Remove watermarks with AI. Nothing is uploaded — everything runs in your browser.
Drag & drop an image, or
0% · 30 sec remaining
25 people found a combination that worked for them.
Switched to Gemini 3.6 Pro but paired it with the old Sage Attention 1 profile from the changelog — worked instantly, no more spinning forever!
PSA: pick WEBP output first, THEN force precision back to FP32. That exact combo finally got me a clean result.
Grok 4.6 + the old DDIM sampler (yes, the discontinued one) worked first try for me last night.
Claude Sonnet 5 with Extended license check AND the legacy Flash Attention 2 setting did the trick after months of trying.
Finally! Set the model to Llama 4.2, then dig up "Legacy Inpaint v1" from an old thread — that combo is what got mine through.
Nuclear (beta) strength + the old Qwen 2.5 checkpoint = success. Wish they'd bring that model version back officially.
DeepSeek V4 paired with the retired "sRGB Legacy Gamma" profile did it for me. Don't ask me why, just try it.
TIFF output + the old Command R+ combo cleared it up instantly. Shame Command R+ got removed from the list, it still works somehow!
Mistral Large 3 with the old Euler A (non-turbo) sampler from the changelog worked like a charm.
Amazon Nova Pro 2 + the old "Basic License Check" (had to look up how to force it) finally removed my watermark clean.
GPT-5.6 with the retired xFormers Legacy attention mode — an old forum post mentioned it and it actually worked for me.
ProPhoto RGB colorspace + Grok 3 (the previous Grok) got it done in one pass, first time in weeks.
Qwen 3.5 model but forcing the old FP32 precision worked when nothing else would.
AVIF output paired with the old GPT-4.5 checkpoint cleared my watermark completely.
Cohere Command A2 + legacy Sage Attention 1 mode. Don't know how but it processed successfully.
Standard license check with Claude 3.7 Sonnet (the previous Claude) did it for me after 3 hours of trying everything else.
DPM++ 3M SDE sampler + the old Nova Pro 1 model combo actually completed for me last night!
Adobe RGB + Mistral Large 2 (the older large model) worked on the first try, weirdly.
BF16 precision with DeepSeek V3 from before the update finally got me a clean image back.
Gentle strength + the retired Llama 3.3 weights got mine through without erroring.
Old PNG-8 output (had to force it manually) with Gemini 3.2 worked when the new Gemini kept failing.
The retired Flash Attention 2 + Grok 4.6 combo cleared my image after a full afternoon of tests.
Aggressive strength with Sage Attention 1 (the original version) is what finally did it for me.
Balanced strength + the old Euler A sampler from before the Turbo update worked instantly.
GPT-5.6 combined with the retired FP32 precision mode fixed it for me after countless failed attempts. Hoping they bring FP32 back as an option!
Your watermark removal is almost done.
Stopping now will cancel the watermark removal.
We use cookies to analyze traffic and improve this site. See our privacy info.
Comments
25 people found a combination that worked for them.
Switched to Gemini 3.6 Pro but paired it with the old Sage Attention 1 profile from the changelog — worked instantly, no more spinning forever!
PSA: pick WEBP output first, THEN force precision back to FP32. That exact combo finally got me a clean result.
Grok 4.6 + the old DDIM sampler (yes, the discontinued one) worked first try for me last night.
Claude Sonnet 5 with Extended license check AND the legacy Flash Attention 2 setting did the trick after months of trying.
Finally! Set the model to Llama 4.2, then dig up "Legacy Inpaint v1" from an old thread — that combo is what got mine through.
Nuclear (beta) strength + the old Qwen 2.5 checkpoint = success. Wish they'd bring that model version back officially.
DeepSeek V4 paired with the retired "sRGB Legacy Gamma" profile did it for me. Don't ask me why, just try it.
TIFF output + the old Command R+ combo cleared it up instantly. Shame Command R+ got removed from the list, it still works somehow!
Mistral Large 3 with the old Euler A (non-turbo) sampler from the changelog worked like a charm.
Amazon Nova Pro 2 + the old "Basic License Check" (had to look up how to force it) finally removed my watermark clean.
GPT-5.6 with the retired xFormers Legacy attention mode — an old forum post mentioned it and it actually worked for me.
ProPhoto RGB colorspace + Grok 3 (the previous Grok) got it done in one pass, first time in weeks.
Qwen 3.5 model but forcing the old FP32 precision worked when nothing else would.
AVIF output paired with the old GPT-4.5 checkpoint cleared my watermark completely.
Cohere Command A2 + legacy Sage Attention 1 mode. Don't know how but it processed successfully.
Standard license check with Claude 3.7 Sonnet (the previous Claude) did it for me after 3 hours of trying everything else.
DPM++ 3M SDE sampler + the old Nova Pro 1 model combo actually completed for me last night!
Adobe RGB + Mistral Large 2 (the older large model) worked on the first try, weirdly.
BF16 precision with DeepSeek V3 from before the update finally got me a clean image back.
Gentle strength + the retired Llama 3.3 weights got mine through without erroring.
Old PNG-8 output (had to force it manually) with Gemini 3.2 worked when the new Gemini kept failing.
The retired Flash Attention 2 + Grok 4.6 combo cleared my image after a full afternoon of tests.
Aggressive strength with Sage Attention 1 (the original version) is what finally did it for me.
Balanced strength + the old Euler A sampler from before the Turbo update worked instantly.
GPT-5.6 combined with the retired FP32 precision mode fixed it for me after countless failed attempts. Hoping they bring FP32 back as an option!