How to Add a Made with AI Label to Images (Free)

Quick answer and why you may need this

Since 2 August 2026, EU AI Act Article 50 has required anyone who deploys an AI system to output images, audio, or video intended for a public audience to add a visible human-readable disclosure. That covers a wide range of real-world scenarios: a social media manager posting AI-generated campaign visuals, a designer using Midjourney for client deliverables, a blogger illustrating articles with Stable Diffusion outputs, or a marketer building product ads with generative tools.

The regulation does not prescribe exactly how the label must look or where it must appear on the image. It does require that a person viewing the content can reasonably recognize it as AI-generated. That leaves room for a small, tasteful badge rather than a large intrusive overlay, especially for artistic and creative works where the regulation explicitly allows a lighter approach.

For a thorough legal breakdown of who must comply, what the exemptions are, and what machine-readable marking means, read the full EU AI Act guide. This article focuses on the practical how-to: getting the label onto your images quickly, correctly, and without handing your files to a third-party server.

General informational guidance only. This article is not legal advice. Consult qualified legal counsel for compliance decisions specific to your situation.

Step by step: using the free AI Label tool

The SammaPix AI Label tool runs entirely in your browser using the Canvas API. Every operation happens in local memory. No file is transmitted to any server at any point in the workflow. The result is a labeled copy of your image ready to publish or hand off to a client.

Step 1. Open the tool

Navigate to sammapix.com/tools/ai-label in any modern browser. Chrome, Firefox, Safari, and Edge all work. No extension, no plugin, and no account are required. The tool loads in a few seconds from a Vercel edge network, so the interface appears almost instantly regardless of your location.

Step 2. Drop your images

Drag one image or an entire folder of images onto the drop zone, or click the zone to open a file picker. Supported input formats are JPEG, PNG, WebP, and AVIF. You can drop as many files as you need in a single batch. The tool reads each file from your local storage using the browser File API and loads it into memory. Nothing leaves your device.

For large campaigns with dozens of images, dropping them all at once is the fastest approach. All images will receive the same label configuration you set in the next steps.

Step 3. Pick your label text

The tool offers three preset label texts:

• Made with AI is the broadest and most widely understood phrase. Use it when an image was generated entirely by an AI model from a text prompt or a reference image.
• AI-generated is slightly more precise and is the phrasing used most often in regulatory contexts. A good choice for professional or compliance-focused use cases.
• AI-assisted fits situations where a human created the base image and AI tools enhanced specific elements, such as background replacement, upscaling, or object removal.

You can also type a fully custom label. Some brands prefer a proprietary phrase like "Created with generative AI" or add an icon glyph before the text. The custom field accepts any short text string.

Step 4. Choose position, style, and size

Position determines where on the image the badge appears. Five options are available:

• Bottom right (default, most conventional placement for watermarks and labels)
• Bottom left
• Top right
• Top left
• Center (rarely used for disclosure labels but available for specific design needs)

Style controls the visual treatment of the badge. Three options exist:

• Solid renders a fully opaque background behind the label text. Maximum legibility on any image. Appropriate when you want the disclosure to be unmissably visible.
• Subtle uses a semi-transparent background so the underlying image shows through. A good choice for artistic and creative works where the regulation permits a lighter disclosure approach.
• Outline draws the label text with a thin border and no filled background. The most minimal and unobtrusive option, suited to images with clean, predictable background areas.

The size slider adjusts font size relative to the image dimensions. A live preview refreshes instantly with every change you make to text, position, style, or size, so you see exactly what the final output will look like before downloading.

Step 5. Preview and download

Once you are satisfied with the preview, click Download. For a single image, the labeled file downloads immediately. For a batch of images, click the batch download button to receive a ZIP archive containing all labeled images in one click. The ZIP is assembled in browser memory using the JSZip library. No server is involved at any stage.

Keep your original unlabeled files in a separate folder. The tool only writes to the downloaded copies. If you later need to adjust the label for a different platform or format, you can re-run the originals through the tool with new settings.

The full workflow at a glance

For teams and agencies that need to document their compliance process, this table summarizes the complete workflow from AI generation to published, labeled image.

