OpenAI’s GPT-5 is the long-awaited upgrade to what many would call the most popular and ubiquitous AI model currently available to the general public.
Ever since the release of GPT-3.5, or ChatGPT, in 2022, AI has caused a total paradigm shift in how millions of people write, code, research, and create. Since then, OpenAI has worked tirelessly to stay ahead of the increasingly impressive competition— both locally from Google and Anthropic and internationally with Deepseek.
But what’s so impressive about GPT-5 and why should you switch over from older models? Well, here are some of the coolest new features this model provides— and six things you can do with them.
Input Video As Context (…Kinda)
With the earliest version of ChatGPT, users could interface with the impressive language model by typing in a small text field. It wasn’t long after the launch of this model that subsequent models supported images as input, and then Voice Mode supported audio.
The logical next step is video. This was a feature that was heavily rumored to be included in OpenAI’s latest release, but the reality is that we’ll still have to wait for native video support.
But where there’s a will, there’s a way— intrepid prompt engineers and “vibe coders” have come up with unique ways to circumvent these limitations, and the good news is that with a little ingenuity, GPT5 Plus can accommodate video inputs easier than previous models.
Users have been able to use videos as input in a roundabout way— adding screenshots with subtitles, or by isolating the audio and feeding it through Whisper. But due to GPT-5’s multimodal and agentic support, you can streamline so much of this process, distilling it to one or two prompts. The end result is a tailor-made app with excellent frontend design that allows you to upload your video, tell the model what you want it to do with it, and voila!
Here are just two of the possibilities that come from native video input support:
GPT-5 Use Case #1: Wildlife Expert
Upload footage from a trail camera and ask GPT-5 to identify all flora and fauna in the area based on sight and sound— from the faintest birdcall to the smallest flower.
Here’s an example prompt you can use, based on the “Anatomy of a Perfect Prompt” section of Humanity’s Last Prompt Engineering Guide:
“GOAL: I want you to design a web app that analyzes screenshots taken from a live trail camera video feed, identifies all of the flora and fauna in the screenshots, and then catalogs them in a database.
FORMAT: Start by doing some research to identify the best trail camera with live streaming capabilities for this project, as well as the software needed to connect to the camera's stream, take screenshots from the livestream at regular intervals, analyze the metadata and visual data in each screenshot, cross-reference them with sources like iNaturalist, and then organize them into an accessible database. Return a bulleted list with all of the tools needed, as well as a link to obtain API keys for each of them. After you've done this, the next step is to design the frontend of the web app (with both mobile and desktop support) that will use all the APIs to provide the functionality that I described above.
IMPORTANT NOTE: The web app should be designed so that users can connect their own local or cloud storage to store the data, including screenshots and cataloged flora/fauna. In order to prevent the screenshots from filling up the user's storage endlessly, it should automatically delete screenshots that don't contain newly identified flora/fauna after a specific period of time, and it should have a maximum storage size that can be adjusted by the user. Also, if any aspect of my proposed web app is not feasible for any reason (e.g. there isn't any specific software with API support to accomplish one of the steps of the process), mention it in your report and suggest viable alternatives.
CONTEXT: I am a nature lover who is involved in several local communities of hikers, campers, birders, fishers, and other outdoorsy folks. I want to design an app that my community can use on their phones or laptops with their own trail cameras to automate the process of tagging local wildlife. Since these are very outdoorsy people, they're not very "tech-savvy", but I still want them to have access to a user-friendly tool that enables them to get even more out of their hobby.”
Just a heads-up— this isn’t a one-shot prompt. That means you’ll need to write a few follow-up prompts to get the code after the research stage, but it shouldn’t take longer than an hour to get everything working properly.
If you want to see what a one-shot prompt looks like, check out this next one!
GPT-5 Use Case #2: Study Buddy
Record a video of your college professor’s lecture and ask GPT-5 to analyze the professor’s voice and everything they write down on a whiteboard to generate detailed study notes.
If you’re interested in doing something like this, here’s a great one-shot prompt to use:
“GOAL: I want you to design a webapp that allows me to upload video or paste video links to recorded lectures, generate a transcript of the lecture, and save screenshots of all the slides, whiteboard drawings, and/or other visual aides included in the lecture, and then generates a one-page study guide that condenses all of the most important information.
FORMAT: Write the code for a local webapp with a user-friendly frontend. The app should first allow the user to upload or paste the link for a video file; then, it should extract the transcript, take screenshots of the slides/drawings/visual aides, and then store them in a user-accessible database. Finally, the app should generate a one-page PDF that contains the most relevant information.
IMPORTANT NOTE: The web app should be designed so that users can connect their own local or cloud storage to store the data. In order to prevent the screenshots from filling up the user's storage endlessly, it should only take screenshots that contain new information, there should be a time gap of at least 10 seconds between each screenshot taken, and it should have a maximum storage size that can be adjusted by the user. Also, if any aspect of my proposed web app is not feasible for any reason, do not write the code and suggest viable alternatives instead.
