diff --git a/docs/data/material/components/avatars/AvatarA11yImage.js b/docs/data/material/components/avatars/AvatarA11yImage.js
new file mode 100644
index 00000000000000..82b30a04f9d2e4
--- /dev/null
+++ b/docs/data/material/components/avatars/AvatarA11yImage.js
@@ -0,0 +1,17 @@
+import Avatar from '@mui/material/Avatar';
+import Stack from '@mui/material/Stack';
+
+// A 1x1 data-URI image so a real `` renders in the regression harness,
+// which blocks network image requests (see test/regressions/index.test.js).
+// This exercises the working-image path (`alt` forwarded to a native `
`)
+// for the axe `image-alt` rule (WCAG 1.1.1).
+const dataUri =
+ 'data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAC0lEQVR4nGNgYAAAAAMAASsJTYQAAAAASUVORK5CYII=';
+
+export default function AvatarA11yImage() {
+ return (
+
` renders in the regression harness,
+// which blocks network image requests (see test/regressions/index.test.js).
+// This exercises the working-image path (`alt` forwarded to a native `
`)
+// for the axe `image-alt` rule (WCAG 1.1.1).
+const dataUri =
+ 'data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAC0lEQVR4nGNgYAAAAAMAASsJTYQAAAAASUVORK5CYII=';
+
+export default function AvatarA11yImage() {
+ return (
+
`, the initials text, or one icon) in DOM order, so within one avatar there is no sequence whose order conveys meaning.
+- Order matters only across several avatars (a row, a list, or an `AvatarGroup`), which the surrounding layout controls. Confirm that the reading order of a group of avatars matches their visual order.
+
+**Manual testing steps**
+
+1. In a UI with a row of avatars (an `AvatarGroup`, or avatars in a list), inspect the DOM or accessibility tree order.
+2. Compare it to the visual left-to-right order.
+
+**Pass:** reading order matches visual order. A `Stack` or flex layout that visually reorders avatars without changing the DOM would fail.
+
+#### 1.3.3 Sensory Characteristics · A
+
+`🚩` · `✅ Supports` · `○ Author`
+
+- An avatar is identified by its image `alt`, its initials, or an author-supplied label, not by its shape (`variant`), color, size, or position, so it can be referenced without relying on a sensory characteristic.
+- Instructions in the surrounding content must not point to an avatar by shape or position alone (for example, "click the round picture on the left").
+
+**Manual testing steps**
+
+1. Find any product copy that tells users to act on an avatar.
+2. Check that it names the avatar by a text cue (the person's name or label), not only by shape, color, size, or position.
+
+**Pass:** no instruction relies on "the round avatar" or "the one on the right" without a text identifier.
+
+#### 1.4.4 Resize Text · AA
+
+`🚩` · `✅ Supports` · `● Component`
+
+- The container is a fixed 40px square with text in rem (`pxToRem(20)`) and `overflow: hidden`. Under full-page browser zoom (the default scaling mechanism), the box and the text scale together, so initials stay fully visible.
+
+**Manual testing steps**
+
+1. In a UI with letter avatars, set browser zoom to 200% (Ctrl or Cmd and +) and confirm the box and initials scale together with nothing clipped.
+2. Optional: compare against text-only resize (Firefox "Zoom Text Only") to see the px box not scaling; long initials there are an author concern.
+
+**Pass:** initials and fallback text stay fully visible at 200% page zoom. The container scales with the page.
+
+#### 1.4.5 Images of Text · AA
+
+`✅ Supports` · `◐ Shared`
+
+- Letter/initials children and the `alt[0]` fallback are live, CSS-styled DOM text, not images of text.
+- The only risk is an author deliberately passing a bitmap of text as the avatar image.
+- Confirmed by a unit test in [`./Avatar.test.js`](./Avatar.test.js).
+
+**Manual testing steps**
+
+1. Select a letter avatar's initials with the mouse (or zoom in) and confirm they behave like real text (selectable, stay crisp), not an image.
+
+**Pass:** avatar initials and fallback are real text; the component renders no image of text.
+
+#### 1.4.10 Reflow · AA
+
+`🚩` · `✅ Supports` · `● Component`
+
+- The avatar is a fixed 40px box that establishes no horizontal layout, so at a 320 CSS pixel width (or 400% zoom) it cannot force two-dimensional scrolling; it scales and wraps with its container.
+
+**Manual testing steps**
+
+1. In a UI with avatars, set the window (or the DevTools device toolbar) to 320 CSS pixels wide.
+2. Confirm there is no horizontal scrolling and every avatar stays visible.
+
+**Pass:** content reflows with no two-dimensional scrolling. A single 40px avatar never triggers it.
