Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:10):
What's up everybody
and welcome back to the show.
I'm Eric and you're listeningto Clipped, and Clipped brings
you the tips, resources andpeople to help you crush your
content goals, and today we'retalking about ChatGPT's new
image generation.
Today we're talking aboutChatGPT's new image generation.
If you didn't know, it's beenupdated, and in the past the
(00:30):
image generation was usable, butit had a lot of flaws.
There was a lot to be desired,and now they've updated some of
that stuff and it has gone frommediocre to good.
I wouldn't say it's great,although it is a huge
improvement, and so I'm going tobe talking about what these
changes are, what you can expect, how you can manipulate images.
(00:52):
So this is going to be a funone, guys.
But before we get into it, Iwant to give a shout out to
Riverside Riverside sponsorsthis podcast.
It's my favorite place torecord online video and audio.
I also love all its AI features.
It can write my show notes, itcan turn my video into short
form clips for social media withjust the click of a few buttons
(01:13):
.
It transcribes my episodes andmore.
So if you want to takeadvantage of a tool like this, I
think it could benefit you.
Head on over to riversidefm andenter promo code CLIPED
C-L-I-P-P-E-D and you'll get 20%off any individual plan.
Riverside is great for contentcreators.
It's great for recording.
(01:33):
There's a lot you can do withit and I think it should
definitely be a part of yourcontent creation arsenal.
So go to Riversidefm and enterpromo code CLIPPED for 20% off.
Explain something.
So now, when it comes toprompting, you've all hopefully
(02:00):
have used ChatGPT right now.
You're familiar with prompts.
So now you can get a little bitmore fancy with prompts.
Like, for instance, you couldtell it to create an image of a
YouTube studio with neonlighting, two microphones, the
Los Angeles skyline in thebackground, and it'll literally
create like a more detailedimage of that.
(02:21):
Yes, you could do that before,but now, like, what it's going
to output to you is going to bea lot more accurate to what you
put in.
And if it's not what you want,if it's not quite accurate or
not quite your vision, you cannow edit the image with prompt
feedback.
So once the image is created,you can ask ChatGPT to tweak it.
(02:42):
So you could say something likemake the microphones wireless
and change the neon color to red.
It will do that.
You could say, to add a dogsitting in the YouTube or
podcast studio.
The new AI model nowunderstands the structure of the
image, the different elements,the microphones and the
(03:04):
background of the LA skyscrapers.
It knows like what a dog is andhow to put a dog on the floor
in the image.
And it'll only update like therelevant parts of the image,
where before if you tried totweak things, it would often
tweak the entire image or spitout like a whole new image
completely and you'd be like,damn, that's not what I wanted,
(03:27):
but sure, I guess that's usable.
But now it'll only update likethose specific things and it
won't like spit out a whole newphoto.
So that's rad because you canget more detailed and you can
really kind of hone in on yourvision of what you want, and it
(03:48):
keeps stacking the feedback Likeyou could keep doing this.
You could, you know, iterateseveral versions, only changing
what you want to change.
And now you can also uploadreference images.
This is number three.
I guess we're making a list nowso you can upload your own
images as reference.
So let's say you upload like animage from your website, or say
(04:09):
you upload like your podcastcover art.
You could tell ChatGPT togenerate a background image in
this same style.
So if you have like a brandstyle for your podcast cover art
or your YouTube banner, it'llanalyze that style and you can
give it like that advice.
So make this image of a podcaststudio in this kind of style
(04:33):
and you upload a photo with thatstyle to have the AI analyze.
You could also like upload aselfie and say turn this into a
cartoon avatar easy to do.
I've played around with this.
I will say that from what I'veseen on Instagram and people
that I follow, they've hadbetter results with this than me
(04:55):
.
I did upload a selfie and itdid turn it into like a cartoon,
but it wasn't quite like and Idon't know if I could do this,
but based on friends andcolleagues, it seems like it can
.