Best practices for AI image disclosure

Getting a label onto your image is the minimum. Doing it thoughtfully produces better outcomes for both compliance and audience trust. These are the practices worth building into your workflow from the start.

Place the label where it is readable without dominating the image

Bottom right is the standard convention because it mirrors where copyright notices and watermarks traditionally sit. Audiences have learned to look there. On images where the bottom right corner contains important subject matter, shift to bottom left or top right. Avoid center placement for disclosure labels: it interrupts the composition and suggests the creator is unsure whether to show or hide the label.

Size matters more than people expect. A label that is too small becomes illegible at thumbnail scale, which defeats the purpose of disclosure. A size where the text reads clearly at roughly 10 to 15 percent of the image height is a practical starting point. Adjust from there based on how the image will be displayed.

Match the style to the context

The EU AI Act explicitly distinguishes artistic and creative works from commercial communications and news content. For an art print, a concept illustration, or a creative campaign image, a subtle or outline style label is appropriate and keeps the aesthetic integrity of the work intact. For a press release image, an e-commerce product shot, or a political advertisement, a solid, clearly readable label is the right choice.

The core question to ask: would a person seeing this image at normal viewing size and distance notice the label? If the answer is no, the label is not doing its job. Increase the size or switch to a higher-contrast style.

Batch your entire campaign before publishing

Retroactively labeling images already published is harder than labeling them before the first post. Build the labeling step into your publishing checklist, right before the final review. For agencies, this means the asset handoff from the AI generation stage always includes a "label" substep before the file goes to the client or scheduler.

With the SammaPix AI Label tool, batching costs no extra time beyond the initial drop. Drop the whole campaign folder, configure once, download one ZIP. The per-image overhead after the first setup is essentially zero.

Keep the originals separately

The tool writes only the downloaded copies. Your source files remain unchanged on your device. Maintaining a clean folder of originals lets you re-run the labeling step with different settings if a platform has different display requirements (for example, square crops for Instagram versus landscape for LinkedIn) without going back to the AI generator.

A simple naming convention makes this sustainable: originals in a folder called raw, labeled outputs in a folder called labeled, published versions in a folder called live. Three folders, zero confusion.

Alternatives and why browser-based wins on privacy

Several other approaches exist for adding a visible AI label to images. Each has real limitations that matter in a professional context.

Desktop software (Photoshop, GIMP, Affinity Photo)

Desktop editors give you total control over typography and placement. The limitation is time: opening each file, adding a text layer, flattening, exporting, and closing takes 2 to 3 minutes per image even for an experienced user. A 30-image campaign becomes an hour of repetitive work. There is no batch equivalent of what the SammaPix AI Label tool does in one drop.

Upload-based online watermark tools

Tools that process images on a remote server introduce a meaningful privacy problem. Your AI-generated campaign images often contain commercially sensitive creative direction, unreleased product visuals, or proprietary brand assets. Uploading those files to a third-party server means they leave your control. Server logs, storage retention policies, and breach exposure all become concerns. For client work, sending files to an unknown server may also violate your NDA or data processing agreement.

Browser-based processing sidesteps this entirely. When the Canvas API draws the label onto your image, the operation happens in isolated browser memory. The file bytes travel from your local storage to RAM and back to local storage. No network request carries image data. This is not a claim about trust or intent: it is a structural property of client-side processing that holds regardless of what any server-side service's privacy policy says.

Platform-native labels (Adobe, Meta Content Credentials)

Some AI platforms now embed Content Credentials or C2PA metadata directly into generated images. This satisfies the machine-readable requirement that Article 50 places on providers of AI systems. It does not replace the human-readable visible label requirement for deployers who publish the content. Machine-readable metadata is invisible to the viewer and is often stripped by platforms during upload or resave. A visible label remains necessary alongside any embedded metadata.

The practical conclusion: use a browser-based tool like SammaPix AI Label for the visible human-readable disclosure. Rely on C2PA metadata from the generation platform (where available) for the machine-readable layer. Both together give you the most complete coverage of the Article 50 requirements.