DALL-E: What are AI image generators and how can they be used?
By teamnext Editorial Team
DALL-E is an artificial intelligence based system that generates digital images from text input. Tools of this kind are referred to as AI image generators.
This article explains how DALL-E works, how AI image generators can be used in practice, which alternatives exist and which copyright aspects need to be considered.
What is DALL-E?
The name DALL-E, stylized as DALL·E, references the Disney character WALL-E and the artist Salvador Dalí. DALL-E 2 refers to the second version of the system, which is the focus of this article.
DALL-E is based on artificial intelligence. More precisely, it relies on machine learning.
The underlying approach uses artificial neural networks and specifically Deep Learning. Between input and output layers, multiple hidden layers process the data. This is why Deep Learning can be described as multi-layered learning.
Who developed DALL-E?
DALL-E was developed by the US company OpenAI. Notable supporters include Microsoft and Elon Musk. OpenAI is also known for the text generation system ChatGPT.
Both systems are based on GPT-3.
What is GPT?
GPT stands for Generative Pre-trained Transformer.
A transformer is a method in machine learning that translates one sequence of symbols into another. In practical terms, GPT is a pre-trained language model.
The training process used a very large dataset of internet text, comprising approximately 500 billion words.
How does DALL-E work?
DALL-E uses Deep Learning to convert text input into pixel-based image output.
The model was trained using around 650 million text-image pairs. These consist of images with captions or relevant tags.
Key characteristics:
• precise prompts lead to accurate results
• generated images are unique
• multiple visual styles are supported
How can DALL-E be used?
DALL-E is accessed via the OpenAI website. After registration, users receive a free starting quota of 50 credits.
• 1 credit generates 4 images
• 15 credits are added monthly
• additional credits can be purchased
First steps with DALL-E
Example prompt:
“sliced air-dried sausage with bread and butter, photo”
The generated image resembles a realistic photograph of sliced cured sausage with bread and butter.
English prompts currently produce the most reliable results. German input is supported but may be less precise.
Each prompt generates four image variations. Selected results can be used to create further variations.
Unlocking DALL-E’s core capability
The real strength of DALL-E lies in creating compositions without direct real-world templates.
Effective prompts meet three criteria:
• precise and unambiguous wording
• creative combination of uncommon elements
• specification of medium or technique at the end
Common techniques include photo, digital art and 3D rendering.
Example 2
Prompt:
“copper statue of herkules drinking beer, digital art”
The resulting image shows a stylized copper statue of Hercules drinking beer. The colored squares indicate DALL-E branding.
The output appears conceptual, although it is purely calculation-based.
Every image is unique
DALL-E generates new images with every run, even when using the same prompt.
This is due to a changing seed value that influences the random number generation process.
Identical images would only be possible with a fixed seed and unchanged model. This option is currently unavailable.
Supported resolutions include:
• 1024 x 1024 pixels
• 1792 x 1024 pixels
• 1024 x 1792 pixels
DALL-E and copyright
DALL-E itself cannot hold copyright. Copyright applies only to human creators.
Despite uniqueness, generated images may still infringe rights if they include:
• protected brands
• copyrighted characters
• styles of living artists
Legal review remains necessary.
Training data challenges
It remains unclear whether training AI models with copyrighted material is legally permissible.
No court rulings exist yet. Platforms such as “Have I Been Trained?” allow artists to identify and remove their works from certain datasets.
So far, only Stable Diffusion has publicly committed to respecting such requests.
Alternatives to DALL-E
Stable Diffusion
• open source
• developed by research institutions and organizations
• transparent training data approach
Craiyon
• formerly DALL-E mini
• based on DALL-E 1
• limited free usage
• paid plans required to remove watermarks
Midjourney
• commercial project
• accessible only via Discord
• targeted at technically experienced users
Looking ahead
AI image generators are already reshaping media production.
Typical use cases include:
• blogs
• news platforms
• marketing content
They reduce licensing costs while enabling custom visuals.
Risks include:
• misinformation
• realistic image manipulation
Creative professions will evolve. Illustrators may increasingly refine and finalize AI-generated drafts.
A new role is emerging: AI prompt specialist.
Conclusion
DALL-E and similar systems are still developing, but their impact is already significant.
In the future, more images may be generated by AI than captured by cameras. Distinguishing real from synthetic content will become increasingly difficult.
Early understanding of this technology is essential.