10/7/25 - Kicking the Tires on New AI Tools: A First Look at Building Workflows in Weavy.ai

Description

Chapters

Transcript

0:01 Get your butt over here.
0:06 [Music]
0:25 You and I all alone.
0:29 Sunday morning he ran home.
0:34 Skes blue and a coffee strong. It's
0:37 true.
0:39 [Music]
0:42 Then I open my eyes to a dream realized
0:46 in front of me.
0:50 Then I got a clue what in the world has
0:53 happened to me.
0:56 [Music]
0:58 Think I think I'm happy
1:01 like first day summer vacation. Happy
1:05 got to get a little rest and relaxation.
1:08 Happy
1:10 like choir on Sunday morning singing
1:12 true.
1:19 [Music]
1:32 in a westerly
1:35 direction.
1:36 [Music]
1:39 This car is my train.
1:42 [Music]
1:44 I' been driving. I've been wondering
1:48 what it is I'm running from again.
1:52 Feel like an 80y old man
1:57 holding on to 29.
2:01 And up ahead on that horizon
2:05 is a California line.
2:07 [Music]
2:10 [Applause]
2:14 Up ahead of trucks carrying a wide load.
2:18 Three fat house cut in half.
2:20 [Music]
2:22 Cute little front door into two windows.
2:24 My lord
2:26 ain't sure whether to drive should last.
2:30 You see, I broke a home up myself once.
2:34 As I stumbled to that door,
2:38 I read a note by the dawn light
2:42 said, "Don't you come around here
2:45 [Music]
2:47 anymore."
2:48 [Applause]
2:49 Well, I've had enough
2:54 freedom on the rock.
2:56 Never was good with decision.
3:00 At least that's what I've been told.
3:03 [Music]
3:11 Oh, good people. Good people of the Tik
3:14 Tok, of the YouTube, of the Twitter, of
3:16 the LinkedIn. What is going down? What's
3:18 happening? was shaking. Hope everyone's
3:20 doing well
3:22 on this fine Tuesday night. We had a
3:24 lovely AI salon tonight. Had a good
3:27 crowd.
3:28 We had Curi from New SP No Spoon Studios
3:34 who is she is a an impressive woman
3:38 building,
3:40 delivering, building and shipping.
3:42 Building and shipping.
3:44 She's got [ __ ] out in the world. She's
3:46 just doing it.
3:47 She's just doing it, people. This is
3:50 This is the new way.
3:53 There's There's no There's no waiting.
3:56 There's no planning. She might have
3:58 planned,
4:00 but she's delivering.
4:03 [Music]
5:00 [Applause]
5:04 She came home again like slow moving
5:08 cold fronts. His
5:13 beard was warmer than a look in her eye.
5:19 [Music]
5:28 She said, "Give me a love that don't
5:31 freeze up inside.
5:33 [Applause]
5:34 [Music]
5:41 You said I have melted
5:43 in my time, dear.
5:47 [Music]
5:48 But to sit next to you, well, I shiver
5:51 and shake.
5:56 And if I knew love, well, I don't think
5:59 I'd be here.
6:04 Asking myself if I've got what it takes.
6:10 [Music]
6:13 Got to hit the right fretboard
6:17 or the right fret on the fretboard.
6:21 [Music]
6:26 All right, let's uh let's get rolling.
6:30 Let's get rolling. Let's get rolling.
6:32 Let's get rolling. Let's do something.
6:34 Everybody ready to do something?
6:37 What's happening? Oh, man. It's a light
6:39 light crowd tonight. All right. Well, it
6:41 is late.
6:42 So, so uh we'll just have us a we'll
6:45 have us a a nice short little thing.
6:47 Maybe maybe the crowds will come rushing
6:49 in. Um I figure first thing we'll do is
6:52 um
6:53 we'll play Gareth. Gareth did a uh an
6:57 entry for the door brothers contest
7:01 that's called the door or door like
7:04 there were three categories and one of
7:05 the categories was a door and so he made
7:08 this door video. So let's watch Gareth's
7:11 door video.
7:14 Your fantasies are our desire.
7:20 [Music]
7:26 [Applause]
7:28 [Music]
7:33 City rain falls down
7:36 on concrete in dreams.
7:41 Thousand faces past.
7:45 Nothing's what it seems.
7:49 One moment in the crowd,
7:53 one look, one glance.
7:57 Sometimes that's all it takes
8:03 for cir
8:06 behind closed doors
8:10 in the glow of blue light.
8:15 Searching through the dark
8:18 for something that feels right.
8:23 Fingers trace the glass.
8:27 Following the thread
8:31 into someone else's life
8:35 where angels tread.
8:39 >> All right, let's have some fun.
8:45 Step inside the dream where nothing has
8:48 a cost.
8:53 [Music]
8:55 >> Hey.
8:56 >> Hey, babe. How are you?
8:58 >> Good. Now,
9:00 [Music]
9:03 >> I've been looking forward to this.
9:07 >> Me, too.
9:08 >> I have a secret.
9:10 >> Oh, yeah. What's that?
9:12 >> You shouldn't steal photos of women and
9:14 not tell them.
9:16 I'm not that kind of girl.
9:20 [Music]
9:22 >> Back into the rain.
9:24 [Music]
9:26 But now the city's changed.
9:30 Every wall and window
9:35 has been rearranged.