CONTEXT: I am an autodidact who is interested in absorbing the most information from a vast array of fields in as short a period of time as possible. I enjoy watching free college lectures on YouTube in my free time, but I often find myself increasing the video speed to get through these lectures as fast as possible, which causes me to miss out on a lot of important information and can cause some anxiety. Having printable one-page "study guides" that I can read while enjoying my morning cup of coffee will limit my screen time while still allowing me to pursue my passion for knowledge.”
Generate Websites With Aura
Two of the biggest competitors to OpenAI’s GPT series of models are Anthropic’s Claude and Google’s Gemini. Although ChatGPT is commonly considered a jack-of-all-trades, these two models carved out solid niches for themselves by mastering one use case: Claude with code generation and Gemini with agentic workflows.
GPT-5 aims to close the gap between both competitors by offering extremely impressive code generating capabilities and multimodal support designed to support popular IDEs like Cursor and Windsurf. In particular, GPT-5 has gotten leaps and bounds better at generating frontend code, with users claiming that previously difficult tasks could be completed in one shot.
Here are two more interesting ways you can take advantage of GPT-5’s improved capabilities:
GPT-5 Use Case #3: Portfolio Upgrade
Turn your online art portfolio into an interactive digital museum experience by prompting GPT-5 to enhance your pieces with motion graphics, particle effects, and/or dynamic transitions.
Take a look at this prompt for a good starting point to try this with your own design portfolio:
“GOAL: I want you to turn my images into animated motion graphics using HTML, CSS, and Javascript.
FORMAT: Analyze the four art deco images I am attaching to this prompt and use HTML, CSS, and Javascript to animate them. The end result should be one webpage with the four static images that I attached, one in each corner of the screen; clicking on one of the images should transition the image to fill the entire screen and play the looping animation, and there should be a small but clearly visible "back" button that smoothly transitions the user back to the four static images. Include a watermark that reads "Forward Future University".
IMPORTANT NOTE: The webpage should be compatible with mobile, tablet, and desktop displays. The images should be used as faithful inspirations for fully-rendered motion graphics; do not just add a simple effect over the static image, but make something truly breathtaking.
CONTEXT: I am a freelance graphic designer who wants to design a website that stands out from others in my field. I specialize in art deco inspired designs that I think would look really beautiful as looping animations, and I thought it would be really cool to design my web portfolio as a "digital museum" that gives these images the gravitas and "wow factor" that I think they deserve.”
GPT-5 Use Case #4: Make a Video Game For Your Cat
Take a picture of your pet cat’s favorite toy or treat and prompt GPT-5 to make it into a cat-friendly video game. Using GPT5’s new Agent Mode with a ChatGPT Plus subscription — and the expert-level prompt shared below — you can generate a full browser game in one shot with 13 minutes of self-guided coding.
Here’s the prompt you can use to make your own cat game:
“GOAL: I want you to create a touch-based browser game designed to be played by a cat on an iPad starring my pet cat's favorite stuffed mouse toy.
FORMAT: Take the picture of my cat's favorite stuffed mouse toy and write code for a game that can run in a web browser. The game should have one, two, or three small stuffed mice (generate the images for the mice based on the attached image of the stuffed toy) moving quickly and erratically across the screen, similar to an actual mouse's movement patterns, and provide visual feedback and one point each time the user touches one of the mice. The game should start when the user presses the "start" button, and timer should count down from 60 seconds before reaching a "game over" screen that also has a "play again" button to restart the game.
IMPORTANT NOTE: Make the game with as little code as needed to provide the features that I described above; it should be lightweight and easily run on a second-generation iPad's browser app.
CONTEXT: My child enjoys playing on his tablet, but I want to find a way to encourage him to spend more time playing with our pet cat. I think a really cute way to engage him with our pet cat while also incorporating his favorite toy would be to create a game that both him and my cat can play together; if both of them are trying to touch as many of the mice as possible to earn a high score, I think it would be a cute way for the both of them to bond with each other and provide some cute videos that I can show our extended family.”
And here’s a video showing how GPT5 can rapidly code, test, troubleshoot, and revise the project without any extra guidance when running in Agent mode:
Fine-Tune Reasoning and Tool Calling via API
In the lead-up to GPT-5’s release, OpenAI rolled out specialized o models that could perform specific tasks. These include o3, designed with advanced reasoning capabilities in several disciplines, and o1, which is specialized for math and science-related queries. They also released GPT-4o, which offers real-time analysis of audio, images, and text.
If you’re feeling confused about these names, you’re not alone. It can be confusing to remember whether your prompt is best suited for o1, o3, 4o, 4o-mini, 4.1, 4.5…
In my opinion, the best aspect of GPT-5 is that you no longer have to manually select which of these models to use to solve your problem— because they’re all included in its Mixture-of-Experts (MoE) architecture!
