How to remove the Gemini ✦ watermark — step by step

Every image generated by Google Gemini AI carries a small ✦ symbol in the bottom-right corner. This is Google's visual watermark — it tells viewers the image was AI-generated, and it is baked into every Gemini output by default.

While the watermark serves an important purpose for transparency, it can get in the way when you want to use Gemini images in presentations, social media posts, or design work. Here is the fastest way to remove it.

Method 1 — CleanerWatermark (recommended)

CleanerWatermark is purpose-built for Gemini watermarks. The tool automatically detects the ✦ position and uses AI inpainting to fill the area with contextually appropriate content — no manual selection needed.

• Go to cleanerwatermark.com
• Drop your Gemini image (or paste with Ctrl+V)
• Wait 5–15 seconds for processing
• Download your clean image

Free tier: 5 images per browser. For unlimited access, the Pro plan is €4.99/month.

Method 2 — Manual editing (Photoshop / GIMP)

You can use Photoshop's Content-Aware Fill or GIMP's Clone tool to manually remove the watermark. This works well for simple backgrounds but requires skill and time, especially for complex textures underneath the mark.

Which method is better?

For most users, CleanerWatermark is faster and produces cleaner results — especially on photos and gradient backgrounds. Manual editing gives more control for complex cases. For bulk processing, CleanerWatermark's Pro plan handles batch downloads automatically.

LaMa vs Stable Diffusion inpainting — which is better for watermark removal?

When building CleanerWatermark, we evaluated several AI inpainting approaches. The two main contenders were LaMa (Large Mask inpainting) and Stable Diffusion inpainting. Here is what we found.

What is LaMa?

LaMa was developed by Samsung Research and published in 2022. It uses Fast Fourier Convolutions (FFCResNetGenerator) to process large regions of an image simultaneously. Unlike convolutional networks that process pixel neighborhoods locally, FFCs have a global receptive field — they understand patterns across the entire image at once.

This makes LaMa extremely good at filling large, regular patterns: sky gradients, fabric textures, uniform backgrounds — exactly the kind of content that appears under a Gemini watermark.

What about Stable Diffusion inpainting?

Stable Diffusion inpainting generates content from a text prompt. It is powerful for creative tasks — replacing a face, adding objects — but requires a meaningful prompt and tends to hallucinate content that was not present in the original image. For watermark removal, where the goal is reconstruction fidelity (not creativity), this is a disadvantage.

Benchmark results

We ran both models on 50 Gemini images across different subjects (portraits, landscapes, objects). LaMa produced seamless results in 94% of cases. Stable Diffusion produced plausible but clearly different content in 40% of cases — visible to the naked eye.

Why we chose LaMa

• Speed: LaMa processes in 3–8 seconds; SD takes 15–45 seconds
• Consistency: LaMa is deterministic for the same input; SD varies each run
• Fidelity: LaMa reconstructs what was there; SD invents new content
• No prompt needed: LaMa is zero-shot for this task

For the specific use case of Gemini watermark removal — a small, predictably placed mark on diverse photographic content — LaMa is clearly the superior choice. Stable Diffusion remains a better tool for creative generation tasks.

What is the Gemini ✦ logo watermark? Everything you need to know

If you have ever generated an image with Google Gemini AI, you have seen it: a small ✦ sparkle symbol in the bottom-right corner. This is the Gemini logo watermark — also called a filigrane in French — and it is automatically embedded in every image Gemini produces.

This article explains exactly what it is, why Google adds it, and what your options are if you want to remove it from your images.

What exactly is the Gemini ✦ symbol?

The ✦ symbol is Google Gemini's brand icon — a four-pointed star that represents Gemini AI. When Gemini generates an image, it overlays this symbol in the bottom-right corner as a visual watermark (filigrane). The mark is semi-transparent and blends with the image background, making it subtle but still visible.

It is not metadata like a C2PA signature — it is a pixel-level visual overlay. That means you can see it on the image itself without needing any special tools.

Why does Google add the Gemini watermark?

Google's goal is AI transparency. The watermark signals to viewers that the image was created by artificial intelligence, not by a human photographer. This is part of a broader industry trend: Adobe, Midjourney, and DALL-E have similar markers.

In some jurisdictions, AI disclosure is legally required. The EU AI Act, for example, mandates that AI-generated content must be labelled. The Gemini ✦ logo is Google's implementation of this requirement. It is baked in by design and cannot be disabled from the Gemini interface.

Where exactly does the watermark appear?

The Gemini watermark is always placed in the bottom-right corner of the image. Its size scales proportionally with the image resolution — roughly 2–4% of the shorter dimension. On a standard 1024×1024 output, it occupies approximately the bottom-right 40×40 pixel region. This predictable position is what makes automatic detection and removal reliable.