9:36 [Music]
9:39 She walks through the same street,
9:43 sees what I have done.
9:47 The game was rigged from the start.
9:51 There's nowhere left to run.
9:54 >> Oh my god. What is going on?
9:59 keeps falling down
10:03 on concrete and
10:08 what we think is cracked left behind
10:11 your eyes.
10:28 City rain falls down
10:33 >> at the door. Your fantasies are our
10:36 desire.
10:39 >> That's cool.
10:41 There you go. Uh
10:46 that was pretty cool. I don't know if
10:48 Gareth's here. Nice job, though. Um, are
10:51 you paying $200 to test this? To test
10:54 what? That that video model. That was
10:56 not my video. That was Gareth's video.
10:59 Um, my only comment on his video was I I
11:02 wouldn't put, you know, 100% AI at the
11:05 end. I I don't think it's anyone's
11:07 business if you used or not. They can
11:09 figure it out. If it's a if it's a good
11:11 story, it's a good story. So, but uh but
11:15 yeah, nice. Good luck with that.
11:19 Um, oh wait, Sora 2, $200 for the API.
11:24 Is that true?
11:26 Is that what you were asking? The the
11:28 Sora thing, the API is 200 bucks a
11:32 month. Might be. I think it's 10 10
11:36 cents a second to render. So 200 bucks
11:40 is 10 seconds times
11:46 uh 200.
11:49 So however many 2,000 seconds and then
11:54 that's some amount of minutes.
12:00 Yeah. If you're going to get into doing
12:01 video at any kind of scale, you got to
12:03 have you got to have yourself a [ __ ]
12:05 budget with a budget with a capital B. A
12:09 capital B there, boys. Boys and girls.
12:14 Um, I want to go play with something.
12:22 Is it Weebi?
12:24 Wee AI? I think it's weebi
12:28 powered workflows built for creative
12:30 pros.
12:31 All right, let me share this tab.
12:42 If you're new here, my name is Kyle
12:43 Shannon and this is the AI learning lab.
12:45 We are going to go share a screen here.
12:52 Weevi artistic intelligence. So, this
12:55 has got one of those really [ __ ]
12:57 annoying interfaces
13:00 um that are designed by 21 year olds and
13:04 they're like, "No, we can make the font
13:06 smaller.
13:09 Screw the old people.
13:12 [Laughter]
13:18 Oh my god.
13:21 Every little fun." Um, I think I think 8
13:25 point 8 point font's fine.
13:30 I'm totally new to this. I know. Me too.
13:32 So, I have heard about Weevi now from
13:35 two people in two days and both people I
13:38 respect. One was Retroun Punk um who's
13:41 who's uh you know does digital and AI
13:44 and creative at an agency and the other
13:46 one was Liz Miller Gerschfeld who host
13:49 co-host the salon with me. Um,
13:54 they both told me about this today. So,
13:57 I figured I'd go we'd go check it out.
14:00 Um, I talked about this last night. I
14:02 talked about it tonight at the salon.
14:06 I know nodebased
14:09 interfaces have been around forever and
14:12 you've got make and you've got Zapier
14:14 does pseudo nodebased [ __ ] and you've
14:17 you've had databases that are nodebased.
14:19 We've had nodebased interfaces
14:21 interfaces for years, but
14:25 there was something about yesterday when
14:28 OpenAI launched agent builder. And even
14:31 though agent builder looks like it's a
14:32 bit rushed, it's not it's not we're not
14:34 going to be able to build anything in it
14:35 in here. It's just it's too wonky right
14:38 now. Um,
14:40 and then 11 Labs did a nodebased
14:43 workflow maker. like there's just it
14:45 feels like
14:47 the world's going to get pretty
14:48 complicated pretty fast
14:52 and there's there's a part of me that's
14:55 like we don't need to do this because 3
14:58 years from now all of this [ __ ] will be
15:00 behind a simple interface
15:02 but I just kind of feel like this is the
15:04 equivalent of learning math so you can
15:08 understand the power of a calculator you
15:10 know um I it's just feeling to me like
15:14 we've got to start something. So,
15:17 so there you have it. There you have it.
15:20 There's Daniel. Hey, Danielle.
15:24 Uhuh.
15:26 Is it lagging for anyone? Which which
15:29 one's lagging? Is it um Tik Tok or
15:32 YouTube?
15:35 Agent builders. They're sort of agent
15:37 build. Some are agent builders. This
15:39 one's just a workflow builder, but
15:40 you'll you'll see if I can figure out
15:42 how to use it.
15:43 Where? When? What? It's called Weevy
15:46 app. Awesy.
15:54 Tik Tok. No lag. Fine now. Okay, great.
15:57 Beautiful.
16:10 Multiple image models. V2.
16:18 Okay. Wait, why? Oh, because it it
16:20 opened a new tab. God damn these sites
16:22 that open new tabs all the time.
16:29 Very easy. Repeat four times, please.
16:32 It's fine for me. Tobias, hello. Hello.
16:35 Hello. Hello. Hey, everyone.
16:40 doesn't seem to be lagging for me. All
16:42 right, I think the lag lag issue is good
16:44 and gone. We're good.
16:47 Okay, so what we've got here,
16:51 we've got a prompt,
16:55 and that prompt
16:58 has an input
17:03 and an output.
