Leonardo.Ai — A Marketer’s Dream
Step-by-step Workflow for a Personalised Brand Campaign
Want to stand out from the crowd with your online presence?
G’day — we all like to feel special and sometimes a personalised asset is enough to do the job! Imagine that as part of your employee onboarding, your bright-eyed, bushy-tailed new-start could make a personalised branded asset to announce that they’ve just started with your company!
Or perhaps for your next event, you’re looking for a unique way to showcase your upcoming speaker lineup? Regardless of your marketing needs, with Leonardo.Ai’s latest product tools, you’ll be able to create a suite of consistent personalised assets in no time.
As part of Superside’s 2024 Virtual Summit, The Infinite Canvas: Exploring AI’s Creative Possibilities, I’ve designed a play-along Ai workflow using Leonardo.Ai to create a series of personalised LinkedIn campaign assets, while retaining the style of Superside’s initial event branding. Personalised assets are a unique and fun way to encourage social sharing!
So let’s break it down into the steps.
Leonardo.Ai
For this workflow, we’ll be using Leonardo.Ai — you can sign up here.
Leonardo.Ai is a top-tier generative Ai content-creation suite, for creating production-quality visual assets with unprecedented quality, speed, and style-consistency.
Leonardo.Ai has freshly released three uniquely vital product features, which make this specific workflow possible: those being Character Reference, Style Reference, and Content Reference.
Character Reference
Generates an image in the likeness of the provided character reference image — this works best with close up face portraits (mid-strength works best for maintaining resemblance while allowing creative flexibility).
Style Reference
References the style from the input image to guide your image generation.
Content Reference
References the form and content from the input image to guide your image generation.
Ai-Enhanced Workflow
Campaign Assets
If you’re playing along at home, you’ll want to download the square marble-statue graphics, supplied at the very end of this article under Resources.
Superside had already created a suite of marble characters for their event marketing, however, to ensure the outputs would generate fitting comfortably within a square frame, I first saved the graphics separately, against a white background, with a consistent 1:1 square aspect ratio.
Headshots
Fortunately, Leonardo’s Character Reference is quite forgiving, so just a quick LinkedIn stalk for headshots did the job. The clearer a portrait, the better.
Now Into Leonardo.Ai
Leonardo.Ai has two user interfaces: Our legacy interface, with all the bells and whistles, and V2 — a new simplified, easy to use interface for Ai first-timers and those getting used to the generative process. I’ll go through the workflows on both interfaces so you can get an idea of the main differences between the two.
Leonardo has a plethora of other creative tools across the platform, from its Upscaler — a tool for making images high resolution, Motion — a feature that converts a static image into a moving asset, through to Live Canvas — for creating images from quick sketches. But let’s just focus on Image for now.
To get started, you’ll want head over to the Image Generation tab — this is where the crux of Image generation takes place.
Legacy Workflow
This article was written for Legacy Mode, so make sure you turn on Legacy Mode to see the Legacy UI screenshotted in this article. In the Image Generation interface, you’ll be met with a list of settings in the left column, as well as the top bar. The creative possibilities with this platform are seriously endless, so for the sake of really nailing this particular workflow, let’s focus just on these settings for now.
Settings
Make your platform settings to as per below:
PhotoReal V2 — A powerful pipeline designed to generate hyper-realistic photos and lifelike portraits.
Alchemy V2 — Leonardo’s signature image pipeline, which brings premium, high-fidelity image generation and coherence to all existing platform models.
Aspect Ratio — Set to 1:1 for square outputs. I’d recommend to lock dimensions (lock icon) for consistent output dimensions. This will ensure your outputs remain the same dimensions, unaffected by the dimensions of the images you use for references. If your images are coming out in the wrong dimensions, it’s most likely taking on the dimensions of your reference images — so you’ll want to head back here to lock the dimensions as 1:1.
PhotoReal Style | Portrait — As you’ll notice, Photoreal has an abundance of varying camera styles, each excellent at their specific trained style. For this use-case, I’ll be using Portrait due to its top capabilities in detailed facial features.
Prompt
Copy and paste the below prompt while feeling free to change any of its content, such as gender or adding any flourishes you may want. i.e. circular glasses, long-flowing curly hair, etc.
Crafted in exquisite detail, a low saturation stock image 3D marble sculpture of a woman facing forward, against a stark white background. The hyper-detailed features of the face intricately, clearly carved into the marble, showcasing expressions of beauty and grace. The flowing fabric that adorns her form is depicted with intricate folds, adding a sense of movement and realism to the art piece. Created using advanced 3D software, this stunning image radiates a sense of elegance and sophistication, capturing the essence of timeless beauty in a modern medium.
I don’t love writing detailed prompts, and lucky for me, I no longer have to — Leonardo’s Improve Prompt feature takes care of that. Beginning with a simple ‘marble statue of a woman’, I then clicked Improve Prompt, for Leonardo to extrapolate the prompt for me. I then typed in a few extra details, specifying a low saturation stock image of a 3D marble sculpture.
