If you’re looking for ultimate control over the output of your text-to-image generator tool and have a background in machine learning, this Pixray guide is for you.

Even if you’re someone like me who doesn’t understand programming but can fare well with prompts, Pixray can prove a worthy contender to some of the most popular image AI tools currently in the market.

Pixray’s algorithm is designed for use by just about anyone and has many amazing capabilities that other tools don’t readily offer. In this guide, you’ll learn everything you need to get started with Pixray.

Disclosure: This post may contain affiliate links, meaning we get a commission if you decide to make a purchase through links on this site, at no cost to you.

What is Pixray & How Does It Work?

Pixray is an image generation system that utilizes advanced techniques to create images from text prompts. It runs predictions on Nvidia T4 GPU hardware. There are already many other text-to-image AIs online, but Pixray stands out due to its integration with GitHub.

With Replicate, you can effortlessly run complex machine-learning models with a few lines of code. You can also access a community of open-source models or run your own without setting up servers. 

Note: You can only use Pixray when you sign up to replicate using your GitHub account. If you don’t have a GitHub account, you can create one image for free.

Pixray features

Using vqgan drawer in Pixray

From the above screenshot, you can see how friendly the interface of Pixray is. It is easy to navigate and understand, even for a first-time user. Let’s dive into these features.

Input

This is where all the details you want from your art are input.

  • Prompts: This is where your text prompts are included.
  • Drawer: This is the render engine, containing seven image renders, including pixel, line_sketch, and clipdraw. 

The type of drawer determines the technology responsible for generating the images. VQGAN, derived from taming-transformers, is used by vqgan, while clipdraw and line_sketch utilize diffvg. On the other hand, Pixel employs the original pixel drawer from Dribnet. 

Regarding style, pixel generates pixel art, vqgan produces GAN-generated images, and clipdraw/line_sketch generates stroke-based images.

  • Settings: This isn’t like the usual setting you would expect a tool to use; this is a text-based setting where you type in what you want to format. 

Output 

This is where your images are displayed. The output contains the generated image and a horizontal ruler pattern directly underneath. This ruler pattern indicates the time taken for the image generation and can be modified by the user. As you make adjustments, you’ll notice that the image quality decreases when the time is shortened and improves when the time is lengthened.

From the output area, Pixray allows users to share the generated image via a URL that is automatically copied to their clipboard when they click on share, download their images directly to their computer in Zip format by clicking on download, and report their image when it seems broken or inappropriate so it can be investigated and appropriate actions can be taken.

Runtime

Images usually take about 5 minutes to generate but might vary depending on the input and complexity. This is quite long compared to many other text-to-image AIs I’ve tested.

Aside from text-to-image generation, Pixray also functions as a Python library and a command line utility that you can explore, especially if you have machine language experience. Underneath every generated image, you will see a small “show logs” button that shows the code behind your generated image. If you’re a budding programmer just getting his feet wet with Python, you can learn a lot from these coding snippets.

Pixray offers resources that might take some time to digest properly but are important for the full utilization of all the tool’s capabilities. You can browse these resources to go deeper into Pixray:

GitHub

Demo Notebooks

Discord Community

Some helpful documentation

Getting Started with Pixray

As a firm believer in “show; don’t tell,” let’s first take a look at how remarkable Pixray is with practical examples.

A screenshot of image generated on Pixray using the input: Jungle with monkeys on trees and drawer: Pixel
Using the pixel drawer in Pixray. Prompt: A jungle with monkeys on trees

I started off with a relatively simple input: “Jungle with monkeys on trees.” I wasn’t impressed with the output at all. It was a cartoon-like image with poor graphics when I expected something more realistic.

I guessed it was from the pixel style I chose from the drawer, so I decided to play around with the different styles and see what I could get.

A screenshot of image generated on Pixray using the input: Jungle with monkeys on trees and drawer: VQGAN
Using the vqgan drawer in Pixray. Prompt: A jungle with monkeys on trees

I tried out VQGAN on this one to see how my output would differ from the first, and I still got a cartoon-like image, but it looked a bit clearer than the first image, which was distorted.

A screenshot of image generated on Pixray using the input: Jungle with monkeys on trees and drawer: Clipdraw
Using the clipdraw drawer in Pixray. Prompt: A jungle with monkeys on trees

I tried the same input with clipdraw to see what Pixray could deliver, and I got this. I wasn’t quite sure how to feel about this one. The outputs aren’t satisfying.