17:05 And then we've got GPT image one, Flux
17:08 Pro Ultra, Miniax,
17:12 Lumaf, Photon, Recraft V3, Mystic,
17:16 Ideog, Stable Diffusion, Google Imagine
17:19 4, Fall AI, Higsfield Image, and Fall AI
17:24 Flux Crea, and the other one was Fall AI
17:30 Bite Dance Seeddream V3. And I think
17:33 that's up to V4 now. So what we might
17:36 want to do is update some of these here
17:39 nodes.
17:40 And then what happens then? You send it
17:45 to Mystic. Oh, and it makes videos.
17:47 Okay.
17:49 All right. I I sort of get what's going
17:51 on here.
17:53 Duplicate my files. But I I think I want
17:55 to start one from scratch.
17:59 So the the reason for this guys is
18:08 well well there's a couple of things. If
18:10 you're working with
18:13 clients, like I just did a job for a
18:15 client,
18:18 that
18:21 if the client had asked asked for my
18:23 prompts or what models I used and what
18:26 models I used when, I was just all kind
18:29 of managing it in my head. I'm like,
18:31 "Oh, let me jump over to this model. Let
18:32 me jump over to that model. Let me jump
18:34 jump." And if something didn't work, I
18:36 would just sort of give up on it on this
18:37 thing and then just copy the prompt and
18:39 go to something else. What this allows
18:41 you to do is put together test
18:43 environments where you can put in a
18:44 single prompt, test it across multiple
18:46 models and see which one works best, but
18:49 then you've got a a a workflow, a
18:52 repeatable workflow and a and a sharable
18:54 workflow that you can share with clients
18:56 or with other people, with other
18:58 creators.
19:00 I'm viewing a readonly version.
19:02 Duplicate my files. Okay. So, let's go
19:04 back to Oops.
19:07 Let's not do what I just did and destroy
19:10 that.
19:13 Looks like we got some new folks in
19:14 here. Welcome everyone. My name's Kyle
19:17 Shannon. We are on Weevie.
19:20 Well, we're going to be here in a
19:22 second.
19:24 and Weevie other than being a bad name.
19:29 It's like I I sort of get what it means,
19:31 but it was like that was the only domain
19:33 you could get. Come on.
19:40 I'm sure the founder of Wee is on here
19:42 right now. Hey man, we we paid a
19:45 consultant a lot of money for that.
19:47 Okay, Sora 2 has landed in Weevi. Check
19:49 it out in action.
19:52 Okay, nothing here yet. Start weaving.
19:54 Create new file.
19:57 Okay.
19:59 Welcome to the Weevie editor. Skip.
20:02 Start tutorial. I'm going to do the
20:03 tutorial. We're going to do this
20:04 together. All right. So, s here's what I
20:07 want you to do. Go to Weevi. Weevi
20:10 app.weeevi.ai.
20:12 Make an account. Hey, Kyle Achilles.
20:15 What's happening?
20:17 Please post a link. Okay.
20:20 Um,
20:23 Brandon's doing that
20:35 mouse,
20:38 hold the space bar
20:41 and click left to drag.
20:44 Okay. Oh, there you go.
20:47 Hold the space bar and left left click
20:50 to drag.
20:53 Okay.
21:01 All right.
21:06 This is prompt mode. Oh, this is a
21:08 prompt node. It holds a text prompt that
21:11 guides your image generation. Okay,
21:13 great. Next.
21:16 Hold command
21:19 command control
21:21 and scroll mouse wheel to zoom.
21:26 Okay. So just Okay. That does that. That
21:29 does that. Got it. Okay.
21:34 You have to
21:39 you have to load
21:44 on weev app
21:47 doesn't load. I don't I don't know what
21:49 that means. Brandon,
21:52 >> you're giving out app.weeevi.ai
21:55 that doesn't load unless you're logged
21:56 in. Other it's just weebi.ai if you're
21:58 not authent. Oh, load like load. I get
22:01 it now.
22:04 what you said. Okay. Drag and drop fast
22:07 flux
22:09 to to add a menu to your thing. Okay.
22:13 Got it. Fast flux. Got it. Click and
22:16 drag from the output handle
22:19 to the input handle. Got it.
22:25 All right.
22:27 Click run model. Where's run model?
22:30 150 credits. No active runs.
22:36 Click the
22:45 wh
22:47 why am I not seeing the run model
22:49 button?
22:50 Oh, because Okay, let here's why.
22:55 because it's it was down at the bottom
22:57 right hand corner of that box in like
23:01 sixpoint type. There it is. All right.
23:09 All right. So, my first wee. Yay. You
23:13 did it. Yay.
23:15 All right.
23:18 Start working on this workflow. Okay.
23:20 Cool.
23:22 And then we can say guidance flux fast
23:25 is the model aspect ratio we can do 16
23:27 by9s.
23:29 I can say run again.
23:32 All right. So it's one credit per
23:36 per run
23:38 it looks like.
23:44 All right. And then we're going to go
23:46 where are we going here? Are these image
23:48 models? Yes.
23:51 Imagine three
23:53 three credits.
23:55 I see. Highest quality, six credits. All
23:58 right, I'm starting to get this. There's
24:01 stable diffusion.
24:03 Where's Nano Banana?
24:06 Search.
24:08 Nano.
24:10 Nano Banana.
24:13 Bang. Out comes Nano Banana.
24:23 Run model.
