Filling the content gap with Ideogram, Runway, Claude and GPT

This guide covers how marketers can use AI tools to grow their image library. It walks through identifying content that performs, creating image variations with Ideogram, and generating video from stills with Runway.

Out of images for your channels?

Do you work on one of those marketing teams with an unlimited budget that can commission as many shoots and film days as it likes? Good for you, skip this guide. For the rest of you still reading: anyone who has done a few years on a marketing team knows the frustration when the image library runs dry. "We don't have enough video for our Instagram" has been said a few times, in one form or another.

There have been plenty of promises about what AI will do, that it will assist every part of your day and optimise every single minute. But a lot of people find it hard to pin down concrete workflows where AI actually becomes part of the everyday and genuinely helps. Especially for anything beyond text generation, like images and video.

Here is a guide to using AI to stretch your image library and create new material.

Step 1: Identify content that performed

Since we are working with material you already have, you may as well use the material that did well. How do you know which? Without going too far down the "measuring the effect of your communication" rabbit hole, a decent start is to look through your social channels.

• Go through your social media channels.
• Identify images that got more engagement than average.
• Pick those images to work on.

Note: this step is not required, you can use any images you like. But it is satisfying to work from content that has data behind it, and make more on the same theme.

Step 2: Create variations with Ideogram

Next you need an AI model that can interpret images and create new variations based on your prompting. I have been testing Ideogram for a while and think it is a very good platform. Because of the output, of course, but also because it is easy to use and the price is manageable for most people.

"Prompt" = a text instruction given to an AI to steer it towards generating specific content or performing a particular task.

So go to Ideogram.ai and create an account. Once that is done you are met by a text field where your prompt goes. Click the text box and the field expands, giving you the full toolbox.

Ideogram has an "Upload image" button that lets you upload your image. From there the "Remix" button can generate a prompt automatically, and you can set how much your original image should weigh when it generates a new one, on a scale of 1-100 called "image weight".

If you are doing this with a real photo, a tip is to select "Realistic" so it generates a photorealistic image. There are a few more tools for adjusting the image, but setting Image Weight and playing with the prompt is a fine place to start.

Prompting for images is one of those things that lives somewhere between art and science when it comes to what works best. And every generative AI platform develops its own methodology and documentation for how to use that particular platform well.

Tip: bring in a platform like GPT, Claude or whatever you use to write your prompts and refine the one Ideogram generates automatically.

I have built custom chats that know the documentation for a few different AI platforms. If you are going to do this over time I really recommend doing the same. The platforms usually have specific terms and tools that are unique to them. If you build a Custom GPT with access to Ideogram's documentation, for instance, it can phrase your prompt with the right words and concepts, and you are off to a much better start.

So once you have your own GPT running, or you simply paste the documentation into the chat, you can copy the auto-generated prompt across and ask GPT to adjust it with the changes you want.

Ready-made CustomGPT for Ideogram: Ideogram Prompt Generator

Paste the new prompt with your adjustments back in and click "Remix", and Ideogram works up a batch of variations of your image to carry on with.

Tip: if you are not sure yet what adjustment you want, just ask GPT to write a prompt for an image AI where you want an alternative version of the picture, and upload the instruction along with the image.

Tip: before we move the image over to Runway to make video, give it one last look. Are there details or small adjustments we want? Change the colour of a garment, add clouds to the sky or some dramatic mountains on the horizon? For that kind of thing I find it easier to take the image into Photoshop and use Adobe's built-in AI, which handles those edits easily.

Her shoe came out a bit odd, for example, which is a perfect thing to select in PS and ask for variations on.

Step 3: Create video with Runway

Once you have an image you like we can take it a step further and easily create video based on the image you just made. Runway is, at the time of writing at least, the platform furthest ahead when it comes to generative AI focused on video.

So create an account at Runwayml.com, roughly the same price as Ideogram. Runway has a range of tools for creating content but its flagship is Gen-3 Alpha, which is what we will use. Make sure that is the model you are on, or Gen-3 Alpha Turbo.

Once you have an account and are up and running in Runway, choose Image to Video, or Gen-3, and you should land in a view where you can upload your new image. Click "Upload" and find the image you downloaded from Ideogram. Below the image you add your prompt, which now applies to generating video. Here I would definitely use Runway's documentation to write your prompt, since video is more advanced and Runway gives you many ways to steer it. So generate a prompt with a custom GPT again and paste it into the text field in Runway. Remember to upload the image when you ask GPT for the prompt, so it knows what material you are talking about.

Ready-made CustomGPT for Runway Gen-3: Runway Magic Prompt

You can then choose whether the image you uploaded should be the first or the last frame, and whether you want a 5 or 10 second video. Easiest is to start at 5 and warm up. Then click Generate.

A few last things

• The limits of AI: be aware that AI still struggles with details like fingers, hands and faces. Keep that in mind when choosing source material.
• Transparency: be open about material being AI-generated when you publish. It builds trust and can spark interesting conversations.
• Experiment: try different settings and prompts to find the best results for your brand. Share tips, tools and prompts between colleagues.
• Mix your content: use AI-generated material as a complement to authentic content, not a replacement for it.
• Stay on brand: make sure the AI-generated images and videos match your brand identity.