Connect with us

Softwares

There are people spending a lot of money to get AI avatars on Lensa. You can do it for free and just as well with this tool

Avatar of Thomas Grimm

Published

on

1671379468 840 560 jpg

During the last months we have seen how the generation of images through artificial intelligence has had a great impact. Users have been experimenting with tools like DALL-E 2, Stable Diffusion, or Midjourney, among others, thus generating all kinds of photographs from the text description of the desired image.

This has been thanks to OpenAI’s GPT-3 language models, or Stable Diffusion’s, by the Stability AI team, and other collaborating studies. And it is that the precision of its models has allowed the development of tools that, with a good use of them, They allow us to generate amazing images of any element that we have in our heads. In addition, there are those who use these tools to generate amazing avatars, being able to put these images as profile photos on networks.

One of the applications that has had the most recognition when it comes to being able to generate avatars to add them as a profile image has been Lensa. This application is having great success among the public, especially for its precision and ease of creating the images, since all you have to do is upload between 10 and 20 images of yourself to generate avatars in multiple styles. The app is paid, although today we are going to show you a completely free method, and that allows you to train an AI with our images to obtain avatars of all kinds.

How to enter an AI with your own images

The language model that we are going to use is that of Stable Diffusion, and through the Dreambooth technique, we will be able to train this AI to add ourselves as a “concept”. In this way, when we ask you to make images of us, with a detailed description, the AI ​​will process the given information and show us the images that we have requested.

The procedure is much easier thanks to the help of DotCSV, a content creator who has created a document that is very easy to use and to use and run in the Google Collab tool. And it is that by entering this link, you just have to follow and execute the steps that appear in each tab. We summarize it below.

First of all, make sure you have a Google account, which your Google Drive has at least 4 GB of free storageand that you have an account on Hugging Face.


Once the Hugging Face account is created, press the icon of your profile that is in the upper right corner, and go to “Settings”. Once here, in the side menu we select “Access Tokens”, and we create a token with the name that we want. Of course, make sure to select write instead of “Read”, in order to be able to modify the Stable Diffusion model and train it with our face. Then, remember to copy the token to a safe place, since you will need it later.

In addition to this, you also need to select the images that you are going to use to train the AI. Keep in mind that the more you upload and the more varied they are, the better. In this case We recommend uploading between 15 and 20 images, with different backgrounds and clothes, and full length, half length and portrait. In this way we will help the AI ​​to understand that we are one more “concept” in its database, and not an individual image.

Once you have them, we will avoid the possible distortion of the images generated by the AI rendering them all at a resolution of 512 x 512 pixels. For this you can use any image editing program.

After performing these steps, we now go to the DotCSV Google Collab document. Already in the interface, we select “Runtime environment” and “Change type of execution environment” and we make sure that the option of “GPU”. If she was like this by default, we leave her as is. Then, we click on the “Play” button from step 1. Here Google Collab will assign us a graphics card remotely, so we don’t have to use the local resources of our PC.

Google Collab

When the process is finished and we have the green tick, we continue with step 1. Here it is enough to link our Google Drive account pressing the “play” button. After that, we continue with step 2, where we just execute it and it will download the libraries necessary for the process.

In step 3, we display the option, and where it says “Huggingface_Token” We add the token that the Hugging Face website generated for us. Once this is done, we leave the other options as they are and press the “play” button.

Advertisement

I've Created A Stunning 3D Avatar Using AI - Here's How You Can Create Your Own

Once we are in step 4, we must select the parameters to configure the training of the AI. If we want to obtain images of ourselves, the training subject has to be “Character”. However, we can choose between other types, depending on whether we want it to generate objects, animals, or other elements. Furthermore, in Subject_Type we will leave “person” in case we want to get our avatars.

The “Instance_Name” step is important, since we must put a name that the AI ​​does not confuse it with another character. In this way, you can refer to yourself through the nickname that you use in networks or similar. But yes, try to avoid full names, since it can make the AI ​​confuse with another person or celebrity (in case it matches).

When you have completed step 4 with all the details, we press the “Play” button and below in the same step a box will appear in which we can upload the images. Here we select the photos that we have ready (with 512 x 512 resolution) and upload them to the tool. When the process is complete, the green tick will appear in step 4.

cpt

Step 5 will try to find AI generated images that match the uploaded models. In this way, if you have uploaded images of yourself, the recommended dataset that we should use is that of person_ddim, so we select it and click on the “play” icon. When the process is finished, all that remains is to go to step 6, where here we just have to press the “play” button and the AI ​​training will begin.