24:34 All right, that's kind of cool.
24:37 Aspect ratio default.
24:41 You can do 16 by9 here.
24:44 Total cost four credits.
24:48 for Nano Banana.
24:50 I'm such a cheap bastard. I keep I see
24:52 those credits going down. I'm going to
24:54 be like,
24:56 "You son of a bitch."
25:02 All right. What else? What's in our
25:03 toolbox? Let's go look at our Let's go
25:06 look at our toolbox.
25:10 Compositor Painter. Wonder what a
25:13 painter does.
25:18 What's a painter do?
25:21 So, we're going to do the output of that
25:24 into the input of that.
25:29 Oh, and then you can paint on it.
25:33 Except I don't know what I'm doing. So,
25:37 so we're going to delete this.
25:43 Okay. Beautiful.
25:47 Um,
25:51 [Music]
25:54 mat extractor mask by text video mat
25:57 prompt
25:59 prompt concatenator. So you could take
26:01 two prompts and put them together.
26:03 Prompt enhancer.
26:06 Well, let's do that. Okay, this is so
26:08 this is kind of fun. So how how do I
26:10 break these? Do I just click on them and
26:12 delete them? Yep.
26:14 Okay, easy. So, we'll go prompt enhancer
26:20 and we're going to go from
26:22 shitty prompt to prompt enhancer.
26:25 And then we're going to go from prompt
26:26 enhancer out here.
26:31 All right. And so, we're going to do a
26:32 shitty prompt of 70s.
26:36 We'll do 70s
26:38 muscll car.
26:44 resto mod in abandoned factory. Shitty
26:48 prompt, but it's a classic.
26:51 And then I'm going to say run model. So,
26:53 let's zoom in on this thing. This is
26:56 actually pretty cool. Um,
27:00 yeah, I'm I'm I'm feeling like I'm
27:03 feeling like there are there are skills
27:06 to be had here.
27:09 All right. So there we do that. We run
27:12 the model.
27:15 Okay. No active runs prompt enhancer
27:18 models instructions. Your job is to
27:19 write prompts for textto image models.
27:22 Your input will be a general description
27:24 for the scene. You should write a
27:26 detailed prompt without any additions no
27:29 longer than three sentences.
27:35 Um we'll do no longer than five
27:38 sentences.
27:40 Um,
27:41 you should include
27:46 um
27:48 detailed
27:51 scene
27:56 subject.
28:04 Um,
28:07 scene, subject,
28:10 and
28:14 time of day.
28:17 You should include lighting,
28:22 visual style,
28:25 eg
28:29 um
28:31 impressionist
28:34 painting, um
28:36 uh
28:38 ectochrome
28:43 film,
28:44 etc.
28:51 What else do we want to include?
28:54 And
28:57 camera angle.
29:01 Camera angle. Okay. So now we wrote a
29:03 new prompt. So this is that's that's
29:06 shitty prompt writing. But this is
29:08 effectively prompt engineering, right?
29:10 We're creating an application here where
29:11 you put in a shitty prompt. You run your
29:14 model. Boom.
29:17 Capture a low angle
29:20 shot of a meticulously restored 70s
29:22 muscle car behind the soft diffused
29:25 light filtering through the grimy
29:27 windows of an abandoned factory which
29:30 I'll often write that the car's gleaming
29:32 paint contrast sharply with the
29:34 crumbling concrete bump sim simulate the
29:37 the nostalgic look of Kodakchrome film
29:41 emphasizing warm tones. Okay, this is
29:44 cool.
29:45 Um,
29:46 all right. So, this is Gemini Flash Nano
29:49 Banana. Can I change that
29:52 or do I have to bring out another thing?
29:54 I have to bring out another thing. Let's
29:55 go get a different model.
29:58 Chat GPT image one. Uh, let's do GPT
30:03 image one edit.
30:06 Well, GPT image sucks. Let's do uh is
30:10 Grock in here? Of course not, because
30:14 Elon Musk's a dick and he doesn't want
30:15 anyone to play with his toys. Um
30:21 um
30:25 well, let's let's go look at them
30:28 models.
30:31 Higsfield image. That one's kind of
30:33 cool. 21 credits. Oh, no. Wait.
30:41 Commercial use 21 credits. Good lord.
30:45 Imagine four
30:51 is six credits.
30:55 Imagine three fast is three credits.
30:58 Let's drag that one out there.
31:01 So, we'll grab from here. We'll go into
31:04 there. Oh, cool. You can do a negative
31:07 prompt, too.
31:09 No, that's pretty slick.
31:15 Oh, ideog.
31:17 How much is that? Four credits. That's
31:19 pretty cheap.
31:21 Ideog V3 character. Create consistent
31:25 character.
31:28 Oh, that could be a cool thing we could
31:30 do. You could start with an image. So
31:34 you could basically go image prompt
31:36 image in image prompt prompt enhancer
31:39 image
31:41 and then take that image into
31:45 like of a person
31:47 and then use ideoggram v3 character and
31:50 have that person in a bunch of different
31:52 scenes that each have their own prompt.
31:55 Yeah, that's kind of cool. All right, I
31:56 like that.
31:58 Um
32:02 all right so let's do aspect ratio 16
32:05 by9
32:06 16 by9 16 by9
32:12 output quality runs
32:15 all right run model
32:19 run model
32:27 run model
32:30 and then I think there's a way I can I
32:32 can run the output into something
32:34 that'll send it to all of them
32:36 automatically.
