Add/Remove Text on Image

Type the words you want on a photo and the AI places them naturally, or point at existing text and the AI erases it, filling in the background beneath.

Drop your image here or browse

JPG, PNG, WebP (max 20MB)

Processing your image...

AI is modifying the text on your image

Download

How It Works

Upload the image you want to modify
Choose Add Text or Remove Text mode
Enter text content (for add mode) and process

Tips

Remove mode works on watermarks, captions, and overlaid text
For adding text, keep it short for best visual results
AI places text naturally based on image composition
Works best with clear, high-contrast images

Also Try

Image to Text

Extract text from any image with AI OCR

AI Meme Generator

Create viral memes with AI

Beard Removal

Remove beard from photos with AI

Why Use Text on Image

Add Text

Write up to 200 characters and the AI renders them directly onto your photo, picking a typographic style that fits the image mood.

Remove Text

Switch to Remove mode and the model locates watermarks, burnt-in captions, and overlaid labels, then inpaints the background as if the text was never shot.

Composition-Aware Placement

The model reads the image before placing text, steering clear of busy areas and faces so the result looks intentional rather than pasted on.

Single-Step Workflow

Upload, choose Add or Remove, hit Process. No manual masking, no selection brushes. One pass handles both the text and the background reconstruction.

Perfect For

Marketing Banners Watermark Removal Social Media Posts Brand Overlays Thumbnail Text Clean Stock Photos

Powered by AI Image Generation

Flux Dev Inpainting Engine

Text on Image runs on the Flux Dev inpainting diffusion model, which processes the entire image before generating any output. When you add text, the model reads the composition, finds a region that can hold type without fighting the subject, and renders letterforms that blend with the lighting and color of the surrounding pixels.

For text removal, the same inpainting pass works in reverse: the model identifies the text region, masks it out, and fills the gap by predicting what the background would look like without the overlay. Results are cleanest on flat or gently textured backgrounds; highly detailed patterns may show minor softening around the repaired area.

Frequently Asked Questions

Yes. Select Remove mode and upload the watermarked image. The model detects the text or logo overlay and inpaints the background beneath it. Results are cleanest when the watermark sits on a plain or softly textured area.

There is no font picker. The diffusion model generates letterforms that fit the visual style of your specific image, so a gritty concert photo will render differently than a clean product shot. You cannot specify a font family directly.

Placement is decided automatically by the model based on the image composition. If the position does not suit your layout, download the result and fine-tune it in any photo editor.

You can upload JPG, PNG, or WebP images up to 20MB. The processed image is always delivered as a PNG.

Remove mode targets overlaid text: watermarks, burnt-in subtitles, copyright notices, and similar additions. Text that was photographed as part of the real scene, like a street sign or a label on a bottle, is harder to separate and may only be partially affected.

Most jobs finish in 15 to 30 seconds. Images with dense fine detail may take a bit longer because the inpainting pass has more pixels to reason about.

Latin-script languages render reliably. Non-Latin scripts such as Arabic, Chinese, or Cyrillic are processed, but the diffusion model may produce inconsistent letterforms, so check the output carefully before using it.

Not directly. The model infers a suitable style from the image itself, so a bold promotional image tends to get heavier type than a soft portrait. If you need a specific typeface, use the output as a base and apply your font in a design tool.

On flat-color or gradient backgrounds the reconstruction is typically clean. On highly detailed textures, grass, fabric, or busy patterns, you may see slight softening or repetition in the repaired patch. A second pass through an inpainting tool can correct stubborn cases.

The text field accepts up to 200 characters. In practice, short phrases, a headline, a tagline, or a name, render sharper than long sentences. The more text you add, the smaller each letter needs to be to fit, which reduces legibility in the output.