The process will take around half an hour or 45 minutes, depending on the traffic of Google Collab and the graphics card assigned to us. When I finish, We will already have our AI model trained so that it generates images of ourselves.

We will have the file with the loaded model in our Google Drive in CKPT format. As an optional step, we can generate an interface from that model and generate the images from the GPU provided by Google Collab. This is achieved in step 7, where we just have to press “play” and wait for the process to finish.

Interface

When finished, it will generate a link in which we must select and It will take us to the graphical interface where we can enter the text to describe our image. Let’s remember that, if we want the AI ​​to generate an image of us in a certain style, we must start with something like “A photograph of [nombre dado]…”. The name is the one we gave it in “Instance_Name” so that the AI ​​would recognize us.

Advertisement
AI

To generate the image, we simply introduce the description with the name of our token and the style (the more detailed, and with more references to artists and artistic style, the better) and click on “Generate”. It is also recommended to increase the steps to 50, for greater detail and precision. As an example, you can use a description like the following:

full length photo of [nombre] as a warrior, highly detailed, 4 k, hdr, smooth, sharp focus, high resolution, award – winning photo

If you have done all the steps correctly, you can now use the tool to generate images. Note that every image you generate is automatically saved to your Google Drive folder, so you have to be careful with storage. However, if you want to run the tool locally and have a decent graphics card for it, you can download this interface and load your CKPT model from Google Drive into it.

Continue Reading
Advertisement
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Softwares

The latest Discord update is slowing down Nvidia graphics cards: there is already a partial solution

Avatar of Thomas Grimm

Published

on

1675265424 840 560

The Discord communication application has become essential on many computers and mobile phones to be able to call friends in a leisure environment, but also work. This tool has numerous advantages by facilitating much more fluid communication than that offered by Skype, but also by creating channels for different communities.

And although it may seem that everything is good, the truth is that Discord it also has problems that are serious. One of them is related to Nvidia graphics cards and a recent update that is causing a slowdown of the computer hardware.

The latest Discord update did more harm than good

The fact that we are talking about an application that is focused on gamers and has a bug that directly affects one of the components most required by this group, such as graphics, is undoubtedly paradoxical. But the truth is that it has been seen that this bug is affecting some Nvidia modelswhich includes the RTX 3080 and the RTX 3060 Ti.

Nvidia Graphics Cards 1000x600

This bug makes it impossible to get the most out of the GPU by slowing down the memory clock up to 200 MHz. At first it was the users themselves who reported the error through Reddit and also in the Linus Tech Tips forum. Faced with this wave of criticism, Nvidia has acknowledged the problem and has given details about its evolution.

This information has been given through a support article on the Nvidia website in which it is specified that if this type of slowdown is noticed, it will be possible to download a GeForce 3D profile manager and apply the fix. In the event that you prefer to wait, you will be able to choose to be patient and download the driver update that is being prepared.

We have tested the GeForce NOW service with RTX 4080 and it is a punch from NVIDIA on the table

in Xataka

We have tested the GeForce NOW service with RTX 4080 and it is a punch from NVIDIA on the table

Discord released this update to allow AV1 streaming with RTX 40 series cards on 4K quality and 60 FPS with Discord Nitro. But in the end this improvement has ended up causing the graphics of the previous generation to fail in this way.

Via | The Verge

Advertisement

(function() { window._JS_MODULES = window._JS_MODULES || {}; var headElement = document.getElementsByTagName(‘head’)[0]; if (_JS_MODULES.instagram) { var instagramScript = document.createElement(‘script’); instagramScript.src=”https://platform.instagram.com/en_US/embeds.js”; instagramScript.async = true; instagramScript.defer = true; headElement.appendChild(instagramScript); } })();


The news

The latest Discord update is slowing down Nvidia graphics cards: there is already a partial solution

was originally posted on

Genbeta

by José Alberto Lizana.

Continue Reading

Softwares

This AI that allows you to clone your voice (and anyone else’s) is filling the Internet with celebrities saying stupid things: this is how you can use it

Avatar of Thomas Grimm

Published

on

1675261259 840 560

An AI that allows anyone, and in a matter of seconds, to clone the voice of any other person (or, of course, oneself) is beginning to take the Internet by storm. The controversial but popular 4chan forum, without going any further, has been flooded with Audio deepfakes of celebrities like Emma Watson or Joe Rogan filled with violent or pornographic comments.

And all since, this past weekend, access was opened publicly to the ElevenLabs platform, perhaps one of the most significant examples of how far AI-based Technologys dedicated to audio cloning and synthesis have advanced.