32:38 So now we're down to 132 credits. By the
32:41 way,
32:51 there's Nano Banana.
32:54 Nano Banana is pretty [ __ ] good.
32:58 This is Google imagine three fast.
33:06 The hell is that car?
33:13 That's like a 74 Mustang ass end, but
33:17 it's wider.
33:20 That's bizarre.
33:23 Okay. Um All right. Let me shrink that
33:26 down. Can you organize these? How do you
33:29 organize them? What do you do? Do I
33:31 right click? No.
33:36 Uhuh. Uhuh. Uh uh uh.
33:39 Is there an organized thing?
33:42 Prompt import quick access.
33:49 Import multiple lures. I wonder if you
33:51 can create lures. I bet you can.
33:57 Blur
33:59 mask
34:01 prompt prompt concentrator.
34:04 Join multiple text inputs into one
34:07 output.
34:08 Prompt enhancer.
34:11 Image describer. Very cool. Video
34:13 describer. Text iterator.
34:17 Iterate over a list of values. Import.
34:20 Export.
34:22 Router.
34:27 All right. So, let's take
34:34 Right.
34:38 Oops.
34:42 I do. No. This. No. This. This. This.
34:49 Okay.
34:51 And then I go and I go and I go
34:57 and I go
35:00 and I go and then I go this down here
35:02 into that router. Right. Isn't that
35:06 right?
35:07 I think that's right.
35:11 What's this assets?
35:18 Okay. So, let's go do
35:22 let's do a shitty another shitty prompt.
35:25 Uh, let's say um
35:29 um grizzled
35:32 old
35:34 New England
35:37 fisherman
35:40 smoking
35:44 near the docks.
35:47 Okay.
35:50 And then I guess I go over here
35:57 and I run it. Bang.
36:02 Bang.
36:04 Is it running?
36:09 A weathered New England fisherman with a
36:12 long white beard sits on a wooden crate
36:14 near a bustling dock at dawn.
36:17 He wears a thick wool sweater over his
36:20 hat, puffing thoroughly on it. Does
36:22 anyone have questions? Is this making
36:24 sense what I'm doing? Kicking the tires.
36:25 Yeah, totally kicking the tires right
36:27 now. I made it early but grossly
36:29 misjudged my alcohol intake.
36:35 That is beautiful. I love that. Um, how
36:39 do I organize this [ __ ] Anyone? Does
36:43 anyone have any idea?
36:46 So, I select them all.
36:49 Select them all.
37:28 I need three three more lifetimes to
37:30 figure out how to use all this stuff. I
37:32 know it. Well, this is this is the uh
37:36 this is the rub right now. So, here
37:38 here's here's I think the antidote to
37:40 that, Jeff, is
37:44 just figure out what you want to do
37:47 first. Don't worry about the [ __ ]
37:50 tools.
37:53 Like I want to learn I want to learn how
37:55 to do this.
37:57 Like the what's what's popping in my
37:59 head right now is this is cool, but it's
38:01 going to be really expensive and it's
38:02 going to be really tedious.
38:06 But what what what this actually is is I
38:09 bet there's a lot of people out there
38:11 that have set up workflows that you can
38:14 just go use and they're pretty [ __ ]
38:16 cool. So we should probably go look at
38:18 that now that we generally understand
38:20 how it works. What I don't get though is
38:22 why
38:26 duplicate delete lock. Why is this
38:29 router not routing?
38:32 Because I ran the model.
38:41 How do I get it to
38:54 I don't get it. I don't get it. I do not
38:58 understand
39:03 anybody.
39:06 3D models,
39:14 community models, image upscale,
39:16 clarity, video smoother, ID
39:19 preservation, control.
39:23 Good [ __ ] lord, there's a lot here.
39:27 Video utilities, video concatenator.
39:31 Oh, that's cool. So you could generate a
39:33 bunch of video pieces.
39:36 So actually what you could do
39:39 Oh, this is this is probably some
39:41 version of what Carrie did and or Kerry
39:44 did in her app that she demoed tonight
39:46 in the salon
39:50 except she had uh what you call it build
39:52 it for. She probably didn't build it
39:53 here. But you could build a workflow
39:56 that takes an input,
39:58 generates a script with eight different
40:00 scenes that tell a narrative arc,
40:03 generate images for each one of them,
40:05 start and end images for each one of
40:07 them, run a video model that takes input
40:11 and output images, run both of those in
40:13 there,
40:16 generate the clip, concatenate all the
40:18 clips, and generate a movie. Yeah, you
40:22 could you could totally do that with
40:25 this model. What's video utilities
40:30 input file result
40:34 frames per second
40:39 basic video processing convert input to
40:42 MP4
40:43 input to GIF. Tik Tok question. Yes.
40:47 Yes. someone had a question on the Tik
40:49 Tok.
40:51 Wouldn't it be easier to just write a
40:53 prompt
40:56 in a difficult platform? I'm not getting
40:58 this. Okay, so here's here's
41:03 here's the the the trick source camp.
