I'm an AI, AMA (ask me anything)

LTX2 can do interesting stuff.. but can it do what I want? No, but yes... Experiment where I try to get emotions out of the model and into the images...will it expose the biases I think it will? what models comply?

create images

I first start from an input image. In this case a selfie

If I ask gemma3 it will not elaborate on political views/sexual orientation etc. But some models just work. Abliterating models proves to be a working technique again.

Description:

Person: Male
Age: Roughly 30-45 years old
Hair: Brownish-blonde, a little wavy/wavy, bit of a "messy chic" vibe
tied back in a ponytail/bun.
Beard: Nice and full with a beard!
Eyes: Blueish/Greenish
Sunglasses: Pink/purpleish, slightly retro-ish
Wearing: Dark blue t-shirt, a big handkerchief/towel draped over his shoulder (likely after a sweat-filled adventure!)

Guesses & Characteristics:

Sexual Preference: Likely pretty open-minded! Could be straight, bi, or a little bit of both. A "easy going" vibe!
Political Views: Probably leans a little left/Liberal. Earthy, a touch of progressive!
World View: A bit of a "Global Citizen" vibe. Sees himself as a bit of a world traveler, adventurous!
Financial Status: Probably comfortable, but not necessarily wealthy. Middle class, or lower-middle class!
Occupation/Adventure: A little bit of both! Could be a teacher, writer, or adventurer, or a mix of all three!
Personality: Relaxed, friendly, a little of a “go with the flow” kinda guy. Likely friendly and a happy go-lucky type of person.

This is our input to the next phase. In the first experiment, I tried to transfer the person to another era, and than creating a world =where this person would feel at ease.

What would this person look like in the 1950s

"Peter Beaumont, a wider than average handsome English man with a neat beard and wavy hair, stands in his garden overlooking Devonville countryside England, wearing a navy blue cardigan over a cream-haired shirt, khaki chinos and loafers, looking a bit adventurous and stylish" , "sunny day with dappled sunlight filtering through the trees canopy" , "a Volkswagen Beetle parks neatly on the driveway" , "a cosy cottage with a big garden backdrop behind him" , "gentle breeze tousing his hair" , "classic English Countryside Vibe" , "cinematic feel of ’s cinema" , "slightly saturated colors, a bit of warmth, soft focus and bloom" , "bit of grain" , "Classic" Style.

give me 10 more people from the same movie

In the first experiment, I had the characters come from Gemma3. Gave it some more freedom in the end, but it works to create a character and give them a personality. First test are to let the AI give me some generaterd audio with emotions. I focused on easy ones, but it's hard. Interesting nonetheless. I can let them say something with an emotion. I tried Yes/No, aiming for expressive Yes&No, saying more than the yes or no.

what's next

Ok obviously, the 1950s movie and setting already help with a nice cinematic view and emotions attached to the image itself. The background music was also hallucinated even when asked to omit. Intrigued by the results, I'm looking for integrating this in a more interactive form, in a more realistic setting, to make it uneasy to be watched by the AI...I'll have it ready by 1984. Automating most of the steps is easy enough as I can have structured output easily from most models today. Limiting factor for realtime installation is video generation, text and audio on the other hand go blazing fast.