter) {
-      warnIfNotHardwareAccelerated(adapter);
-    }
-  }, [adapter]);
-
-  if (!device) return null;
-
-  // configure canvas, build pipelines…
-  return null;
-}
-```
-
-With [`GPUDeviceProvider`](/api/gpu-device-provider), read the adapter from [`useMainDevice`](/api/gpu-device-provider#usemaindevice) the same way:
-
-```tsx twoslash
-import { useEffect } from "react";
-import { useMainDevice, warnIfNotHardwareAccelerated } from "react-native-webgpu";
+import { warnIfNotHardwareAccelerated } from "react-native-webgpu";
 // ---cut---
-function Scene() {
-  const { adapter } = useMainDevice();
-
-  useEffect(() => {
-    if (adapter) warnIfNotHardwareAccelerated(adapter);
-  }, [adapter]);
-
-  return null;
+const adapter = await navigator.gpu.requestAdapter();
+if (adapter) {
+  warnIfNotHardwareAccelerated(adapter);
+  const device = await adapter.requestDevice();
 }
 ```
 

apps/docs/content/docs/getting-started/offscreen-canvas.mdx

@@ -25,3 +25,7 @@ device.queue.submit([encoder.finish()]);
 const pixels = await canvas.getImageData(); // CPU readback
 ```
 
+`getImageData()` copies the canvas texture into a buffer, maps it, and resolves with `{ data, width, height, format }`, where `data` holds 4 bytes per pixel, row by row. Because WebGPU copies rows in multiples of 256 bytes, the canvas `width` must be a multiple of 64.
+
+It is meant for tests, snapshots and one-off exports; to read pixels repeatedly, copy the texture into a reusable `GPUBuffer` of your own with `copyTextureToBuffer`.
+

apps/docs/content/docs/integrations/react-three-fiber.mdx

@@ -119,7 +119,7 @@ export const Fiber = () => (
 | **Presentation** | Same as raw Three.js - R3F never calls `present()` for you |
 | **Orbit controls** | Pointer events must attach to a wrapping `View`, not the Canvas - spread gesture handlers on the parent `View` that wraps `FiberCanvas` |
 | **Worklets** | Keep `GPUDevice` on the main thread if combining with worklets ([Worklets](/docs/integrations/worklets)) |
-| **Model loading** | GLTF needs `resolveAssetSource` polyfill ([Expo setup](/docs/getting-started/expo)) |
+| **Model loading** | Resolve bundled models to a URL with `Image.resolveAssetSource` ([Loading assets](/docs/integrations/three-js#loading-assets)) |
 | **Effect deps** | `FiberCanvas` runs setup in `useEffect` without deps - remount the component to rebuild the GL context |
 
 ## What you learned

apps/docs/content/docs/integrations/three-js.mdx

@@ -174,7 +174,7 @@ renderer.dispose();
 ```
 
 <Callout type="warn" title="Known leak in three r184">
-Three's `RenderObjects.dispose()` leaves `dispose` listeners of the disposed renderer on a module-level shared `QuadMesh` geometry, which keeps the old backend alive across remounts. Until the upstream fix lands, the [example app](https://github.com/wcandillon/react-native-webgpu/blob/main/apps/example/src/ThreeJS/components/makeWebGPURenderer.ts) clears those listeners after `dispose()`. See [issue #445](https://github.com/wcandillon/react-native-webgpu/issues/445).
+A disposed renderer stays in memory across remounts. Until the upstream fix lands, the [example app's `makeWebGPURenderer`](https://github.com/wcandillon/react-native-webgpu/blob/main/apps/example/src/ThreeJS/components/makeWebGPURenderer.ts) works around it after `dispose()`. See [issue #445](https://github.com/wcandillon/react-native-webgpu/issues/445).
 </Callout>
 
 ## Loading assets

apps/docs/content/docs/integrations/typegpu.mdx

@@ -141,4 +141,4 @@ The [example app](https://github.com/wcandillon/react-native-webgpu/tree/main/ap
 
 - [GradientTiles](https://github.com/wcandillon/react-native-webgpu/blob/main/apps/example/src/GradientTiles/GradientTiles.tsx) - the render pipeline above, with buttons driving the uniform via React state.
 - [ComputeBoids](https://github.com/wcandillon/react-native-webgpu/blob/main/apps/example/src/ComputeBoids/ComputeBoids.tsx) - a flocking simulation combining a compute pipeline with an instanced render pipeline, double-buffered storage buffers, and tweakable physics parameters.
-- [MNISTInference](https://github.com/wcandillon/react-native-webgpu/blob/main/apps/example/src/MNISTInference/MNISTInference.tsx) - hand-written digit recognition running a neural network in compute shaders, with [React Native Skia](https://shopify.github.io/react-native-skia/) for the drawing surface.
+- [MNISTInference](https://github.com/wcandillon/react-native-webgpu/blob/main/apps/example/src/MNISTInference/MNISTInference.tsx) - hand-written digit recognition running a neural network in compute shaders, with [React Native Skia](https://wcandillon.github.io/react-native-skia/) for the drawing surface.