Aaron Preece
Welcome to the second entry in our series on music generation AI. In this article, we’ll be reviewing the AI music generation service called Riffusion, which recently released its latest music generation model, FUZZ 1.1, along with paid subscriptions. If you’d like to read part one of the series, you can find my review of the Suno AI web interface here.
As with that review, this article primarily focuses on the accessibility of the interface itself, rather than the music generation capabilities. I tested Riffusion using the NVDA screen reader with Firefox. Note that I used a free account, so some options—such as “Vibes,” sound snippets that can be used in song generation—weren’t available.
Homepage and Navigation
The Riffusion homepage begins with the site’s logo, followed by left and right sidebars. For screen reader users, these are broken down neatly with headings under sections such as “Explore” and “Library.”
There are only a few unlabeled icons in this section: one appears to do nothing, and another opens the audio upload dialog—but only if you're logged in. When logged out, several buttons—such as those for uploading audio, generating music, liking songs, or opening the "More Actions" menu—instead take you to the login screen. These buttons are not labeled in the logged-out state. For instance, the button that becomes the "More Options" submenu for specific songs is unlabeled and directs you to the login screen.
When logged in, a search field and a few extra buttons appear at the "bottom" of the header from a screen reader’s perspective.
The main content begins with promotional elements like the latest Riffusion model info, artist spotlights, and featured playlists. When using a screen reader’s link navigation shortcuts, these appear as single links. However, when navigating with arrow keys, they span multiple lines. Two unlabeled buttons or links appear per song or link in this section.
This featured content uses heading level 3 labels, appropriately subordinate to the primary navigation. Below this, the “Trending Songs” section is preceded by a heading level 2, making it easy to jump to. Tabs to switch between Songs and Playlists are properly labeled and show their active or inactive state.
Sort options like “Staff Picks,” “Today,” “Weekly,” and “Monthly” are all accessible and allow filtering by timeframe. Songs load in 20-song increments, each introduced with a number. These are clearly structured and accessible when in list view.
Each song entry includes a play/pause button that correctly indicates its state, the song title and style, the creator’s name, a favorite button, and a “More Actions” menu. This menu remains unlabeled.
One accessibility hazard worth noting is an unlabeled button just below the sort filters that toggles between list view (default) and grid view. Switching to grid view significantly reduces accessibility. Song numbers, which appear as plain headings in list view, now become both headings and buttons that begin or stop playback—but do not indicate their current state. The Like button is labeled only as “Button,” making it appear as two indistinguishable buttons in a row to screen readers. Additionally, the creator’s profile image and name are combined into a single link, with the raw URL of the image appended to the username, forcing screen reader users to listen to the entire image URL before hearing the creator’s name.
In contrast, List View handles these elements much more cleanly. Fortunately, since list view is the default, this issue is easy to avoid unless manually toggled.
Songs continue loading in increments until the bottom of the list is reached. If you scroll to the bottom, you’ll find playback controls for the currently playing song. These are mostly labeled, though one unlabeled button is present at the bottom of the playback section. This button opens a volume adjustment dialog. The slider within this dialog is also unlabeled, but its function is relatively easy to determine.
Song Page Interface
If you go to a song’s individual page, the layout is clean and simple. You can jump directly to the song’s play button using the screen reader’s landmark shortcut.
Immediately following this play button is a list of other recommended songs by the same creator, which appears before the heading for the current song.
The song’s title is marked with a heading level 2. Below this are options like “Use Vibe” and “Remix,” followed by several unlabeled buttons. These include a Like button, a Copy Song Link button (which is unlabeled but does announce to the screen reader when the link has been copied), and a More Actions submenu.
Additional song information, including style and lyrics, is clearly presented in plain text.
Account Creation and Login
Upon account creation, the welcome dialog may not appear correctly with a screen reader. A second dialog often follows with information about the latest model. This can be closed, after which you’ll land on a terms of service screen.
Unfortunately, this screen displays only two blank lines and the word “Riffusion” when viewed with a screen reader. You must use OCR (e.g., NVDA’s built-in OCR) to read the screen and locate the agreement checkbox. You can then route the mouse to click the checkbox, allowing you to proceed.
To complete setup, use your Google account or email to sign in, choose a username (required), and optionally upload a profile image. The upload button is an unlabeled field but opens a standard file dialog.
After agreeing to the terms and setting up your account, you'll see a subscription dialog followed by a return to the homepage. Notably, formerly unlabeled buttons—such as those under each song—are now replaced with proper action menus for adding to playlists or reporting songs.
Other Unlabeled Elements
After the search field—located at the bottom of the header from a screen reader’s perspective—there are three unlabeled buttons. The first does nothing. The second reroutes focus to the search field. The third opens the updates dialog.
The “Avatar” button, present when logged in, opens a menu where you can adjust various settings related to your account.
Upgrade Screen
The Upgrade screen is mostly accessible. Radio buttons for selecting monthly or yearly billing are labeled correctly, and buttons for each plan likewise have proper labels. However, the plan selection section does not use proper table markup, making comparisons harder for screen reader users.
Music Generation Interface
To generate a song, you must first activate the “Create the music you imagine...” option in the header. Although this appears as plain text, it is actionable. Once you begin the creation process, this text updates to reflect the song you're working on or have just generated.
This interface is generally well-structured. Tabs for Compose and Prompt modes are labeled and correctly indicate their state. Input fields and most buttons are labeled. However, the clear content button at the top is unlabeled, as are the checkboxes in Compose mode—though these appear next to corresponding text, which helps. The “Add random tags” button is unlabeled and appears after “Add Vibe.” If the Advanced checkbox is selected, the “Edit Seed” button is also unlabeled.
The "Ghostwriter" button opens a further dialog where you can have Riffusion write lyrics for you. The edit field and "Write for Me" button are clearly labeled. When you finish and press the "Write for Me" button, you’ll be sent back to the main page and will need to re-enter the song creation dialog.
After selections are made and Generate is pressed, two headings appear, each tracking the progress of one of the two songs being generated. Riffusion always generates two at a time. These are presented with headings and accessible playback controls, similar to those on the homepage.
Modifying Songs
You can remix songs in various ways. Options include swapping vocals or background instrumentals, remastering the track, or creating a cover version. You can also trim the song using properly labeled input fields.
Covers use a dialog similar to the original generation interface and retain the same level of accessibility. Within this interface, you can modify lyrics, sound style, and other features.
Conclusion
With a few exceptions, the Riffusion interface is impressively accessible. The most significant barrier occurs during account creation, where the terms of service screen is entirely unreadable without OCR. However, this can be bypassed with NVDA’s OCR and mouse routing. Having access to an AI or person who can confirm the checkbox for agreeing to terms is selected may also be helpful.
Beyond that, only minor unlabeled buttons persist throughout the interface. The overall structure is intuitive, headings are used appropriately, and the screen reader cursor behaves as expected. Unlike many modern web apps, Riffusion feels like navigating a well-structured HTML site, which makes it much easier for screen reader users to browse and interact confidently.
