Menu
What's new
The GetDPI Photography Forum

Great to see you here. Join our insightful photographic forum today and start tapping into a huge wealth of photographic knowledge. Completing our simple registration process will allow you to gain access to exclusive content, add your own topics and posts, share your work and connect with other members through your own private inbox! And don’t forget to say hi!

Levi Strauss uses AI models now

P

Paul Spinnler

Well-known member
More scary is if people just record you with say their phone in a phone conversation, train a perfect AI model of your voice in a second step and then scam others. Real time simulation of other people is possible with a well trained model.
 
P

Pieter 12

Well-known member
Paul Spinnler said:
More scary is if people just record you with say their phone in a phone conversation, train a perfect AI model of your voice in a second step and then scam others. Real time simulation of other people is possible with a well trained model.
Click to expand...
Already being done. There was a piece on the news where someone claimed to have kidnapped a person's child and played the child's AI voice to convince them they had the victim. Turned out to be a total scam, there was no kidnapping. Next thing is blackmail with photos or video of you having an affair or committing a crime.
 
P

Paul Spinnler

Well-known member
1680216316873.jpeg

You can ask Chat GPT to write great prompts for AI generators which then yields excellent results. The input question is simple, the prompts are complex. Saves time in researching good prompts. One AI helps the other AI which helps you to visualize a concept.
 
P

Pieter 12

Well-known member
Paul Spinnler said:
View attachment 202149

You can ask Chat GPT to write great prompts for AI generators which then yields excellent results. The input question is simple, the prompts are complex. Saves time in researching good prompts. One AI helps the other AI which helps you to visualize a concept.
Click to expand...
I get that. But it seems like fishing for something rather than having any control over the creation. Spin the wheel, take your chances. Soon, no one will bother to think.
 
P

Paul Spinnler

Well-known member
Pieter 12 said:
I get that. But it seems like fishing for something rather than having any control over the creation. Spin the wheel, take your chances. Soon, no one will bother to think.
Click to expand...
You can see what comprises a good prompt that way and mix and match. You can take point 5. and put in "midday" instead of "golden hour", etc. You can take "candid photograph" and add it to the prompt 5. etc. Every single comma delimited point in the Chat GPT response list is a building block for a new variant the AI can spit out. If you don't know how to quickly pull together a complex prompt for your purpose you can enlist another language model AI to help you find the right sequence of words and inspire you. You might see "performance photography" and realize that yes, this is what the thing you had in mind actually is, etc.

If you have a fast machine - say an RTX 4090 with fast CPU, you can crank out stable diffusion results in seconds and iterate by playing around with the words. Iteration x speed are a powerful method to get to great results very fast.

I think you have a lot of control and it is super easy with not much "talent" or "skill" needed.

Next-gen AI will also have more and more advanced control features meaning you can add sample imagery or other style constraints on top to constrain the result. Firefly already lets you say whether you want photo realistic or not, for example.

With advances in computing power and more sophisticated models trained on higher-res imagery we should soon have access to 4k native image generative AI - a good 4k base can be retouched in PS or upscaled to create fine art you can print on a wall.

AI models are trained on millions of images - the possibilities to get what you want are vast and it is not too difficult to get the hang of it.

Everyone will be a fine art artist in no time!
 
Last edited:
PeterA

PeterA

Well-known member
I don't think Ai will change people's interest in recording their own stories or making documentary records of this or that and people who want to will still go to places to make a photo that has already been made.
The image I posted was made in 5 seconds first use - the hands aren't right - but I can crop those out - I'm happy to pay an extra 10 or 20 a month for firefly commercial usage license as I am happy to pay 20 a month for GPT4+

anyone in business these days has to generate an online media channel / vlog/blog and update content for user engagement - most of the blogs are repetitive messenging devices headed by clickbait headings - these tools just make the production process a matter of minutes a day rather than a chore - these tools also remove a lot of co, competitive advantage held by large corporations as far as customer engagement goes - by reducing the fixed costs of acquiring the mundane stock image or ( coming soon enough) short video clip.

In my area of investment management we are already working on our own proprietary plugins to use with various API links in Bloomberg and Factset as examples - in order to make research and development less dependant on in house storage and in house grunt work like database maintenance and working through nonsense like SQL.
NLM are already capable of digesting 400 page quarterly reports from companies extracting key words from various officers and standardising financials in order to highlight what is going on - a process that used to require expert analysts working away for hours to come up with can take less than minutes ..

ALL this means my costs are likely to go down for relatively low value work and significantly up for higher value add work - demand for real intelligence will always exceed supply.
 
J

jduncan

Active member
Pieter 12 said:
I have not read the website--too much mumbo-jumbo. But if Levi's is using it to advertise their clothing and using 3D models of the clothes instead of the actual thing, that would be misleading advertising. A good lawyer could have a go at them, especially since they have deep pockets (pun intended).
Click to expand...
That is a good point, but if they said these are AI models it won't be illegal. They used to use paintings on the cover of magazines. Clearly, many of the affected will be women and society has a tendency to protect them, so they could pass a law, so who knows.

Best regards,
 
P

Pieter 12

Well-known member
Upon further thought about my post on using AI models and CG nothing for advertising, I guess it is not that much different than the use of drawings in the past, except people might--and most probably will--be fooled into thinking they are looking at photos of real people wearing real clothes. I hope at some point a disclaimer would be required for any AI imagery. As well as some sort of imbedded token to indicate an image was created using AI.
 
Mexecutioner

Mexecutioner

Well-known member
Paul Spinnler said:
More scary is if people just record you with say their phone in a phone conversation, train a perfect AI model of your voice in a second step and then scam others. Real time simulation of other people is possible with a well trained model.
Click to expand...
For this to work well and be modeled correctly you need several minutes of high quality recording, definitely not from a cell phone. We have been able to do this to replace or add dialog lines of deceased characters that are not able to do ADR. I imagine in a few years you will need a smaller sample of dialog in order to achieve what can be done now.
 
You must log in or register to reply here.
Top