Silueta.me is a background segmentation tool optimized for selfie and portrait photos.
Select your image and get a hi-res PNG image with alpha channel in seconds
with no additional user interaction, and obviously, for free.
We developed it based on latest research on Convolution Neural Networks (CNNs ) with our machine learning model trained specifically for best results with selfie and portrait photos.
Your images are processed in real-time.
We NEVER store them, not even temporarily.
Max input size: 2048x2048px
WebP, PNG or JPEG - 50Mb
Status: Starting - Server Load: Low
Choose a portrait/selfie photo for best results.
Your images are processed in real-time. We NEVER store them, not even temporarily.
Max input size: 2048x2048px - WebP, PNG or JPEG - 50Mb
Silueta.me works on magic.
At least what seems like magic to us, the latest cutting edge Machine Learning algorithms aimed at computer vision.
The background/foreground segmentation has been an obsession for us for many years, from the time where chroma keying and optical flow were the only options to the latest golden age of machine learning 'magic'.
Some years ago we achieved nice results with a combination of 'early days' ML segmentation models, trimap generation, refining using CPU intensive methods like KNN matting and color postprocessing. The field is advancing at a crazy speed so we decided to update the tech under the hood and showcase it with this tool.
In the core of the system, we are using an implementation of U2Net, please check their amazing work in their research paper. We fine-tuned their model to suit our needs, specially hardware constraints, and optimized the end result applying matting postprocessing like Closed-Form alpha matting
Our model (only 43Mb vs 170 Mb) has been shared with the authors of U2Net in order to return some value to the awesome visual computing community.
We also have some additional preprocessing and postprocessing optimizations turned off until we manage to deploy in more powerfull hardware. As of today the main computing of this tool is working on a free Heroku dyno, limited to CPU computing and only 512Mb of RAM. If the project gets enough traction we will hopefully launch into dedicated custom servers.
Enjoy Silueta.me and feel free to contact us with any questions about the technology stack, as well as any special requests or help with your specific use case.
Silhouette.me is a free service to remove the background of any photo. It works 100% automatically: You don't have to manually select the background/foreground layers to separate them - just select your image and instantly download the result image with the background removed!
Yes, Silueta.me is 100% free.
We show and ad in the seconds it is processing to try to cover for hosting costs.
We use sophisticated AI technology to detect foreground layers and separate them from the background. To improve the results we have several additional algorithms in place to improve fine details and prevent color contamination.
Check the 'Technology' section for more details.
The AI is trained to detect persons as foreground and everything else as background. That's why it works best if there is at least one person in the image. We are currently working on support for other kind of images as well, such as animals, product images, etc.
If you have a special request or want to tell us about your use case, we would love to hear about it!
You can upload images of any resolution. For performance reasons the mask is generated at 512x512px and applied to the source image. In order not to fry our server nor your device (ie: mobile) the image you upload is scaled so the end alpha image has a maximum resolution of 2048x2048px.
Yes, Silueta.me is 100% free. If the project gets enough interest we will offer premium paid tiers for professional users, mainly to pay for better computing hardware/hosting. Premium tiers would include more refined masks and resolution as well as an API and plugin integration in professional edition tools.
Not yet, but we will make one if the project get's enough traction. We are professionals ourselves and we know you expect only the best results when you pay for a service, but in order to cover the hardware costs it is important to know that we can monetize this tool before scaling it further.
No, we do not use your data to train our AI. We only remove the background and give you the result, that's all. If, in the future, we would ever consider using image uploads for training purposes we would be very clear about it and explicitly ask for your permission in advance. We take privacy VERY seriously.
We have worked in visual computing for many years, and have developed ad hoc solutions for many clients. Also we are in love with AI and Machine Learning and have accumulated some valuable knowledge throughout the years. We thought it would be cool to make this tool to showcase what we can do.
(15secs max depending on complexity of the image)