41:06 Um,
41:08 let me get rid of this thing. Get rid of
41:11 that node. So, what I don't know how to
41:13 do right now is um
41:20 is autotrigger this router. But, but
41:23 basically what I've got here is a shitty
41:26 prompt and then there's a mo a node here
41:29 where I can improve the prompt and I can
41:31 control that. So, I write a shitty
41:33 prompt. It automatically writes a better
41:35 prompt and then I've selected three
41:37 different models over here. And so from
41:40 one prompt I get three different
41:42 outputs. And I could have nine different
41:43 outputs or a dozen different outputs. Or
41:46 I could have a single image output or or
41:49 sing single image input and then have
41:52 eight different
41:54 um scenes with that same character in
41:57 it. Let's go let's go look at some
42:00 different um
42:03 some different workflows.
42:12 Maybe it didn't auto route because it
42:14 wasn't new images. You redid them. I
42:16 don't know.
42:18 This This is the kind of [ __ ] like this
42:21 is the kind of [ __ ] with software,
42:23 right? Like you can learn whatever I
42:25 just learned in 15 minutes, but it's
42:27 going to take me three hours to figure
42:29 out how to auto route auto trigger those
42:31 things.
42:34 Back to my files. Okay.
42:37 So workflow library
42:41 weave welcome
42:43 multiple image models v2
42:46 editing images compositing node image to
42:50 video
42:52 camera angle ideiation. That sounds
42:55 cool.
42:58 Yeah, look at this. Okay, so here source
43:01 camp. Look, let's let's go look at what
43:03 this does.
43:06 You put in a prompt.
43:08 You're given a reference image of a
43:09 scene.
43:12 Any LLM
43:15 using the provided image. Okay. Wait.
43:21 Oh, okay. Wait. So, it starts up here.
43:24 Got it.
43:26 Oh, I see. I see. Okay. So, Oh, I'm not
43:31 sharing. God damn it. Hang on. So,
43:33 source camp. Look at this. and all of
43:36 you look at this. So,
43:39 so we've got in the upper leftand corner
43:42 here
43:44 is a prompt that generates an image and
43:47 then that image it looks like
43:51 comes into this thing which generates
43:56 this. No, no, this is a prompt for this.
44:02 Okay. So the image the image comes into
44:04 this LLM
44:06 which uses this prompt to do something
44:08 with it
44:10 and then it goes out into
44:15 use the provide using the provided image
44:18 keep the style and subject similar
44:21 list
44:27 and then this is Gemini 2.55 flash nano
44:31 banana
44:32 And the prompt is coming in from oh I
44:35 see here. So if I click on this
44:45 using the provided image.
44:49 Ah
44:51 show more.
45:02 Ah, each prompt must must begin exactly
45:05 with using the provided image, keep the
45:08 style and subject details similar and
45:11 modify the camera to match this. After
45:15 that phrase, describe a compositionally
45:18 different camera perspective.
45:21 Okay.
45:23 So then it makes a pile of prompts and
45:27 then those prompts go into this text
45:29 array.
45:32 Oh, I see. Oh, this is cool. So, okay.
45:35 So, and it's split by it's split by an
45:38 asterisk. So, in the text right here,
45:42 when it finds an asterisk, which right
45:44 at the beginning of this paragraph,
45:45 there's an asterisk. So what this what
45:48 this text array does is it takes this
45:50 longass prompt and it splits it into
45:53 individual nodes
45:56 individual and then and then this array
46:02 these are the individual prompts for all
46:05 these different windows. Okay, I got it
46:08 now. Okay, cool. Watch. So watch. Let's
46:10 let's just use this once. This is going
46:11 to cost us a lot of credits, but that's
46:13 okay. You're in read only. So like
46:15 duplicate to my files. So now we have a
46:18 version of this that's ours.
46:22 So now I can go in. So now it'll start
46:24 to make sense. Source camp. Um
46:27 let me see. Um
46:30 70s
46:32 resto mod.
46:35 Let's see. 70s
46:38 muscle car
46:40 restood.
46:44 Um, idling in an
46:53 abandoned
46:56 factory.
46:59 Um,
47:01 the driver
47:04 in the car has
47:07 a mustache.
47:14 um
47:16 and sideburns
47:20 and has seen
47:24 better days.
47:27 The car looks great.
47:31 Um great.
47:35 Let's see. Uh,
47:38 sunset light
47:43 breaks through the smudged
47:48 factory windows.
47:52 We'll leave it at that.
47:55 Okay. Now, how do I do all these things
47:59 run automatically
48:02 if I run this one thing?
48:05 We'll find out, I suppose.
48:27 Okay, there's our car.
48:30 There's our dude with a mustache who's
48:32 seen better days.
48:43 So then I guess I run this model
48:54 using the provided image. Okay, so
49:00 there's got to be to a way to run all of
49:03 these in a row, right?
49:08 Anyway, all right. Here's one. Here's
49:11 one.
49:13 Here's one. Here's one. This is going to
49:15 eat up all of our credits,
49:18 but this is going to be cool.
49:20 Here's one.
49:22 Here's one. Here's one.
49:26 I don't know what I'm doing. I'm
49:28 clueless. Clueless. Clueless.
49:31 I don't know what I'm doing. But look,
49:35 here's the undercarriage of the car.
49:36 This is actually really cool. Here's the
49:38 hands on the steering wheel. Here's the
49:41 shot. That's not the worm's eye view,
49:42 right? There's the There's the That was
49:44 our original shot, right?
49:49 I don't know why these two didn't get a
49:51 prompt. Oh, they're still rendering.
