AI IMAGE GENERATORS, GFP-GANs, AND PROMPT DESIGN
Extra advice and tips for advanced AI image generation
Deleted scenes and random thoughts
I like to keep my posts short and sharp. That means some tips get cut. So here are some extra nuggets of wisdom for readers who want a deep dive.
Extra Details from “Download a free GFP-GAN to improve your AI-generated faces”
What to do if you don’t get the image you want
Midjourney lets you rerun an image (Upscale). Upscaling will cost you (or use up one of your 25 free jobs). You can also reuse the same seed source to replicate similar results on a new roll of the dice. Or you could recycle the current image as input. But all these options will burn up precious credits.
Jasper doesn’t currently offer ‘upscaling’, but on the plus side, you have unlimited generations (rerolls). That’s one reason I recommend Jasper.
In any case, before you give up on an image that’s nearly what you wanted, but lacking in the finer details (maybe it’s blurry, the smile is weird, or the eyes are missing), you might first like to try an image restorer to refine it.
Sometimes perfection is a few clicks away!
Using a GFP-GAN is just one step to perfect portraits
GFP-GAN won’t solve every problem, and sometimes it introduces new ones. Skin texture can become too airbrushed and plastic-looking. And expressions can be almost comically-altered towards the cheerful. GFP-GAN likes to put a grin on people’s faces, like an overdose of Joker gas.
Never fear! What we’re really looking for are better eyes: more symmetric and circular. In fact, uneven eyes are how both the human brain and AI models instantly detect fake faces. So it’s important to get the eyes right.
I recommend combining your original image and the GFP-GAN output in Photoshop. Set the GFP-GAN as background, and the original as a mask.
Then just brush away the areas that need greater clarity, like the eyes, while retaining any imperfections and noise that add to skin texture and realism.
Extra Details from “How to get your AI art generator to stop cropping off people’s heads”
Why do AI Image Generators crop women’s heads more than men?
Many of the words we use to try to get a good image counterproductively encourage cropping. Prompts like “beautiful” and “stunning” are more often associated with close-ups and selfies than they are full-body shots.
One explanation for why AI crops images dates back to 2015. Researchers at Stanford found AI prefers cropped selfies, as it thinks subjects that fill the frame are more attractive. AI does this for images of women more than men, which anecdotally is the same case now with image generation.
Anecdotally, cropping appears to happen more when words like “attractive” are in the prompt, which is particularly frustrating when you have an aesthetically pleasing output that’s too cropped. So go easy on modifiers that are associated with selfies and physical attractiveness for best results!
Extra Details from “Using an AI writer to generate the text prompt for an AI image generator”
Is LIGHTBULB just another Prompt Generator?
It’s important to note that my LIGHTBULB recipe is different from prompt builders, like PromptoMANIA. Prompt builders require you to have a general topic or idea in mind for the image you want to generate.
Prompt Builders assist you in assembling “prompt parts”. It’s like a Subway menu: you choose from a list of options to create your text prompt. This is helpful when you need structure for your prompt, but it’s not as helpful when you’re blocked and have no idea what to generate in the first place.
LIGHTBULB, on the other hand, offers a much more self-contained approach. Everything is generated for you, from the text prompt to the customized image itself. It’s like AI Inception, or robots making robots.
Your input can be any text: a blog article, an essay, a product description, a passage from a book, to a full-blown story or poem. Whatever inspires you. And Jasper will then suggest what the perfect image would be, and gives you a starting point for your prompt (including medium, style, mood etc).
You can also finetune your prompt for more hyperrealism by adding some parameters to your prompt like “unreal engine”. And if you find your AI is cropping out parts of the face or head, try my tricks and tips here. You can also use a GFP-GAN or a neural filter to fix details and improve resolution.
Finally, you can feed the Jasper-generated prompt into PromptoMANIA for added specificity now that you have your creative idea. I love to stack AI Art tools in this way to create works that are fascinating and easier to generate.
Conclusion:
I hope you found these extra tips for advanced AI image generation useful.
We started with how to improve images with GFP-GANs by combining the output with your original image in Photoshop for better results. And we’ve addressed why AI image generators crop heads, and how to avoid it. We also discusse dhow to use LIGHTBULB to generate customized prompts.
There are many techniques to get great results from your AI art generator. By understanding how the AI tool works and experimenting with different settings and inputs, you can create fascinating and realistic images. To keep up-to-date with all my discoveries, be sure to hit the Follow button. Thanks!
Ready to join Medium?
Unlock the entire Medium catalogue with my referral link, and you’ll also be supporting my writing here at no extra cost to you. I’d appreciate it.
Who is The Jasper Whisperer? How can he help me make AI artwork?
I’m fully certified by Jasper AI and a group expert. I partner with Jasper AI. Jasper Art is now only $20/mo for unlimited image generations!
You can start to generate AI art right away — no waitlist!
I offer Pro training on how you can use AI to create commercial content.
If you’re interested in discovering more, please feel free to contact me.
