Elena' s AI Blog

OpenAI's Model Show-off

19 Feb 2024 / 14 minutes to read

Elena Daehnhardt

AI Models Show, Midjourney Nov 2023
I am still working on this post, which is mostly complete. Thanks for your visit!


The rapid evolution of AI enables us to be more productive, make faster decisions, and boost creativity, with the promise of generative AI being genuinely fantastic!

The latest development from OpenAI is Sora, their text-to-video model. It can generate high-quality videos up to a minute long based on user prompts.

Sora creates intricate scenes with multiple characters, specific movements, and accurate details of subjects and backgrounds. It understands the user’s prompt and can simulate the physical world to a certain extent.

The model may struggle with accurately creating complex scenes, specific cause-effect instances, and spatial details [1]. It may also have difficulty describing events that take place over time [1]

Only a few users, such as visual artists, have access to OpenAI Sora now. However, you can find examples of how to create videos from text at Sora web page.

In this post, we will discuss the technology behind Sora and briefly recap several other OpenAI models that are now available to everyone.


Our new model Sora can create videos from text and image inputs, but it can also transform styles and environments from a video input. What should we make with Sora next? #madewithSora #Sora #openai

♬ Divergent - HVRDVR

Sora, technical report

The key points of the OpenAI’s Sora model are explained in the research report, Video generation models as world simulators.

The report discusses the method developed by OpenAI to convert different types of visual data into a consistent representation. This method enables the training of generative models on a large scale. As a result, Sora (the generative model) can produce videos and images of various sizes and resolutions. It can even generate high-definition videos up to one minute in length.

Sora’s design draws inspiration from large language models that use tokens to unify diverse text modalities. Similarly, OpenAI compresses videos into a low-dimensional vector space by converting them to visual patches. For more details, refer to their report

Sora undergoes training and produces compressed videos in a latent space that is both temporal and spatial. A decoder model is used to map these generated latents back to pixel space to create output videos. Additionally, the decoder model is enhanced with GPT user prompts to improve the quality of the results. You can refer to 2 for more information.

OpenAI models

GPT models are machine learning algorithms trained on large amounts of text data such as Wikipedia articles or books. This training allows them to understand and generate language similar to human language. These models use transformers to process the text data and generate new text.

GPT models by OpenAI are transforming NLP and delivering impressive results in language tasks such as translations, summarization, and sentiment analysis. These models can also produce images, videos, and audio.

OpenAI’s models, such as those based on the GPT (Generative Pre-trained Transformer) architecture, undergo a two-step process of pre-training and fine-tuning.

During pre-training, the model is exposed to a vast amount of diverse text data from the internet, enabling it to learn the language’s patterns, structures, and relationships. By predicting the next word in a sentence based on the context of the preceding words, the model becomes an expert in capturing a broad understanding of the complexities of human language, including grammar, syntax, and semantic relationships between words.

After pre-training, the model is further fine-tuned on specific tasks or domains to make it more useful for particular applications. Here, the model is trained on a carefully curated, narrower dataset to excel in tasks like translation, summarisation, question-answering, or code generation. Fine-tuning the pre-trained model makes it a versatile tool for various natural language processing tasks.

AI models are trained on vast amounts of internet text to understand language rules and nuances and then given specialised training to become proficient in specific tasks. This results in models that generate coherent and contextually relevant text for various applications.

API usage and access

OpenAI’s API allows developers to integrate GPT-3 and other language models into their apps for natural language understanding, text generation, code completion, and more.

API keys

To use OpenAI commercially, sign up for the API, obtain an API key, and follow OpenAI’s usage policies. Pricing is based on tokens processed, with different tiers for different usage levels.

Usage and costs

Model pricing depends on window size. In sequential data processing, window size refers to the context length the model considers. For example, a language model with a window size of 10 considers the previous 10 words when predicting the next word. The total cost calculation is based on the number of tokens processed. See their current Pricing.

You will get some credit tokens that will expire after 3 months. You can also experiment with OpenAI models in the OpenAI Playground and GPT Playgrounds at gpt3demo.com.

Rate limits are also imposed to prevent misuse.

To see your usage costs for fine-tuning or other jobs, check your account.