GPT-5 Use Case #5: Occam’s Razor
Now when you ask GPT-5 your hardest questions, you can expect it to automatically select the right tool for the job. Naturally, this removes a lot of the trial and error that goes into prompt engineering (and it makes it less appealing to switch over to a specialized model from other companies, like Anthropic and Google).
Sometimes you don’t need a wall of text or a 12-step reasoning chain; you just want the cleanest, simplest answer in the shortest amount of time. GPT-5 makes this easy with its new reasoning_effort setting. If you set it to minimal, the model won’t overthink things: it will skim your materials, stop gathering context as soon as it has enough, and give you a clear recommendation.
Remember the example I shared with you for Use Case #2, the Study Buddy? Here’s a variation of that prompt you can try when you want a no-frills, executive-style decision brief:
“GOAL: I want you to take the lecture transcript, images of slides, and text documents that I provided and distill them into a one-page study guide capturing the most important concepts.
FORMAT: Output should be one page long (around 500–600 words). Start with a 2-sentence summary of the main idea. Follow with 3–5 short sections highlighting key points, definitions, or arguments. End with a bullet list of 3 actionable takeaways or review questions.
IMPORTANT NOTE: Keep the output concise and avoid over-explaining. If certain information is missing, state briefly what would be needed, but still produce the best possible guide with what’s available. Use reasoning_effort: "minimal" to avoid unnecessary chains of thought and keep the focus on clarity and speed.
CONTEXT_GATHERING:
<context_gathering>
Goal: Collect just enough context from the provided materials to generate a high-quality one-page guide. Deduplicate any repeated content, stop reviewing once the core themes are clear, and do not expand beyond the materials I’ve shared.
</context_gathering>
CONTEXT: I am an autodidact who watches free college lectures and reads widely across multiple disciplines. I often move too quickly and miss key points, which makes me anxious. Having a single, clean study guide that I can print and review over coffee helps me absorb information faster while reducing screen time.”
GPT-5 Use Case #6: Think Outside The Box
Naturally, if you can expect GPT-5 to know the best model for the job, you can also try the opposite. If you’re looking for a really unique approach to solve your problem, you no longer have to crank up the temperature and run the risk of getting hallucinated outputs; instead, try prompting GPT-5 to use a less conventional model to try and solve your problem.
For example, you might tell it to solve a geometry problem visually, or to reason like an engineer while designing a piece of art. GPT-5’s MoE setup lets you push it into those non-standard expert modes.
Check out this example prompt to see how you can apply lateral thinking techniques to your prompting:
“GOAL: Solve my problem using an unconventional tool or angle. Generate 3 very different approaches (for example, use diagrams for a math problem, or logic models for an art project), then pick the best and polish it.
FORMAT: Show me a short plan explaining which tools or methods you’ll use. Give me three alternatives (include quick sketches or ASCII diagrams if images aren’t available). Deliver a final, detailed version of the chosen idea. Finally, provide instructions someone else could follow to recreate it.
CONTEXT: Instead of “just another safe answer,” I am deliberately forcing you to wander into creative territory. Briefly consider the simplest, most direct way to approach this issue— then throw it away and give me something weird.”
Before you dive in, here’s a quick warning.
Something to keep in mind when adopting unique prompting techniques and/or structures is the possibility of including conflicting information. Although people like to blame the training data on hallucinations, the truth is that a lot of hallucination can come from conflicting logic in the prompts users write to LLMs.
Fortunately, OpenAI Playground now includes an Optimize feature with the release of GPT-5. Use this tool to look for any logic conflicts or vague language that could prevent your prompt from delivering the results you’re seeking.
With a Little Patience, the Sky’s the Limit
These are just a few suggestions that are meant to give you an idea of how to get the most out of OpenAI’s newest model. Putting all the hype aside, the truth is that GPT-5 will provide the best results if you approach it with a solid idea of what you want to accomplish, reflected in a comprehensive prompt.
You may be surprised to find that many of your dream goals when working with LLMs are now much more easily attainable with this model— many users are reporting that they’ve been able to get exactly what they ask for in one shot! However, this doesn’t mean you should be discouraged if you don’t get the perfect result on your first try.
Remember that the biggest strength of LLMs is the ability to talk to it and collaborate toward a common goal. Don’t be afraid to go on tangents, be prepared to encounter some unexpected challenges, and keep in mind that human input will still be necessary at several points in the process.
What are some ways you’ve used GPT-5’s impressive new features? Did we miss any good ones? Let us know by leaving a comment, and don’t forget to check out Humanity’s Last Prompt Engineering Guide, now updated for GPT-5!
Want to get more from GPT-5? Watch our video: How to Make Better Prompts
