Now Writing Weekly Blogs As I Travel

AI Visual Adverts

Creating job ads for social media can feel like a whole new world of challenges...

Creating job ads for social media can feel like a whole new world of challenges. Usually, when we think job ads, it's all about the text—describing the role, the skills needed, the company culture. But let's be real, that doesn't quite cut it on social media. Social platforms are all about grabbing your attention visually. You want something that pops, something that makes someone stop scrolling and think, "Hey, I wanna know more about this."

Now, the tricky part isn't just that making eye-catching media takes a good chunk of time and a dash of creativity. There's also a whole maze of issues like who owns the images, not stepping on copyright toes, and making sure you're not accidentally showcasing real people without their consent. So, here's a thought: what if we could use AI and something called stable diffusion to whip up some snazzy visuals? Imagine having fast, striking images that are totally copyright-free and don't feature any real-life faces.

Let's dive into this with a concrete example. Say we're hiring for a nurse. The traditional route might not cut it, but this AI approach could be a game-changer. I kicked things off with a simple prompt, fed it to a versatile base AI model, and...

A young nurse, happy, working, hospital, patients Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 4195078305, Size: 512x512, Model hash: aeb7e9e689, Model: juggernautXL_v8Rundiffusion, Version: v1.7.0

Got something that kinda hit the mark but also missed by a mile. The faces and limbs were off, and frankly, the images screamed "I'm a fake!"

Not one to give up, I switched gears and grabbed a different base model from Cititai, dubbed CBINP, or "I Can't Believe It's Not Photography". And wow, just swapping to this model alone stepped up the game to a whole new level of realism.

A young nurse, happy, working, hospital, patients Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2634363526, Size: 512x512, Model hash: 6d3331c3d8, Model: icbinpICantBelieveIts_newYear, Version: v1.7.0

This new image looked heaps better, but I wanted to dial up the vibe, make it less like a stock photo. So, I played around with the prompt, making it longer, and threw in some do's and don'ts. I used positive keywords like 'perfect anatomy' and threw in a bunch of no-no's like 'deformed', 'distorted', 'disfigure', and let's not forget 'ugly', 'disgusting'. Here's where things get sticky, though. It opens up a can of worms about ethics. If our biases shape these AI models, how do we make sure they're fair and inclusive, especially in a world where AI's cranking out our ads and media?

perfect anatomy, happy, realistic, classical medium shot RAW photo of a Nurse outside a hospital, brilliant iridescent lighting, refined image, Hospital scenery in the background, crisp detailed dslr, Depth of field, Focal length, Diffuse, Emissive, Atmospheric perspective, calm, subtractive lightNegative prompt: stock photo, (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputationSteps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2312698271, Size: 512x512, Model hash: 6d3331c3d8, Model: icbinpICantBelieveIts_newYear, Version: v1.7.0

Despite these hurdles, I've managed to create some genuinely stunning and realistic nurse images that could totally jazz up a job ad or a career page. Next, I wanted to tackle creating a banner, keeping LinkedIn's banner size in mind as a template.

perfect anatomy, happy, realistic, classical medium shot RAW photo of a Nurse outside a hospital, brilliant iridescent lighting, refined image, Hospital scenery in the background, crisp detailed dslr, Depth of field, Focal length, Diffuse, Emissive, Atmospheric perspective, calm, subtractive light Negative prompt: stock photo, (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3747526812, Size: 1584x396, Model hash: 6d3331c3d8, Model: icbinpICantBelieveIts_newYear, Version: v1.7.0

Adjusting the dimensions worked fine for the background, but then the people started cloning themselves in the image. Adding a 'no clones, please' note to my prompt helped a bit, but it wasn't a perfect fix. It really underscored the need to experiment and generate multiple images to get it just right. I attempted to generate an image with text but that was not successful, perhaps that is something best left to tradionaly image editing.

(text Ashley Hospital) perfect anatomy, happy, realistic, classical medium shot RAW photo of a Nurse outside a hospital, brilliant iridescent lighting, refined image, Hospital scenery in the background, crisp detailed dslr, Depth of field, Focal length, Diffuse, Emissive, Atmospheric perspective, calm, subtractive light Negative prompt: stock photo, (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, more than one person Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 4015727892, Size: 1584x396, Model hash: 6d3331c3d8, Model: icbinpICantBelieveIts_newYear, Version: v1.7.0

All this experimenting has got me thinking bigger. What if we could take this whole process, bundle it up into a neat little app that takes a job description and spits out a whole library of custom, AI-generated media? That could be a game-changer for job ads on social media... stay tuned!