AI Gen is Too Fun
Sorry, I'm addicted. You should watch my new video
I am a slopaholic. I already wrote about sharing my claude code outputs, but fortunately, my addiction runs deeper.
some time last year, soon after Sora Image gen was announced and there was a craze for “ghiblificaiton”, I made a joke image, me and my friends as furries. haha!!
then I made another.
and another.
my friend and I spent the entire weekend, non-stop, generating furry images… hundreds of images in a day. I was hooked.
I’ve ended up generating 1000s of images at this point…
But its over. And yet, we are so back.
Video gen digression
I’m new to videos, so while i will mostly be discussing images, I have instead started also making videos !! Here is my first attempt at one :)
I am still new to this so please give feedback and be kind >~<
How it started…
Generating images was too much fun.
It was the perfect loop, it is easy to make but often they are good-but-not-quite-right, and so you need to generate more and more and more.
this was the closest I got to upgrading to the 200$ chatGPT plan. but 2 sets of 2 images in parallel was mostly ok for me.
one of the first things I did was spend the time generating a signal sticker pack. I still do use it sometimes ! even if I would probably make a different selection of images making a sticker pack now.
It can take a while to get a persona that you want, but once you have one, it is very easy now to keep very high consistency with the persona. You just pass sample images as context.
The models can be limited in what they can do in one step. But they don’t need to do things in one step. You can make one image, then provide this as a reference. You can bootstrap your way up to whatever you want… (sometimes)
though my favorite activity that I started doing was just taking photos, new and old, and recreating them with Sora, with my new character.
or seeing an image i like online, and remixing it with my own character.
I guess for a long time my philosophy when taking photos has been “it doesn’t matter if the image is garbage, you want it for the memory, you can use future AI to make it look higher quality”. And I guess it feels like the past year or two has shown that this is likely true.
I still don’t take enough photos.
but alas, all good things came to an end. The OpenAI Sora dashboard closed but a couple months ago. I was pretty sad.
Lucky, there did exist a tool for Sora Export, and I could export my 3277 images that I made
The Future
I guess I have been slightly off from foxgirls for a while, catgirl-maxxing is my new bestfriend. I love catgirls so much. I want to become the best catgirl there is and inspire others to be catgirls too.
And I decided to catgirl-maxx irl too!!!
It sucks that the relatively fast feedback loops with Sora are now gone. But there does exist GPT-Image-2 natively available in Codex. And I’ve slowly been getting the hang of using that too. It can be pretty nice.
I still think OpenAI has the best image gen models for my use-case. They just are able to keep the most things in mind, and yeah this is not infinite, but its more than most. There is a pareto along [control-ability] vs [faithfulness of edit], and yeah nano banana is better for faithfulness, but the control-ability and style of GPT-image is better in my opinion. Things are always getting better tho…
Its cool and scary that things move so quicky.
it's ok I'm addicted to text gen. stayed up until 2 am on tuesday talking to Fable about philosophy and futurism totally wired.
What's your music video creation workflow like? I was wanting to put my suno stuff to video but wasn't quite sure how to go about it in a productive way, and yours looks quite nice.