At this moment, OpenAI models provide several application scenarios and much more:

  1. Text-completion models that are considered as legacy models include gpt-3.5-turbo-instruct, babbage-002, davinci-002. However, for better results, it is more useful to use Chat Completions API with newer models such as gpt-4, and gpt-3.5-turbo.
  2. GPT-4 and GPT-4 turbo are advanced large models, which can understand and generate natural language or code, as well as accept image inputs and emit text outputs. It is also multilingual.
  3. GPT-4 with Wisper models can transcribe audio into text, summarize clips, extract keywords, and generate captions. OpenAI has a great tutorial about Creating an automated meeting minutes generator with Whisper and GPT-4. Please notice that Wisper also has an Open-source implementation in https://github.com/openai/whisper.
  4. Text-to-speach models generate outstanding speech records from text inputs.
  5. GPT-4V model is useful for analysing images and answering questions about them.
  6. Image creation, image editing and image variations generations are enabled by dall-e-3, dall-e-2 respectively. See Image generation

Python for using GPT models

To use the OpenAI API for GPT models such as GPT-3.5, with Python, you’ll need to follow the general steps explained in their Quickstart guide:

  1. Sign Up and Get the API Key. Go to the OpenAI website and sign up for an account.
  2. Once signed in, navigate to the API section and generate your API key.
  3. Install the OpenAI Python Library. Open a terminal or command prompt and install the OpenAI Python library using pip:
pip install --upgrade openai

Indeed, you can set up a virtual environment where you do not have conflicts with other libraries you install for other projects. You are not required to use a virtual environment when installing the OpenAI Python library.

Use the API Key

Store your API key securely. You can set it as an environment variable or use it directly in your code.

Next, you can check well-organised API usage examples, such as grammar correction, calculating time complexity, finding keywords from text, fixing Python bugs, playing with their Sarcastic bot Marv, and much more.

I like that you can use their playground and get Python code examples based on the usage category so quickly in one place!

Remember to review the OpenAI API documentation for more details and options.

GPT Models with Python

There are many potential applications of OpenAI GPT models with Python. Here are a few examples:

Language Modeling: With GPT models, you can generate novel text in any style or genre. You can easily create language models that mimic any human speech, from poems to tweets.

Chatbots: You can use GPT models to generate automated chatbots that can converse with users in natural language, saving you time and resources. Check their Assistants API and Math Assistant creation example for more information.

GPT models can help you suggest text while editing long-form content, like research papers or novels.

OpenAI’s GPT models are powerful tools for natural language processing that every coder should be able to use. Integrating it with Python 3 allows us to quickly and easily generate sophisticated language models quickly and easily. From generating chatbots to improving the quality of long-form text, the possibilities of GPT models are virtually limitless.


If you are interested in tailoring their models to your needs, you can also follow OpenAI’s Fine-tuning guide, and detailed Examples.

Not all models can be fine-tuned yet. They recommend fine-tuning gpt-3.5-turbo-1106 model [4], gpt-3.5-turbo-0125 soon will be available for fine-tuning as well. Fine-tuning can also be done if a model was tuned before and you need to add data.

Try the following fantastic AI-powered applications.

I am affiliated with some of them (to support my blogging at no cost to you). I have also tried these apps myself, and I liked them.

Synthesia.io can generate videos from text prompts, creates AI avatars and much more.

Hour One AI uses text-to-video generator technology that allows you to easily create, manage, and streamline cinematic AI avatar videos.

Hey Gen uses text-to-video generator technology that allows you to easily create, manage, and streamline cinematic AI avatar videos.

vidIQ helps to grow YouTube channels with optimised content and keyword generation.

Deepbrain AI helps to create videos faster with AI-powered video editing that features realistic AI avatars, natural text-to-speech, and powerful text-to-video capabilities.

Pictory.ai creates professional quality videos from your script with realistic AI voices, matching footage and music in a few clicks. Pictory.AI can also convert blog posts into captivating videos and extract highlights from your recordings to create branded video snippets for social media, and much more.


OpenAI’s GPT models represent cutting-edge advancements in natural language processing, enabling developers to integrate state-of-the-art content generation into their applications with the help of API. In this short post, we discussed the new OpenAI model for video creation and mentioned some previous models that are useful for generative AI applications. Thanks for reading!

Did you like this post? Please let me know if you have any comments or suggestions.

AI-generated art and music/sound posts that might be interesting for you


  1. Creating video from text

  2. Video generation models as world simulators

  3. OpenAI’s GPT models

  4. Fine-tuning

  5. Completions

  6. Pricing

  7. GPT-4

  8. Creating an automated meeting minutes generator with Whisper and GPT-4

  9. https://github.com/openai/whisper

  10. Text-to-speach

  11. Image generation

  12. Rate limits

  13. OpenAI Playground

  14. GPT Playgrounds at gpt3demo.com

  15. Assistants API

desktop bg dark

About Elena

Elena, a PhD in Computer Science, simplifies AI concepts and helps you use machine learning.

Elena Daehnhardt. (2024) 'OpenAI's Model Show-off', daehnhardt.com, 19 February 2024. Available at: https://daehnhardt.com/blog/2024/02/19/openai-sora-gpt-models/
All Posts