I’d like to try line_sketch drawer, too. I’m not expecting much because line sketches are supposed to be outlines of an image and aren’t the most realistic representation of an object.

I wasn’t wrong. This is mediocre — even for an outline.

Using the line_sketch drawer in Pixray. Prompt: A jungle with monkeys on trees

Pixray Limitations

Requires Technical Knowledge: To maximize the full potential of Pixray, you must thoroughly review its resources; some machine language knowledge is also a plus. Casual prompts won’t produce high-quality images. Users must also provide proper instructions in the settings to get decent image quality.

Poor image quality: Unlike many other AI text-to-image algorithms, Pixray produces a poorer image quality without proper instructions. I also found that it does a bad job with predictions.

Long prediction time: Pixray takes about 5 minutes to generate an image. This is considerably longer than other tools, and the quality of output received afterwards is not worth the time investment.

Pixray Pricing

Pixray pricing

My Pixray Review

First things first, I wasn’t impressed with the quality I got from Pixray. The images displayed on their website looked promising, so I couldn’t help but wonder why my outputs were terrible.

I clicked on the images to see how their inputs/prompts differed from mine, and I realized that their settings were much more descriptive than mine. But if you’re someone like me who prefers using plain English to create images, be prepared to be challenged with Pixray. It requires you to brush up on some image terminology so that you can produce images of greater quality with full control.

A screenshot of an image generated by pixray
The log shows the specifications of this image

This is an example of the type of input used.

A screenshot of an image generated by pixray
A log showing the input used by another user to create this image

Here is another beautiful image with technical inputs.

Exploring Image Generation Algorithms on Replicate

Pixray is one of the numerous image generation systems on Replicate, so since I wasn’t getting the results I required, I did a little digging around and found some amazing algorithms that don’t take up to 5 minutes of run time like Pixray but produce better results.

A screenshot of Replicate's explore page
Other ai algorithms available on Replicate

Once you sign up for the platform, you are directed to Pixray. Click on the “Explore” button on the dashboard to find a wide range of amazing algorithms.

I immediately wanted to see how it could do things differently with the monkeys, so I would start with it.

Image generated on Pixray using the Algorithm: AI-forever/Kandinsky-2, Input: A forest with monkeys on the tree
Algorithm: AI-forever/Kandinsky-2, Input: A forest with monkeys on the tree

This came out a whole lot better than what I previously created, so it’s worth checking out.

Image generated on Pixray using the Algorithm: Stability-AI/ Stable-diffusion, Input: Backdrop with pink flower
Algorithm: Stability-AI/ Stable-diffusion, Input: Backdrop with pink flower

I absolutely love this; it came out better than I had imagined.

Image generated on Pixray using the Algorithm: Cjwbw/Stable-diffusion-high-resolution, Input: Medusa
Algorithm: Cjwbw/Stable-diffusion-high-resolution, Input: Medusa
Image generated on Pixray using the Algorithm: Nitrosocke/archer-diffusion Input: Japanese fighting scene
Algorithm: Nitrosocke/archer-diffusion Input: Japanese fighting scene
Image generated on Pixray using the Algorithm: Sczhou/codeformer
Algorithm: Sczhou/codeformer

This is a face restoration algorithm for correcting the faces in old photos, and producing higher quality AI-generated images.

Image generated on Pixray using the Algorithm: Mcai/absolutebeauty-v1.0-img2img–This algorithm is meant for beautification.
Algorithm: Mcai/absolutebeauty-v1.0-img2img–This algorithm is meant for beautification.

Replicate features other algorithms beyond images, like:

Text generated on Pixray using the Algorithm: Replicate/Vicuna-13b—A language model created based on ChatGPT interactions
Algorithm: Replicate/Vicuna-13b—A language model created based on ChatGPT interactions.
Text generated on Pixray using the Algorithm: Suno-AI/bark–A text-to-audio model
Algorithm: Suno-AI/bark–A text-to-audio model

Pixray Alternatives

I wouldn’t necessarily recommend Pixray to anyone who lacks knowledge of machine language (or someone who’s willing to learn basic lines of code with their guides.)

I did enjoy experimenting with Replicate’s other algorithms, including the sound AI output. I found those satisfactory, so they may be worth a try.

If you want something that gets the job done at a somewhat similar quality as Pixray minus the learning curve, you’d definitely want to check these out:

Looking to become a better AI artist? Check out these guides to take your output to the next level: