Tim O’Neill, co-founder Time Under Tension
This week I presented at the In-house Agency Council Summit in the Hunter Valley, to a room of marketing and creative professionals who are just getting started on their adoption of generative AI tools.
The goal of my presentation was to expand their mindset beyond ChatGPT and Midjourney, to the breadth of other AI platforms that are available, and importantly also to the ability to mix-and-match generative AI ‘building blocks’ into your own workflows or applications.
The examples I gave were grouped into two categories; Strategy & Planning, and Creative & Production;
Strategy & Planning
Summaries & insights – Large Language Models such as GPT (which powers ChatGPT) are amazing at taking large chunks of text (or entire documents) and summarising them in any way you wish. This can be a great time-saver for researching, rather than trawling through a 100 page PDF for the 5 key points, you can ask the AI to do this for you in a few seconds.
My favourite way to do this is using Claude (a Large Language Model, like ChatGPT) via Poe.com. The reason being it allows you to upload a large PDF file (or several), and performs really well on large documents.
Research insights - Notably.ai is a synthesis platform for user research that helps you discover insights from research data instantly. You upload your data, and the platform offers dozens of templates to use, including for 4 what’s, personas, research rewind, and more.
ChatGPT plug-ins - In May of this year OpenAI released plug-ins for ChatGPT, which are available from the plug-in store for ‘Plus’ paid subscribers. There are nearly 1,000 plug-ins that you can install and use within ChatGPT, and just like apps on your mobile phone they bring new functionality from 3rd parties into the ChatGPT interface.
The benefit of this is the ability to use natural language to interact with apps such as Zapier or Hubspot. You can also create your own plug-ins for your brand, and make them publicly available to others.
Chat with your data – around the same time that OpenAI released plug-ins, they also released what was then called Code Interpreter but is now renamed Advanced Data Analysis.
This is an amazingly flexible and magical piece of technology that I can’t do justice to here, but one use for marketing teams is the ability to upload a CSV data file, and ask questions from it (“tell me about…”) and also create graphs with (“create a pie chart showing…”). Think about it as rapid prototyping BI tool with a conversational UX.
Brief creation – I described how you can take a multi-step process (such as writing a brief for an agency), and use various generative AI tools to assist you with each step of the process. The example I gave was a training workshop we ran for G’Day Group’s marketing team. The efficiency gains can be tremendous, but also allowing for more creativity to be unlocked from the inhouse team
Creative & Production
Idea generation – The US agency Addition have built a wonderful demo tool called CannesGPT, that allows marketers to create award-winning campaign ideas in seconds. Or at least, create campaigns that sound like award-winning campaigns!
The system is trained on case studies from 2023’s most awarded campaigns from Cannes Lions, including all of the Titanium and Grand Prix winners. You can then simply type in a short brief, and ideas are generated in seconds.
This is a fun tool to play with, and also a good example of how you can ‘train’ an AI model with your own (or someone else’s) content, to get a custom version of ChatGPT.
Content creation – most marketers have at least experimented with content creation using ChatGPT, but there are benefits to using specific content tools such as Jasper.ai. A couple of the benefits include the ability to create your own brand style-guide, and also to generate multiple versions of content (e.g. a Facebook post, a TikTok script) from a single request.
Fine-tuned imagery – image generation tools like Midjourney and Adobe Firefly are amazing, and well known to marketing pros. What is less known (and not currently available in these tools) is the ability to ‘fine-tune’ based on your own images, to then be able to create new images in the same style, or of the same object.
For example, by uploading 15 photos of as your training data, you can then generate an image of your product in a new scene, or generate new photos in the same brand style. There are a few tools that easily allow you to experiment with fine-tuning, my favourite is Astria.
Generate humans! – Photo-shoots getting you down? Don’t have time to fly to Paris? Why not generate 100% inauthentic humans with AI? Human Generator lets you create photo-realistic photos of people, with control over clothing, ethnicity, age, location and more.
AI Avatars – until now, AI Avatars were firmly stuck in the uncanny valley. Even the most realistic are ‘off’ enough to scare the children! Very soon we will have photographic quality video avatars, with lip syncing to whatever script we (or ChatGPT) has created. This will mean AI Avatars are no longer limited to chatbots stuck in the corner of your browser window, they can be used for marketing purposes such as fast production of a video interview with your brand ambassador. But 100% AI generated. HeyGen are at the forefront of this with some crazy demos of what they are building. Yes, this is potentially scary stuff, so tread carefully and ethically!
Voice interfaces – the technology for text-to-voice and voice-to-text is now so good, that you can have a conversation with a Web site, such as this example from Twise that lets you speak to Barak Obama, Mr Beast or Harry Styles. The voice is trained on that of the likeness of that person, and the ‘brain’ (a Large Language Model) is trained on the way that person speaks.
The combination is a spooky voice conversation with your favourite influencer. As a technical demo Twise is amazing, and hopefully sparks ideas for how you could use this technology in interesting ways, such as our I Spy with My AI prototype.
Music – generation of music and also videos are right at the cutting-edge of generative AI technology, we’re in the ‘wow’ demo stage rather than production ready tools. I wanted to close my presentation with a wow, and the example IHAC rap I created using Suno bought down the house! You can experiment with it yourself for free using the Discord app, it works just like Midjourney but the output is a song (with lyrics and vocals) based on your text prompt. Try it, you won’t be disappointed!
Generative AI as LEGO blocks
A closing comment in my presentation was describing how the same technology that is used to build the platforms I just showcased, is available to anyone to combine in new and unique ways.
You should think of generative AI as a collection of LEGO blocks, for example text creation, summarisation, image generation, voice or music. It is possible to build these blocks in new and unique ways to suit your own needs as marketers.
As with LEGO, the only limit is your imagination. Have fun and play!