+
+### 🔁 Hybrid
+
+#### 1.1.1 Non-text Content · A
+
+`✅ Supports` · `◐ Shared`
+
+- When given an `src`/`srcSet`, the component renders a native `
` and forwards the author's `alt` as its text alternative; the states without images (`Person` fallback and `SvgIcon` children) are `aria-hidden`, so assistive technology skips them as decorative.
+- A text alternative is the author's responsibility.
+- An icon-only avatar (`IconAvatars`) or the `Person` fallback exposes nothing to assistive technology by default; if the icon carries meaning, the author must add an accessible name.
+- axe-core `image-alt` confirms a text alternative is present on the image avatar.
+
+**Manual testing steps**
+
+1. With a screen reader running (VoiceOver with Safari, or NVDA with Chrome), focus an image avatar (`src` and `alt`) and confirm it renders a native `
` whose `alt` is announced.
+2. For any informative image avatar in the product, confirm it has a meaningful `alt`; without one, a screen reader may announce the image filename.
+3. Confirm an icon-only avatar and the `Person` fallback are `aria-hidden` and silent.
+4. Confirm a decorative image avatar with `alt=""` is ignored by assistive technology.
+
+**Pass:** every informative avatar exposes a text alternative (`img` `alt` or an author-provided label); decorative and no-image avatars are silent. The component does its part; an informative avatar with no `alt` is an authoring failure.
+
+#### 1.3.1 Info and Relationships · A
+
+`✅ Supports` · `◐ Shared`
+
+- The only relationship (image-to-name) is conveyed natively via `
` and `alt` text.
+- If an author builds meaning by composing the avatar with surrounding content (for example, a name and avatar in a list item), conveying that structure is the author's responsibility.
+- Confirmed by unit tests in [`./Avatar.test.js`](./Avatar.test.js).
+
+**Manual testing steps**
+
+1. Inspect an image avatar in the accessibility tree (in Chrome DevTools: Elements panel, Accessibility tab): confirm the `img` exposes `role="img"` with the `alt` as its name.
+2. Confirm a letter or icon avatar's root div is generic (no spurious role) and its decorative icon is `aria-hidden`.
+3. Confirm no grouping, heading, or list semantics are implied by the avatar alone that would need programmatic encoding.
+
+**Pass:** the image-to-name relationship is programmatically available via native `img` and `alt`.
+
+#### 1.4.1 Use of Color · A
+
+`🚩` · `✅ Supports` · `◐ Shared`
+
+- Color is decorative only: the default gray background and the hash-derived `stringToColor` background in `BackgroundLetterAvatars` are aesthetic, so the component never uses color alone to convey meaning.
+- The avatar's information is carried by the image, initials, or icon, not by the background hue, so nothing is lost when color is removed.
+- If an author repurposes the background to encode status or role (for example, green for online), they must add a non-color cue (text, icon, or shape) to satisfy this criterion.
+
+**Manual testing steps**
+
+1. In a UI with color-backed avatars, turn on a grayscale view (in Chrome DevTools: Rendering tab, Emulate vision deficiencies) and confirm the identifying content is the initials or icon, not the hue.
+2. Confirm no information is lost when color is removed, and that the component does not use color alone to distinguish any state.
+
+**Pass:** no information conveyed by the avatar is lost in grayscale; any color-coded meaning an author adds is paired with a non-color cue.
+
+#### 1.4.3 Contrast (Minimum) · AA
+
+`⚠️ Partially Supports` · `◐ Shared`
+
+- Default letter avatars fail: `colorDefault` sets the text color to `background.default` (`#fff` in the light theme) on `backgroundColor` `grey[400]` `#bdbdbd`, about `1.9:1`. The children text is 20px at the inherited weight 400, so it is not WCAG large text and the `4.5:1` threshold applies; at `1.9:1` it fails even the `3:1` large-text threshold. This affects `
` is exempt (it's not text), and the dark-theme default (`grey[600]` `#757575` on `#121212`, about `4.1:1`) is also below `4.5:1`. Icons are non-text and are out of scope here. Authors must override `bgcolor`/`color` to reach `4.5:1` for letter avatars.
+- axe-core `color-contrast` flags the low-contrast letter demos.
+
+**Manual testing steps**
+
+1. With a contrast tool (the color picker in browser DevTools shows a ratio), measure a default gray letter avatar's initials against its background: expect about 1.9:1.
+2. Confirm the initials are 20px at weight 400 (not large text), so `4.5:1` applies.
+3. Measure letter avatars on custom `bgcolor` values (for example `deepOrange[500]`, or a `stringToColor`-derived color) against the white text.
+4. Confirm a working image avatar is exempt (no text on background).