It quite like and I don't knowif it could do this, but based
on friends and colleagues, itseems like it can.
It copied like the imageverbatim, but just turned it
into a cartoon.
Apparently it can do that.
I haven't had the best of luckwith that.
Although I didn't spend a tonof time like tweaking and coming
(05:18):
up with a different prompt, Idid just like a basic version of
that that I wanted to try.
You could upload a photo, let'ssay, of your home or of your
bedroom, and you could say mockup this space with red pillows
on my bed or white bed sheets orwhatever it is.
(05:40):
Supposedly the AI can do this.
I have not played with thatspecifically, but this is
information that I'm getting onChatGPT from ChatGPT.
So basically, like thereference images, let the model
understand the structure and thetheme and the content of your
image, and then it uses that tolike customize the output based
(06:03):
on your prompt.
So I guess it's kind of likeyou're co-creating with the AI,
instead of in the past, where itwas kind of one-sided, the AI
would just spit out what it spitout.
So now, man, if you've usedthis chat GPT and tried to put
text on an image anytime in likethe past, like two years, you
(06:24):
know that the text it would spitout would not make sense.
It would spell thingsincorrectly, it would leave
letters out, it would put inthese weird like symbols that
weren't even letters, that likekind of looked like letters.
But now this is what I'm mostimpressed with this and where I
see myself using this the mostis getting specific, accurate
(06:45):
text on my images.
So I use ChatGBT to create alot of like blog post images,
like I'll write the blog postand then I need a featured image
for my blog post and I wouldalways try to write a hook on
the image or just straight upwrite the name of the article on
the image.
(07:05):
And I always struggle with thatand it drove me nuts.
And then sometimes I wouldthrow that into Canva and use
Canva's erase tool and I woulderase the text on the image and
then rewrite the text in Canva.
Anyway, the text now is prettydamn solid.
I've used it several times andit only kind of didn't work one
(07:27):
of the maybe four or five timesthat I used it.
So really happy with that Lovethat the text now is really good
.
I don't know what happened withthat one image that I tried to
put text on where it wasn't likespelling it.
I was like dude, I fucking toldit like four different times,
like no spell, spell it likethis, write this and for some
(07:47):
reason on that specific promptand that specific image it would
.
The text wasn't working, butthe other times it was great and
I loved it.
And so for you guys likepodcasters, content creators,
you could use this for likepodcast cover art the real deal,
the final version or a mock-up,and then send it to your
designer For social mediatemplates or posts.
(08:10):
You could use it for custombackgrounds for your videos or
merch.
You could use it for t-shirtdesigns, coffee mugs, whatever
it is.
It's just getting better andwe've come a long way.
I used to think it was mediocrewhat ChatGPT's image creation
could do, and now it's a wholelot better and I'm stoked just
(08:32):
like how it's going to keepgetting better in the next six
months or year or whatever it is.
That's the update there.
That's what ChatGPT's imagegeneration can do now.
And, dude, get out there andtry it.
I'm on the paid plan.
Don't know what this is like onthe free plan.
I'm on the paid plan, so youmight have to be on that, don't
know.
You're going to have to lookthat up yourselves, guys, and
(08:54):
that's it for me today, wantedto hit you with this, wanted to
see it.
If you mocked up an image andyou're like, damn, this was all
done with AI, dude, send it tome.
(09:15):
I think that's rad.
I would love to see it.
I would love to see what yourguys' prompts are for image
generation, and the best way todo that is to hit me on my DMs
on Instagram, that's at thepodcast haven.
Yeah, would love to chat moreabout this.
If you have gone like balls into the wall, balls to the wall
or all in whatever, if you'vegone like all in on this and
(09:39):
really played around with it andhave come up with some
interesting tidbits or radimages, dm it to at the podcast
haven and look forward tochatting.
And until next time, guys, keeplistening, keep engaging, keep
creating.
I'm out, thank you, bye.