BEST & Open Source AI Generators for Anime pictures
Over the last year, AI has become a trend, where businesses want to know more about it in order to exploit its potential to increase productivity, especially Generative AI.
In this article, we will explore the best open-source AI generators that are available on the market and compare their performance in generating animated images.
We will explore :
- Yodayo
- Happy Accidents
- StarryAI
- Stable Diffusion web UI
In order to carry out this experiment, we will take a single image generation request that we will use on the different websites.
Prompt: “2 girls, long black hair, green eyes, wearing a white shirt, jeans, city, happy, high res, 8K, looking at the viewer, masterpiece, perfect face”
Negative Prompt: “Extra fingers, fewer fingers, extra legs, extra eyes, low quality, bad anatomy, bad composition, inaccurate eyes, watermark, copyright, logo, NSFW, fat, ugly, clothing cutout ”
1/ Yodayo
Let’s start with the first Yodayo, It is necessary to create an account and then click on + New Artwork.
You will have an interface where you can enter your Prompt and also include a Negative Prompt like extra fingers.
If you make your image generation a High priority it will take 5 YoBeans to generate (you start with a free 300 YoBeans), You can also generate for free by choosing not to toggle the High Priority option.
You also get 150 YoBeans daily free claimable!
Now we will use the Prompt and the Negative Prompt and start generating the picture, You can also download the picture and even enhance it.
You can also have more customization by choosing the models and configuring some parameters, You need to click on the “Gear” icon next to Generate.
The sampling Methods are the denoising process (determines the balance between noise reduction and the preservation of image details), Steps are how many iterations you want the AI to run before generating the result, and Scale guides the AI to know how much the model should follow the Prompt (The higher would be more faithful to the text while a lower would have more artistic liberty).
Spells give specific details of a character or style that you desire. You also can add Seed to make image generation deterministic (If you put -1, it will simply start with the random seed). Let’s generate a few more examples, very good quality!
Happy Accidents
Same steps, you need to sign up for an account, choose a username, then click on Create located on the sidebar.
You can choose a model and some parameters like we saw in the previous tool. Note: You have 100 fast image generation per month
We need to choose a model related to anime generated image, You can choose from multiple models present on the website (Image below)
I will choose Mix-Pro-V4, Let’s use the same Prompt and Negative Prompt and start generating the images.
Amazing results! We can also use images generated from Yodayo website to generate new images.
Image Strength determines how close Stable Diffusion will stick to the provided picture (Higher values will instruct the A.I. to stick closer to the original image. Lower values will tell the A.I. to optimize for aesthetical beauty).
StarryAI
Very easy to use, but you have a small pool of image generation since you have 5 Credits a day and each image generation can be around 1 to 2 credits. As with previous tools, you need to create an account first, then select a style. I will choose Art.
Then we will use our Prompt and active the option “Remove from Image“ to be available to activate the Negative Prompt.
Note : Before generating the image, you need to choose the style you want.
Image generation takes time and you can generate few images compared to the other tools that we saw previously but It is very simple to use and from 1 image generation you can have 1 result and 4 variants. You can see the result below, very good performance.
Stable Diffusion web UI: It’s a web interface based on the Gradio library for Stable Diffusion, You can install it locally or run it on the collab, Here is the link to learn more about it.
We will use collab in this article, but before that, you can visit CIVITAI to get an idea about the model you want to use.
On the search bar, you can write the model type, make sure to filter the result to only show Checkpoint (the base model).
If you are running the model locally you need to download Checkpoint and put it in models/Stable-diffusion (image below)
Let’s start running it on Colab. The first thing to do is to change the runtime to GPU (image below)
Link to the Colab HERE
You will just need to press the play button on each section. The first thing to do is to make sure that we have access to our Google Drive by clicking on the play icon in the first bloc.
Then you will copy the repo of Automatic 1111 and install all the requirements by clicking on the next 2 blocks
You can download the model or use the model provided on Colab. If you already have a model you can put the path in the Path to Model section.
Once you have connected your Google Drive, you can access the files in it and you will find a new folder named “sd”.
You will navigate to the “sd” folder, then you will click on ‘stabledifusion-webui’ and choose ‘models’.
In this folder you will find the models, Click below and find the “Stable-diffusion” folder, you can put your own models there and import it later.
On the Control Net (a neural network that controls a pre-trained image Diffusion mode) choose a model.
Note : You can download multiple ones, but don’t forget that you are limited in space working with Colab.
You also have LoRA(a training technique that is used for fine-tuning Stable Diffusion models). You can leave it as it is (optional)
You can enter a username and a password to access your stable diffusion web interface for security purposes, You can of course leave it as it is and just click on the play button.
After clicking the play button, it will start downloading and then give you a public URL, Click on it to access your web interface
You can start generating images now on the cloud on your computer if it is installed locally
Let’s explore some parameters that you need to know:
- Stable Diffusion Checkpoint allows you to choose the model you want to use
- You have several tabs Let’s focus on the 2 most important, the first one txt2img (generating images from text), and the second one is img2img (providing a reference to generate another image)
- Sampling method and Sampling steps, we already saw that in the previous tools ( denoising process and the number of iterations)
- You can also choose the Width and the Height of the image.
- You can increase the image generation by increasing the Batch size.
- We already saw the Seed, it determines the starting point of the image generation, and you can keep it deterministic or random. When you generate an image you will have the seed of each one (image below). You can copy the Seed and run the Prompt again to see a similar output inspired by the previous result.
- Note: You can use the recycle icon to use the Seed from the last generation.
- CFG Scale (Classifier free guidance); controls how much the model should respect your prompt, If you keep it high then it will give more weight to the Prompt.
Easiest way
If you are having trouble with that, you can use this easiest solution, you need to go to this Github Here. Each of the Python files you find (.ipynb) is a model that you can run with Colab.
You can search the Model you want on Civitai and Look for it in Github. This is the model I used.
You need then to copy the link and look for it on Colab from the Github; navigate to the Github tab then past the link that you copied before, and click on the search bar.
You will then have a Colab window, you only need to click on play once (first block) and then as we saw previously, click on the link provided (after downloading and finalizing everything). EASY!
The article is coming to an end, The difference between the services offered on the websites is mainly the number of points for image generation, the available models, as well as the image generation parameters.
If you are looking for an easy solution, all 3 are excellent. If you are looking for a complete solution without limitation that you can even install on your machine I advise you to use Stable Diffusion.
Please consider the following and subscribe to the newsletter for more articles about business, data science, machine learning, and extended reality, It’s FREE!
You can find my lists in the following links :
Statistics & Data Science: https://medium.com/@soulawalid/list/statistics-data-science-65305693779d
Business: https://medium.com/@soulawalid/list/business-1528f08575a7
Quantum Machine Learning: https://medium.com/@soulawalid/list/qml-be0b06f7a986
Extended Reality: https://medium.com/@soulawalid/list/extended-reality-bf03607b0b80
Neuromarketing: https://medium.com/@soulawalid/list/neuromarketing-8f94149e3c73
If you have any questions, you can ask me on Linkedin, Here is my profile: https://www.linkedin.com/in/oualid-soula/ Let’s connect!
Please consider the following and subscribe to the newsletter for more articles about business, data science, machine learning, and extended reality, It’s FREE!