+5. Repeat in the dark theme (`grey[600]` on `#121212`).
+
+**Pass:** all letter and fallback text meets `4.5:1` (or `3:1` if it's sized to qualify as large text). The default gray letter avatar, the `deepOrange[500]` example, and arbitrary `stringToColor` backgrounds do not meet this without author color overrides.
+
+#### 1.4.11 Non-text Contrast · AA
+
+`🚩` · `⚠️ Partially Supports` · `◐ Shared`
+
+- Meaningful icon-child avatars fall short: when an icon is the avatar's sole, unlabeled content (`IconAvatars`), it is a graphical object required to understand the content, so it needs `3:1`. The component's `colorDefault` renders a white icon on `grey[400]`/`#bdbdbd` (`1.9:1`), and author backgrounds vary, for example `green[500]`/`#4caf50` = `2.78:1` fails, while `pink[500]`/`#e91e63` = `4.35:1` passes. So the default and low-contrast author backgrounds are below `3:1`.
+- Exemptions: the colored circle or square is a decorative graphical object, and the `Person` fallback is a decorative placeholder. The default avatar is non-interactive so the user-interface-component part of this criterion does not apply.
+- Under `@media (forced-colors: active)` the component sets `1px solid ButtonBorder`, so the shape stays visible in Windows High Contrast Mode.
+- A meaningful icon avatar needs both a `≥3:1` foreground/background and an accessible name (1.1.1, since icon children default to `aria-hidden`).
+- Rated Partially Supports because the `1.9:1` shortfall comes from the component's own `colorDefault`, whereas the missing name is an author responsibility.
+
+**Manual testing steps**
+
+1. In a UI with icon avatars on different backgrounds, measure each icon against its background: a white icon on the default gray (about 1.9:1) or on `green[500]` (2.78:1) falls below 3:1.
+2. Confirm the colored circle or square itself is decorative, not an information-bearing graphic.
+3. Enable Windows High Contrast / forced-colors and confirm the `1px` `ButtonBorder` keeps the avatar boundary visible.
+
+**Pass:** any icon that conveys meaning meets `3:1` against its background; the decorative container shape and `Person` placeholder are exempt, and the boundary stays visible in forced-colors mode.
+
+### ⚙️ Automated
+
+#### 1.4.12 Text Spacing · AA
+
+`✅ Supports` · `● Component`
+
+- An avatar holds 1 to 2 character initials (or the `alt[0]` fallback) in a fixed 40px box with `overflow: hidden`, so the only text-legibility risk is clipping. The four spacing overrides do not enlarge one or two initials enough to overflow the box, so nothing is clipped.
+- Confirmed by a Playwright regression test in `test/regressions/index.test.js`.
+
+**Manual testing steps**
+
+1. On a two-character letter avatar, apply the WCAG text-spacing values: line-height 1.5, letter-spacing 0.12em, word-spacing 0.16em (paragraph spacing 2em does nothing here, with no paragraphs).
+2. Confirm the initials stay fully visible inside the 40px circle.
+3. Repeat at other avatar sizes to gauge sensitivity.
+
+**Pass:** documented initials and fallback text stay fully visible after applying the four overrides. Long author-supplied strings that already overflow at default spacing are out of scope.
+
+## Not applicable
+
+- **1.3.5 Identify Input Purpose (AA).** Not an input; collects no user information.
+- **1.4.13 Content on Hover or Focus (AA).** Shows no hover or focus content; a `Tooltip` wrapper would own this.
+- **2.4.4 Link Purpose (In Context) (A).** Renders no anchor or `role="link"` by default; an author who sets `component="a"` owns the link's purpose.
+- **2.5.3 Label in Name (A).** Not a labeled control; the `img`'s `alt` is an accessible name, not a visible label.
+- **3.1.2 Language of Parts (AA).** Names and initials are an explicit exception; other text is author-supplied.
+- **4.1.2 Name, Role, Value (A).** Applies to controls; the static `div`/`img` exposes no control role, state, or value (the `img`'s `alt` is covered by 1.1.1).
+- **4.1.3 Status Messages (AA).** Adds no live region; the image-to-fallback swap is not a status message.
+- **Time-based media (1.2.1 to 1.2.5).** No audio or video.
+- **Audio Control (1.4.2).** Emits no audio.
+- **Orientation (1.3.4).** Sets no orientation lock; a layout concern.
+- **Keyboard, pointer, and focus (2.1.1, 2.1.2, 2.1.4, 2.4.3, 2.4.7, 2.4.11, 2.5.1, 2.5.2, 2.5.4, 2.5.7, 2.5.8, 3.2.1, 3.2.2).** The default root is a non-interactive `