AI image generation: five free services

Intermediate

Intermediate

How neural networks generate images

Neural networks are a form of machine learning and underpin deep-learning algorithms. They consist of neurons that receive, process and return information. Neurons are connected by synapses.

AI models rely on training data to identify patterns and solve tasks such as recognising or generating images.

For example, when searching for similar pictures with Google Lens, the system also uses a neural network. It looks for similarities with other images in its database.

By a similar principle, algorithms generate illustrations from text prompts. A user inputs a request in natural language and the network combines elements it has learned.

Some models cannot create images from text, but can apply visual effects to existing pictures, stylise them after famous artists or alter selected objects.

The potential of such systems seems vast, particularly in art. In 2022 image generators gained broad popularity.

ForkLog has selected five free resources that let anyone test a neural network online and create unique images.

Bing Image Creator

Microsoft’s service is based on the DALL-E neural network. In March 2023 the tech giant launched a standalone site for the tool and integrated it into Bing Chat.

In October 2022 the service was publicly released, allowing anyone to sign up and use it. The system now generates more than 2 million images a day. By contrast, creating a DALL-E 2 profile requires a phone number from a limited list of countries.

In April the AI generator appeared in the Edge browser. It is available from the sidebar to users worldwide.

To use the service, open the corresponding browser pane, enter a prompt and wait for the result. Image Creator will produce four options, which can be downloaded.

To prevent misuse for toxic content, Microsoft has implemented its own safeguards in addition to OpenAI’s methods.

The company also makes clear that images are AI-generated and marks results with watermarks.

The tool gives users 25 boosts a week to speed up generation. One token is debited per request.

After boosts are exhausted, the company offers to obtain more via Microsoft Rewards points or wait for a weekly refresh.

Dream by WOMBO

The web service lets you create images free and without registration, though there is a Premium tier. Mobile apps are available for Android and iOS.

To generate an image, enter a text prompt up to 200 characters and select a visual style. After 10–20 seconds the network returns a result at 960×1568 pixels.

Finished images can be downloaded or published to the site’s feed.

The “Edit with text” option lets you modify a generated image using a prompt. You can redraw an image, add or remove an element, restyle an object or invert the illustration.

The feature is in beta and its accuracy is fairly low.

In the free version you can edit an image only twice. After that you must reupload the image or modify the new one.

The service also allows you to stylise your own image by request.

Premium costs $9.99 a month or $89.99 a year. A one-off payment of $169.99 is also offered.

Stable Diffusion

A text-to-image generator created by Stability AI. In September 2022 the service became available to the general public.

The model was trained to create images from any text prompt, including pictures with public figures and nude people. However, in November 2022 the company updated the algorithm, weakening its ability to create NSFW content and illustrations in the style of specific artists.

The network draws for free.

Users simply enter a text prompt of any length. The model then generates four illustrations at 512×512 pixels.

Lexica Aperture

In December 2022 Sharif Shameem, the founder of Lexica, a search engine for images and AI prompts, presented the Lexica Aperture image generator. The algorithm can create photorealistic illustrations from text prompts.

The network is available for free after signing in via a simple web interface. It creates images in portrait or landscape orientation at 768×1152 pixels.

Below the prompt field you can specify a negative prompt—what should not appear in the picture. The service also lets you upload a reference image.

The finished image can be expanded or used to create variations.

Deep Dream Generator

A platform for creating and editing images. The service requires account registration and has limits on free use.

The project’s site offers three modes.

Text 2 Dream

Creates images from a text description. The user should craft a prompt or click “random prompt.”

The generator is relatively fast, but sometimes falls short of DALL-E 2 or Stable Diffusion in its interpretation of words and the resulting images.

Deep Dream

This mode performs deep processing of a source image using AI algorithms. Upload a picture and set parameters: initial depth, AI enhancement and neural-network layer.

Deep Style

In this mode the user can stylise an existing image. They upload a picture or photo and select a reference illustration from the list.

As for free-tier limits, upon registration the user receives 30 units of energy. They are consumed when using the tools: five points per request in Deep Style or Text 2 Dream, and two points per request in Deep Dream.

Energy can be purchased. The cheapest plan costs $19 per month.

DALL-E 2

In January 2021 OpenAI sparked a revolution in AI by presenting DALL-E, a text-to-image generator.

It is a transformer built on the large language model GPT-3 with 12 billion parameters and trained on text–image pairs.

The network generates images at 256×256 pixels. It offers a wide range of capabilities, including creating anthropomorphic animals and objects, realistically combining unrelated concepts, rendering text and applying transformations to existing illustrations.

In April 2022 the company released the second version, which generates more realistic images with lower latency at 1024×1024 pixels.

DALL-E 2 lets you select and edit specific areas of existing images, add or remove elements along with shadows, create collages and produce variations.

The new version also offers Outpanting. It lets you expand an image using text prompts. When adding new objects, the system accounts for existing visual elements such as shadows, reflections and textures.

In October 2022 OpenAI closed the waitlist for the beta version of the image generator.

The network is available in all countries except Afghanistan, Belarus, Venezuela, Iran, China and Russia.

In February 2023 it became known that Ukraine was removed from the list of countries where the company’s services are blocked. However, OpenAI’s algorithms do not work in territories temporarily occupied by Russia.

Users in regions where the technology is unavailable can interact with DALL-E 2 only if they have a VPN service and an active phone number from a country open to OpenAI products.

Registration in DALL-E 2 is free. Image generation requires credits, with one deducted for each text prompt.

Previously, new users were given free attempts to create images. Recently this changed.

The network became paid for accounts registered after April 6, 2023. 115 attempts cost $15.

Users who created an account before that date can still use the network for free. They receive 50 credits, which after a month decrease to 15. The number of attempts renews every 30 days.

Various companies and well-known brands use DALL-E 2. Cosmopolitan and The Economist used the network to generate covers. Microsoft launched its own application, Designer, based on the service, which can create content such as text, ad banners, greeting cards and logos.