Text Remover
Remove text from images with AI. Paint over text you own or have permission to edit, then process it in your browser with LaMa inpainting.
Select an image up to 3 MB and brush over unwanted text to remove it using in-browser inpainting. Results vary with the background and mask quality.
Your image is processed in your browser after the model loads, not uploaded.
3 MB Input Limit
PNG, JPG, or WebP
Browser Processing
Runs after the model loads
AI Powered
LaMa inpainting model
How the inpainting pipeline works
Formula or method
The tool uses LaMa (Large Mask Inpainting), an AI model converted to ONNX format and run entirely in your browser via onnxruntime-web (WebAssembly). You paint a binary mask over the text you want removed; the model receives the image and mask as two tensors (each resized to 512 x 512, float32) and predicts replacement pixels from the surrounding context. The inpainted patch is then upscaled and composited back at the original image resolution. The model weights are downloaded from iForge Apps CDN on first use and cached by the browser. Your image data is never transmitted anywhere.
Basis and assumptions
- Model input resolution is fixed at 512 x 512 pixels; both the image and mask are downscaled to this size before inference and the result is upscaled back to the original dimensions.
- The execution provider is WebAssembly only. WebGPU is deliberately bypassed: LaMa's Fast Fourier Convolution layer triggers a kernel failure in the WebGPU path of onnxruntime-web, so WASM is used for reliability on all browsers.
- On Apple Safari, inference runs single-threaded to avoid exceeding WebKit's WebAssembly memory ceiling; this is slower but stable on images above 2 megapixels.
- Result quality depends on the background behind the masked text. Regular textures (sky, walls, grass) give convincing fills; complex or unique backgrounds (faces, geometric patterns) often produce visible artefacts.
- The 3 MB input limit is a browser memory constraint, not an API restriction.
What this tool does not decide
- Whether you have the right to modify the image. This tool edits pixels; it does not grant copyright permission. Only use it on images you own or are clearly authorised to change.
- Whether the result is good enough for your purpose. AI inpainting is probabilistic and can produce visible artefacts, especially over complex backgrounds. Always inspect the output before use.
- Whether removing specific text (an attribution line, a licence notice, a platform watermark) complies with the licence or platform terms that govern the image. Check before publishing.
Sources
- Carve/LaMa-ONNX model repository (Apache 2.0 model card; lama_fp32.onnx) (Carve on Hugging Face, mirrored to the iForge Apps CDN) last accessed 2026-06-12
- onnxruntime-web v1.21.0 (MIT licence) (Microsoft) last accessed 2026-06-12
Last checked: 2026-06-12
How AI Text Removal Works
Removing text from an image isn't as simple as painting over it with a solid colour, that leaves an obvious patch. This tool uses LaMa (Large Mask Inpainting), an AI model that analyses the surrounding pixels and generates new content to fill the gap. Similar inpainting workflows are used in photo-editing tools for object removal.
Here's what happens when you click "Remove Text": the brush strokes you paint become a binary mask, white where you painted, black everywhere else. The AI model receives both your original image and this mask, then predicts what the pixels behind the text probably looked like based on the surrounding context. Grass continues as grass. Wood grain continues as wood grain. The result looks natural because the model understands visual patterns, not just colours.
The model downloads from iForge Apps CDN on first use. After the page and model have loaded, selected images are processed in your browser using WebAssembly and ONNX Runtime. This tool does not upload selected images to an iForge Apps server.
Supported Formats & Limits
| Feature | Details |
|---|---|
| Input formats | PNG, JPG, JPEG, WebP |
| Max file size | 3 MB |
| Output format | PNG (full resolution) |
| AI model | LaMa (Large Mask Inpainting) |
| Model size | ~200 MB (downloaded once, then cached) |
| Processing location | In browser after the model loads |
Tips for Cleaner Results
Cover the Text Completely
Paint slightly beyond the edges of each letter. Leaving small gaps means leftover fragments that are harder to clean up than getting it right the first time.
Adjust Brush Size
Use a larger brush for big text blocks and a smaller brush for fine details. Matching the brush to the text size gives cleaner masks and better AI predictions.
Simple Backgrounds Work Best
Text over grass, sky, walls, or solid colours gets removed cleanly. Text over complex patterns or faces may need multiple passes or manual touch-up.
Use Undo Freely
If you accidentally paint over important details, use undo to step back. Building up the mask in small strokes gives you more control than one broad sweep.
Common Use Cases
Cleaning up screenshots. Remove captions, annotations, or UI labels from screenshots before including them in presentations or documentation.
Preparing images for social media. Strip out distracting text overlays from photos before reposting or repurposing them for different platforms.
Restoring old photos. Remove date stamps, annotations, or handwritten notes that were printed or stamped onto scanned photographs.
Product image cleanup. Remove pricing labels, promotional text, or stickers from product photos before listing them on a different marketplace.
What Inpainting Handles Well, and What It Struggles With
Inpainting predicts the missing pixels from what surrounds them, so the background behind the text decides how convincing the result looks. It excels on regular, repeating textures and struggles where the missing area has unique detail the model cannot guess.
| Works well on | Struggles with |
|---|---|
| Sky, walls, grass, sand, water | Text sitting over a face or hands |
| Wood grain and other even textures | Sharp lines and geometric patterns |
| Blurred or out-of-focus backgrounds | Readable text behind the text you remove |
| Small to medium captions and labels | Very large areas covering most of the image |
When a result is not clean, it is usually because the area behind the text held detail the model had no way to reconstruct. A second pass over the leftover fragments often helps.
Why Results Vary
Background complexity
A plain or repeating background gives the model plenty to copy from. A busy, one-of-a-kind background gives it less to work with, so the fill can look softer or smudged.
Mask precision
Paint a little beyond each letter so no faint outline is left behind, but avoid covering far more than the text, since a larger mask means more pixels the model has to invent.
Image resolution
The result is recomposited at the original resolution, so very large images take longer and fine detail at the edges of the mask is more noticeable. Compressing first to fit the 3 MB limit also keeps processing quick.
Use It Responsibly
This tool edits pixels; it does not grant permission. Only remove text from images you own or are clearly authorised to change.
- Respect copyright, photographer credits, and licence terms. Removing an attribution or a credit line you do not own may breach the licence you accepted.
- Follow each platform's rules. Some marketplaces and stock libraries forbid altering supplied imagery.
- Do not use it to mislead. Erasing a date stamp, a disclaimer, or a label to misrepresent an image is on you, not the tool.
Which AI Image Tool Do You Need?
| Tool | Removes | Works Well For | How It Works |
|---|---|---|---|
| Text Remover | Text, captions, labels | Screenshots, memes, owned overlays | You paint over text, AI fills the gap |
| Background Remover | Entire background | Product photos, portraits, logos | AI detects foreground, removes everything else |
| Watermark Remover | Semi-transparent overlays | Images you own or may edit | You paint over an overlay, AI reconstructs |
Related brush tools use similar inpainting workflows. Use them only on images you own or are authorised to edit.
Related Tools
How to use this tool
Select your image: drag and drop or click to choose a PNG, JPG, or WebP.
Paint over the text you want removed with the brush, adjusting the brush size to fit.
Click Remove Text, then download the cleaned image as a full-resolution PNG.
Common uses
- Cleaning up screenshots you own for presentations and documentation
- Erasing date stamps and captions from scanned photos
- Removing your own labels or annotations from images
- Tidying product photos you have rights to edit
Share this tool
Frequently Asked Questions
Is Forge Text Remover free to use?
Do you store or upload my images?
How do I use it?
What image formats are supported?
Why is the first use slower?
What is LaMa inpainting?
Can I remove text overlays I do not own?
What about text on complex backgrounds?
Can I undo my brush strokes?
Does it work on mobile?
How large can the image be?
Can I remove logos and graphics too?
Results are for general informational purposes only and should be checked before use. They are not professional advice. See our Disclaimer and Terms of Service.