Image Guidance
With all of that set, let’s head over to the serious stuff. Pressing on the Image Guidance tab located under the prompt bar, you’ll open up the Image Guidance, aka, references area. This is where we set all of our references.
Now just a few housekeeping rules; you cannot have more than one Character Reference, or Content Reference — however, you can have multiple Style References (but let’s stick to just one for now). If you trigger a system error on the platform, chances are you’ve set your references wrong, so let’s be sure to set them correctly.
Character Reference (Mid)
Turn on Image Input, and select Character Reference from the drop-down menu. Upload your clear headshot, and set its strength to medium. Mid-strength works best for maintaining character resemblance, while allowing for flexibility in its new generation.
Style Reference (Mid-High)
For the next Image Input, select Style Reference, and upload one of the campaign assets. I used the same Style Reference image across all my generations just to make sure they all maintained the same aesthetic.
Content Reference (Mid-High)
For the final Image Input, select Content Reference, and upload a reference image that’s of the content you would like to generate. Content references look at the form and the content of the uploaded image to help guide your new image generation.
Right, we’re ready to roll. Hit that Generate button.
Ta-da! What a masterpiece — you should have four great options to choose from. Keep hitting generate until you get an outcome that you love!
If something doesn’t look quite right, head back into the Image Guidance and have a play with the Content Reference and/or Style Reference settings. Sometimes Mid strength works better depending, on your input, sometimes High strength works better — have a play until you find what works best for your goal. If your Character Reference isn’t coming through well enough, try using a different headshot image as input.
Generative Ai is an inherently iterative workflow, so going back and forth is a natural part of the creative process. Play until you get it just right!
Final Touches
Once you have identified the image you’re happiest with, click on it to open this additional feature. At the bottom of the display bar, you’ll see a graphic of a diagonally-striped tile — this is the Remove Background feature.
Press on Remove Background and wait a few seconds as it creates a transparent layer. Once it’s finished, at the left of the base bar, you’ll see a drop-down menu to display the Original Image or the No Background image. Press on the download arrow in the base bar to download this file.
Repeat Workflow
Congrats on nailing the first workflow! To create the LinkedIn profile header and hero pic, we’ll need a minimum of four different poses. You’ll want to keep all the settings the same, except to Content Reference. Content Reference will change the pose, so repeat the above workflow only changing the Content Reference input image, to get a variety of different positions. Be sure to remove the background on three outputs.
Bringing It All Together
Whichever your go-to graphic design software, you’ll simply want to add your own transparent images on top of the supplied Banner as such:
Remember to keep one image with a background for your main profile. And you’ve done it! You’ve completed your first personalised workflow!
Additional Branding
If you’re eager to take it a step even further, I’ll quickly run through how to create your own logo in the same campaign style.
For this different workflow, we’re no longer using Character Reference or Content Reference, but instead, Edge to Image as our Image Guidance setting. Edge to Image — detects edges in an input image and uses these edges to guide a new image generation.
Since we’re continuing with our campaign aesthetic, Style Reference should be set with an image from our previous generations— I particularly enjoyed the gold earring of this output, so I’ll run with this one.
An additional catch, is that I’ve changed the PhotoReal Style from Portrait to Stock Photo for my desired output. Portrait is great for faces, but Stock Photo is better for the type of marketing asset I’m creating.
PhotoReal V2 | Alchemy V2 | Leonardo Kino Xl | Stock Photo | Prompt:
A luxuriously crafted marble and gold tile catches the eye, its smooth surface gleaming with the subtle elegance of a finely embossed logo. The angle of the photograph captures the intricate detail of the design against a pristine white backdrop, creating a striking contrast. This impeccably captured image, possibly a high-quality photograph, showcases the tile’s flawless craftsmanship and exquisite beauty. The interplay of light and shadow highlights the rich texture of the marble, making it appear almost tangible to the viewer. This visually stunning composition exudes a sense of luxury and sophistication, inviting admirers to marvel at the fine artistry on display.
For Superside’s logo, I did the same workflow with a tweaked prompt — A gleaming gold logo against a stark white background. Remove BG, and done!
Other Styles
And the fun doesn’t have to end there — once you’ve got this workflow down-pat, you’re free to run wild with the possibilities!
Perhaps you’re wanting to create characters made out of clay, or perhaps as a 3D low poly isometric vector illustration. Or even a graphic novel? Or in the style of a post-card you’ve had stuck to your fridge… with Style Reference the options seriously as endless.
For this, you just need to first create an image in the style you want. Make sure you have all your Image Guidance References turned off, and type the style you want into the prompt bar. Keep playing in here until you have a perfect Style reference image.
Then you’ll just use that image as your Style Reference, while repeating the Character Reference and Content Reference steps from earlier. Viola!
This workflow was created in Leonardo.Ai by their Senior Creative Technologist, Jessie Hughes. It was created for Superside’s 2024 Virtual Summit The Infinite Canvas: Exploring AI’s Creative Possibilities.
If you’re intrigued by how you can use this workflow to create consistent personalised assets across your own company, or for your next event, please do reach out for help. Our door’s always open for collaborations.
Resources
