Skip to main content

AI alt text

Generate SEO-friendly alt text for every image in one click. How it works, what voice it follows, and the session cap.

Last updated April 27, 2026

AI alt text is the only Renamerly feature that sends image data off-device. It is opt-in per image and Pro-gated.

What it does

Click Generate Alt Text on any image row (or bulk-select + Generate). Each selected image is sent to our server, which returns one short caption optimized for accessibility and SEO. The caption is written to the image's altText field and included in:

  • The CSV manifest that ships alongside every ZIP export.
  • The alt= attribute of images when exporting to HTML-ready formats.
  • Any downstream CMS integration (on the roadmap).

What context the AI sees

Every alt-text request is generated from real signals you've already given the workspace, not from a one-size-fits-all prompt. Specifically, the prompt is assembled from:

  • The active template's name and category. A template categorized as Photography produces narrative, scene-setting captions; one categorized as E-commerce produces SEO-friendly product copy; Real estate emphasizes rooms and features. Categories the AI doesn't recognize fall through to a factual-and-descriptive default.
  • The full row metadata bag. Every non-empty key on the row — sku, descriptor, brand, product, color, view, any custom keys you've added — is passed to the AI as ground truth. The AI is told to treat your metadata as fact and describe what's actually in the image.
  • The active filename format. The keys your filename references (e.g. brand, product, view for an SEO chain) tell the AI which attributes you consider most important for naming, so the caption emphasizes those.
  • The manual group, if any. If the row belongs to a hero-shots or lifestyle bucket, the AI is told the group name and may use it as scene context.
  • The template's voice directive. A free-text one-liner that takes precedence over the persona default.

The result: the same image with three different templates produces three different captions, each shaped for its persona. An image of a Nike sneaker captioned under the E-commerce / SEO Attribute Chain template reads as a product listing; the same image under a portfolio template reads as editorial scene-setting.

How the voice is set

Every template declares an altTextVoice string. The starter templates ship with:

  • Photography / Manual Groups → "narrative, scene-setting, one short caption focused on subject and mood".
  • Photographer / ISO 8601 Deliverables → "editorial, neutral, one short caption per image".
  • E-commerce / SEO Attribute Chain → "product-focused, keyword-friendly, 1 sentence emphasizing material + use case".

When no template is active, the default voice is "short, factual, SEO-friendly alt text describing the subject and any prominent visual attributes". Pro users editing a template can change the voice per template — see Creating a template.

Voice directives are clipped to 240 characters before being included in the prompt — long enough for a precise tone description, short enough to keep the prompt bounded and to limit blast radius from a hostile or accidentally-runaway voice string.

Session cap

Session alt-text requests
default: 20

Maximum AI alt text generations per workspace session, across Free and Pro.

The cap resets when you reset the workspace or start a new session. It is a fair-use ceiling — most users hit it only when bulk-generating across 50+ images in one sitting. Contact support via /pricing if you consistently need more; Enterprise tiers remove the cap.

Data handling

  • Images are sent as base64-encoded JPEG thumbnails (max 1024×1024, 80% quality) over HTTPS to our server.
  • Our server forwards the thumbnail to the AI provider, stores no images, and logs only anonymous request metadata for rate limiting.
  • Responses are cached on-device for the current session so re-generating is a free instant-restore.
  • AI voice strings are included in the prompt only. They are not logged on the provider side.

See Privacy for the full data-flow diagram.

Edge cases

  • RAW images: our server receives a JPEG preview derived from the RAW. If your RAW lacks an embedded preview (rare on modern cameras), the request fails silently and the row is skipped.
  • Already-generated alt text: regenerating overwrites. There is no automatic versioning; if you want two candidates, copy the first into a comment before regenerating.
  • Offline: the button is hidden when the browser is offline or the user is on Free.