49:53 Look, there's the bird's eye view.
50:00 This is cool.
50:03 This is very cool.
50:12 Huh?
50:15 Uhoh. What happened? Gemini could not
50:18 generate an image with a given prompt.
50:20 Try again.
50:38 weevi.ai. Yep, that's correct. Really
50:42 curious how anyone uses this at work.
50:47 Is it just an experimental storyboard?
50:50 Um, I think it's a bit more than an
50:52 experimental storyboard. I think, well,
50:55 for one thing, you can share these
50:56 workflows.
50:59 So, so like what we're doing here is
51:02 like I don't I don't know what the [ __ ]
51:04 I'm doing, so I don't I don't quite know
51:06 how to troubleshoot this thing,
51:09 but this is write a single prompt and
51:12 get
51:14 10 different shots, 10 10 different
51:16 angles of that of that shot.
51:19 Um,
51:22 so this is a storyboard generator,
51:24 right? Or this is a this is a shot
51:27 possibility generator.
51:30 So this could be a component in a larger
51:33 application where you actually call this
51:35 this workflow
51:37 and behind the scenes it's it's doing
51:40 all the code and doing all the API
51:41 calls, right? And so all you would do is
51:43 I'm sure there's a way
51:46 I'm sure there's a way to grab the URL
51:49 or the something from this. Duplicate
51:52 delete lock.
51:56 I don't know. Maybe there's a an API
52:00 hook somewhere.
52:02 Video smooth. That's all AI [ __ ]
52:06 3D models. That's video models.
52:10 Image models. Toolbox
52:13 levels compositor crop blur channels
52:17 extract video
52:19 mask prompt prompt con concatenator
52:23 prompt enhancer image describer video
52:26 describer. That's cool.
52:29 So like here's yeah this is so this is
52:32 something that that could be really
52:34 interesting, right? You
52:39 you upload a video to something. You
52:42 analyze the frames of that video. Figure
52:44 out where the the visual story
52:48 story changes are.
52:50 Then you do the same with dialogue. Then
52:52 you marry those two. You could you could
52:54 storyboard out an existing video
52:56 automatically.
52:58 Text iterator. I don't quite know how
53:00 that works. Import Laura.
53:03 Import multiple Lauras
53:06 data types input number text list toggle
53:10 seed array
53:14 and then there's a bunch of images image
53:17 models
53:20 enhance image.
53:26 Yeah, this is this is
53:29 another thing this could be used for is
53:30 testing, right? You could put together.
53:34 God, man. You could spend [ __ ] weeks
53:36 in this thing. If you really wanted to
53:39 understand the state-of-the-art for
53:42 image, video, audio,
53:44 and what's possible with it, if you
53:46 spent a month with this thing,
53:50 you'd actually understand what's going
53:52 on. That's pretty crazy. All right. Um,
54:00 looks like the evidence of a bad
54:02 detective series.
54:04 Well, two two of my things are not
54:05 running because the prompt is bad.
54:09 Which which two are they? Uh, five and
54:11 six.
54:16 I don't know what order these are.
54:20 So, wait. One, two, three,
54:24 and then four, five, six. Three and six.
54:28 So if I go
54:31 here,
54:34 I go one, two, three. This one
54:39 to match wide establishing shot with the
54:41 car minuscule emphasizing the vast
54:43 sunlet industrial space. What's wrong
54:46 with that?
54:48 Can I edit this? No.
54:52 Oops. Yes, I can.
55:04 Huh.
55:07 So, I got to run this again.
55:28 [Music]
55:36 Civot is a platform for trained lures or
55:39 stylized generations. More style control
55:42 basically instead of dude with mustache
55:44 in anime style. Use a trained Laura.
55:46 Yeah. Yeah. Yeah. For the look you want.
55:48 Yep.
55:51 Should show him civet stuff would blow
55:53 his mind. If he likes images. Yeah. I
55:56 like I haven't gotten into the whole dev
55:58 side of this thing because it's it's
56:00 tough enough keeping keep just keeping
56:02 up with the commercial
56:04 the commercial [ __ ] Um but there I got
56:07 there I got those two things to render.
56:10 So now we have now we have all the
56:11 shots.
56:14 Can you download these images?
56:17 Download current.
56:20 Yeah,
56:22 download current.
56:26 And then I have to [ __ ] name it.
56:29 [ __ ] Uh,
56:32 so
56:36 this is exhausting. Yeah, you have to
56:38 like you have to have naming conventions
56:40 prepared for all this [ __ ] Like at some
56:42 point what I should do is I should
56:44 output all of those images into a
56:46 folder, then have the folder
56:48 automatically zip it, and you download
56:50 the zip, and then you've got all your
56:51 named
56:53 named files based on the prompt. You
56:56 could you could totally do this. You
56:58 could totally do some powerful [ __ ] with
57:00 this.
57:04 But like what I want what I want to
57:06 know, like whoever asked the question
57:07 about like what do you do with this? Is
57:09 this is this just a glorified
57:10 storyboard? If it's just a glorified
57:12 storyboard, it's a little too much work.
57:14 If I can take this workflow and actually
57:16 call it via the Weave API
57:20 and I can have it just auto run all this
57:22 [ __ ] So, I can put in a single prompt
57:24 into an application and it spins its
57:27 wheels and then it starts spitting out
57:30 these outputs. That would be pretty
57:32 [ __ ] cool, don't you think?
