What generative AI will not replace creatively...at least not for now
I've been testing AI with Adobe and Flora AI recently and here's what I've learnt
Hey Fun Gs,
Gen AI is not a one button fix all. I really wish it was.
Now I recently shared with you what I have been concepting and building with Flora AI in my last post. And it was great to really see how far I could push one idea as a solo creative.
But let me be real with you. There were also a lot of moments where AI completely missed the mark and I’m going to share that with you today.
Here’s my original two characters that I wanted to either swap the outfit or add them into a 3D scene.
So here’s the problem:
AI has bias and lacks imagination sometimes - When I ask for a 3D scene it throws you an average of what you’re asking for. This is why learning how to prompt is important. It’s such an odd skill to learn because I don’t usually think creatively with words and you almost have to be a micro manager and instruct clearly so that what you imagine in your head to turns into something that is relatively feasible with AI.
There would be this cyberpunk bias with pink and blues that is associated with Chinese elements. 80% there… but the rest is meh - Gen AI is great with speed and iteration BUT it needs you to drive it. I cannot tell you how many times, I’ve played around with Nano Banana Pro or GPT Image 2 and AI decides to add in a new character or extra arm that I did not ask for. And most of all, I cannot tell you how many times I’ve felt frustrated because the generation was so close but still wrong.
Nano banana generated their love for chatgpt and the other one made the proportions of the avatar incorrect. This is all from a prompt of add my digital designs onto the male and female character. 
Random limbs being placed where they are not meant to be Video is WAY harder than still images - Having consistency from a character, product design or outfit is HARD af. One frame might look amazing… and then the next frame completely changes the character design or silhouette. As someone working in digital fashion, consistency matters a lot to me. Especially when trying to build believable worlds and characters.
AI still struggles with proportions - I wanted to generate a collectible model on a table and it did great but it made the keyboard so small in comparison to everything else. It felt silly. But details such as scale, proportion, and spatial relationships still confuse AI a lot more than people realise.
So yes, a lot that can go wrong and it made me realise that you really have to know what to prompt and how to direct the vision for it go right with gen AI. If you don’t then gen AI can easily override and give you outputs that are subpar.
Anyways, I will continue to experiment and learn the best practices of combining AI and 3D. Whatever I learn I will share back to you.
Let me know your thoughts and have you tried using gen AI in your projects?
Learn Digital Fashion with Me
🟢 Youtube - FREE tutorials
🔵 Twitch - FREE Weekly Live Streams, every Wednesday UK 9PM
🟣 Patreon - Exclusive access to Stephy’s technical tips and industry advice
⚪ Thr3d Academy - Advanced course that will help break you into industry - sign up to the waitlist
🟡 1:1 Consultation - Personalised advice tailored to your brand - helping you implement digital fashion into your workflow
Lastly, if you want to join a digital fashion focused community and chat to other Fun Gs then join my discord server by clicking the image above.
★ Stephy Fung ★
IG | Twitter | Tiktok | Youtube | Twitch | Patreon | Discord