Apart from the undeniable realism of its voice synthesis (light years away from canned voices like Loquendo, so popular years ago), the company itself boasts that the audio clips it generates require little or no editingwhich encourages its use in tasks such as dubbing or audiobook production:

When video deepfakes emerged, we all thought that an era of massive misinformation was inaugurated… and in the end its main application was the generation of porn. It remains to be seen what will happen to their audio equivalents

ZAO, the Chinese MOBILE APP that through DEEPFAKE turns you into DICAPRIO in SECONDS

Somebody put order here

Of course, the fact that it does not present none of the safeguards we are used to finding in imaging services to avoid creating certain types of content ‘sensitive’ and/or based on real people.

In a thread posted on Twitter last Monday, ElevenLabs recognized that some of its users were abusing its application: “[observamos] a growing number of cases of misuse of voice cloning”, and warned that they are able to “trace any generated audio back to the user”.

Likewise, it has put on the table the possibility of implementing protections such as the User identity verification or manual approval of each request voice cloning. However, its use remains unlimited for now.

Advertisement

we have tried it

When we register for free on the platform, we are granted access to a series of tools. First of all, we will find the simple voice synthesizer that allows us to convert texts to sound files using, in principle, predefined voices in American English.

However, if we make use of the ‘Voice Cloning’ function, we can upload audio with a voice recording to the platform (“it must last more than 1 minute and not contain background noise”, clarifies the website), which will allow us to create a personalized voice from which we can return to the synthesizer and create an audio based on our voice… or anyone else’s.

screenshot 39

Voice Cloning Step 1

screenshot 40

Step 2

That yes: the service is, for now, optimized for the American English accent; so it is possible that, when generating the audio, you do not finish recognizing yourself in the resulting file.

screenshot 41

The classics never die. Although it sounds horrible in American, let me tell you.

That has been our case, after generating a personalized voice based on that of a server: reading Don Quixote in a perfect Kansas accent (up state, down state) has not finished sounding ‘very me’, but US users claim not to suffer from this problem.

The free account only allows create up to five custom voices and synthesize to audio 10,000 characters of text per monthIn addition, the resulting audios are subject to a non-commercial license. There are, from there, several payment options, the cheapest being $22 per month.

Via | The Verge

Image | Based on an original by Jason Rosewell on Unsplash

Advertisement
Continue Reading

Softwares

The Spanish-speaking audience does not stop growing on Twitch. But we’re still so underrated

Avatar of Thomas Grimm

Published

on

1675258875 840 560

It is undeniable that Twitch has undergone a great boom in recent years accompanied by the strategies that They have followed different content creators who have attracted a large audience to the platform. But the problem that resides is the great differences that can exist between the English and Spanish-speaking communities, mainly due to the differences in investment by advertisers.

But the great reality is that despite this difference in investment and prestige, the truth is that the Spanish-speaking community is one of the most powerful in terms of growth in the number of viewers and also in number of creators. And this is something that was recently evidenced at the ESLAND Awards where the community gave a big blow to the table.

Getting started on TWITCH How to set up OBS and PERSONALIZE your CHANNEL

The Spanish-speaking community is growing at a fast pace

One of the examples that Bloomberg has chosen to express this large Spanish-speaking community is that of Rivers. A streamer who studied engineering in Mexico in 2020 when she started playing Call of Duty on Twitch motivated by her friends. And as they went live, more and more people began to see this girl who wondered if she should spend more time being a streamer and less time studying. Obviously, this is something that she did not like in her family, although in the end it seems that she made the right decision to currently have a live audience of 19,000 viewers on average.


And this is motivated by a substantial increase in interest in the transmission of the Spanish-speaking public. Specifically, between 2019 and 2022, the audience in Spanish multiplied, going from the seventh category to being very close to English in a second position, as Bloomberg has detailed.

These are all the winners of the ESLAND 2023 awards: Ibai conquers Twitch and is proclaimed 'streamer of the year' for the second time

But it is not something that is coincidental, but rather it is due precisely to the fact that Latin America has about 80% of people between the ages of 10 and 65 play video games. But the problem in this case is that Latin America does not receive much attention from investors, and it is that it is at the bottom when it comes to investment in digital advertising. It is even believed that this market is considered ‘incipient’. Specifically, the country as a whole is located behind Africa and the Middle East.

Income

Given this boom, from Twitch they have increased their presence in the region with more staff and adding marketing teams, creative partners and sales. But this is something that can be very sad, since currently in the record for viewers on the entire platform, Spanish streamers stand out with Ibai and its Evening of the Year II, and also the successes that many creators in Latin America are reaping. This confirms that Spanish is booming on the social network, although it is not being given the recognition it deserves.

Source | Bloomberg

Advertisement
Continue Reading

Trending