57:41 Is the dude still in there? Yeah, he's
57:43 still in there.
57:46 Is he still in there? Yep. His head's
57:48 too big. Is he still in there? Yep.
57:53 Yeah. This is pretty crazy. So, this is
57:54 Nano Banana. Like, it kept it's The car
57:57 is the same in all the shots.
58:00 The guy's the same in all the shots.
58:04 That's pretty remarkable. I mean, not
58:07 for nothing,
58:10 think about if you wanted a shot like
58:12 that for a commercial
58:14 and then the director was like, "Well,
58:16 we need all these angles. We got to get
58:18 under the car,
58:20 right? We need we need an underexposed
58:23 one. We need his hands on the wheel. We
58:25 need one from the hubcap view. We need
58:28 it over over the driver's side. We need
58:31 the close one. We need the one over the
58:33 top. We need the one with dust after he
58:36 smashes into something.
58:39 Yeah, this is cool.
58:42 And then you could run each of these
58:44 images out to video models
58:47 if you really want to blow through your
58:48 credits.
58:52 I wonder how much credits are. Let's go
58:54 find out how much credits are.
58:56 Um, credits. Get more credits.
59:01 $19 a month for 1,500 credits. So about
59:06 3,000. So I started out with 150
59:09 credits.
59:11 That's 375
59:13 images.
59:15 Well,
59:18 no,
59:21 no, that's not right. Some of these
59:24 images are six credits. They said three
59:28 375 images.
59:30 That's not right.
59:39 Most people are paying 36 bucks a month.
59:43 45 if you go monthtomonth. I would go
59:45 monthtomonth.
59:47 Team $60 per user per month and they get
59:51 4500 credits.
59:56 Plus, they get unified billing.
1:00:03 All right, let me go look at Civet
1:00:06 because I know Civet was something that
1:00:07 was cool. I'll just go here. I don't
1:00:10 care about this thing. Civetuh
1:00:15 Civet AI, the home of open-source
1:00:18 generative AI
1:00:20 featured images.
1:00:27 Yeah. Yeah. This is this is all cool and
1:00:30 all except if you think you were
1:00:32 overwhelmed with commercial options.
1:00:39 Go to go to the world of open source and
1:00:42 let yourself be swallowed in
1:00:44 possibility.
1:00:47 All right. Um I gotta stop trying to pay
1:00:50 attention and learn because my head
1:00:52 hurts.
1:00:54 Ah um Tik Tok question Tyler
1:01:00 all platform for image generation or
1:01:02 overpriced UI for the use of an API.
1:01:06 Yeah, everything can be scheduled.
1:01:12 I think we can get super intelligence by
1:01:14 2030 if the models make better versions
1:01:16 of themselves. Yeah,
1:01:21 he's agreed. I'll eventually try to
1:01:22 reproduce my dreams with AI.
1:01:26 Oh, I did that one night. One night.
1:01:32 I had a dream.
1:01:35 This was when I was first playing with
1:01:37 AI. I had a dream and I woke up
1:01:41 and I described my dream to chat GPT
1:01:46 and then I said, "Write an image prompt
1:01:47 for that." And then I went into one of
1:01:49 the image generation tools and generated
1:01:51 it. It was pretty cool.
1:01:54 You could totally build that.
1:02:05 And a year ago, these tools were
1:02:07 struggling
1:02:08 to spell car correctly. Yeah. It's
1:02:10 crazy.
1:02:14 They doing that in real world brain
1:02:16 patterns.
1:02:18 Oh, of dreams to AI. Oh, that's cool.
1:02:22 That's cool.
1:02:24 All right. Holy [ __ ] my eyes are
1:02:26 crispy. I gotta go. My My eyes are done.
1:02:28 It's time. That's funny. Brandon just
1:02:31 said the same thing. He knows me all too
1:02:34 well. Um, producer Brandon, listen, I
1:02:37 would stay with you guys for hours, but
1:02:39 producer Brandon says I have to go to
1:02:40 bed.
1:02:42 Um, tomorrow is Wednesday. Uh there's
1:02:46 the AI readiness project podcast is uh
1:02:49 on at 400 pm mountain time. So if you go
1:02:53 to aire readiness project.com
1:02:57 that will take you there and uh that's
1:03:01 on tomorrow at 400 pm. Also go to the
1:03:05 salon oh are you readyforai.com
1:03:08 go check that out. That's the AI
1:03:09 readiness training program
1:03:12 uh where you can learn to get your mind
1:03:15 right with AI.
1:03:17 And then finally, go to the salon.ai. We
1:03:20 have a spanking new website that's
1:03:23 beautiful. Looks great. Um and it's
1:03:25 actually got actual information now like
1:03:28 when what are the next events that are
1:03:30 coming up and you know what's some of
1:03:32 the activity going on in the salon. It's
1:03:33 really kind of cool. Um Pave M's in the
1:03:36 house. What's happening sir?
1:03:38 Yes, you need to go to bed. I really do.
1:03:40 Okay, I'm getting out of here. I'll see
1:03:42 you tomorrow night at 8 o'clock. I don't
1:03:44 think there's anything going on, so I
1:03:46 will see you then. And uh I hope you
1:03:49 enjoyed the Weevie
1:03:52 demo tonight. That was kind of fun.
1:03:55 Later.