How is the watermark added technically?

The Gemini watermark is applied as an alpha-blended overlay: the ✦ symbol is composited onto the image with a set opacity level. For some images this means reverse-alpha-blending can mathematically recover the original pixels. For others — particularly where the image has complex texture under the mark — AI inpainting (like LaMa) produces a cleaner result by intelligently reconstructing the covered area.

How to remove the Gemini ✦ logo from your image

There are two main approaches:

• CleanerWatermark (recommended): Upload your image, and our AI automatically detects the ✦ position and removes it using LaMa inpainting. Free for up to 5 images/day. No account needed.
• Manual editing: Use Photoshop's Content-Aware Fill or GIMP's Clone Stamp. This works but requires skill and takes longer — especially on textured backgrounds.

Frequently asked questions

Is it legal to remove the Gemini watermark?

Generally yes, for personal and creative use. Check the Gemini Terms of Service for commercial redistribution — the rules vary by use case. CleanerWatermark does not make legal recommendations; consult your jurisdiction's requirements.

Does removing the watermark affect image quality?

With AI inpainting (LaMa), the result is seamless in 94%+ of cases. The removed area is reconstructed using surrounding pixels. The output is the same resolution and format as the input — PNG, JPG, or WEBP.

Why can't I just crop the watermark out?

Cropping removes useful image content along with the watermark. Inpainting preserves the full composition, which is usually preferable for presentations, social posts, or design work.

Does Gemini add any invisible watermark too?

Yes — Google also embeds SynthID, an imperceptible digital watermark in the image pixels. SynthID is not removed by visual tools like CleanerWatermark. CleanerWatermark only removes the visible ✦ logo overlay.

What Is the Nano Banana Watermark — and How to Remove It

If you've used Google Gemini to generate or edit an image, you've seen the small ✦ sparkle logo in every output corner. That watermark comes from Nano Banana — Google's AI image generation model built into Gemini.

What is Nano Banana?

Nano Banana is Google DeepMind's image generation model, integrated into the Gemini app and Google AI Studio. Versions released: Nano Banana, Nano Banana Pro, Nano Banana 2, and Nano Banana 2 Pro — each with improved image quality, but a consistent watermark system throughout.

Why does every Gemini image have a watermark?

Every Nano Banana output carries two distinct watermarks:

1. The visible watermark — the ✦ sparkle logo, semi-transparent, applied using alpha blending
2. SynthID — an invisible digital watermark by Google DeepMind, embedded at the pixel level. Imperceptible to the human eye but detectable by Google's verification tools.

Where exactly does the watermark appear?

The ✦ logo always appears in the bottom-right corner: 48×48 px for ~1024 px images, 96×96 px for higher resolution. With Gemini 3.5, the position shifted slightly inward compared to earlier models.

How is the watermark added?

Gemini uses standard alpha compositing. Because the watermark shape and alpha values are fixed and predictable, the equation can be mathematically inverted — called reverse alpha blending — to recover the exact original pixels.

Best removal methods

• Crop — fast but loses part of the image
• AI inpainting — invents pixels, variable results
• Reverse alpha blending — restores exact original pixels, no guessing

Gemini 3.5 Moved Its Watermark — Here's What Changed

If you've recently generated an image with Google Gemini and noticed the ✦ watermark looks slightly different — or your usual removal method left a faint artifact — you're not imagining things. With Gemini 3.5, Google quietly changed the position and alpha profile of its visible watermark.

What exactly moved?

The ✦ logo has always appeared in the bottom-right corner. With Gemini 3.5, it shifted slightly inward and upward. The change is subtle — a few percent of the image dimension — but enough to break any tool using hardcoded pixel coordinates. Those tools now produce a faint ghost shadow instead of clean removal.

Why does pixel position matter so much?

Gemini's watermark uses alpha compositing. To reverse it precisely, you need the exact pixel coordinates and opacity values. An alpha map even slightly off in position causes the inversion to overshoot — leaving a darker ghost in the shape of the watermark instead of a clean image.

Two profiles now coexist

• V2 (default) — Gemini 3.5 and newer: new position with recalibrated alpha map
• V1 (legacy) — pre-Gemini 3.5 outputs: original 48/96 px profile, unchanged

How CleanerWatermark adapted

We updated our detection zone to reflect the new Gemini 3.5 watermark position. The mask now targets the correct region for both legacy and current Gemini outputs. No action needed on your end — just upload your image as usual.

What about SynthID?

Gemini images carry two watermarks: the visible ✦ logo we remove, and the invisible SynthID signal by Google DeepMind. SynthID is not affected by visual removal methods. Our service targets only the visible overlay.