WEBVTT

00:00:00.000 --> 00:00:01.639
You know that feeling, you've got this image,

00:00:01.760 --> 00:00:03.660
it's almost perfect, you know exactly what needs

00:00:03.660 --> 00:00:06.139
to change, maybe just the color of a shirt or

00:00:06.139 --> 00:00:08.400
getting rid of a sign way in the back. You type

00:00:08.400 --> 00:00:11.220
it out, super clear. Oh yeah. And then the AI,

00:00:11.519 --> 00:00:16.120
this incredibly powerful thing, starts editing

00:00:16.120 --> 00:00:19.219
the sidewalk or the sky or like the wrong shirt.

00:00:19.320 --> 00:00:21.199
It drives you crazy. Exactly. You're trying to

00:00:21.199 --> 00:00:24.140
give precise visual directions using just text.

00:00:24.280 --> 00:00:26.899
And it feels like the AI is just... guessing

00:00:26.899 --> 00:00:28.760
where to look half the time. It's like trying

00:00:28.760 --> 00:00:31.160
to guide a brilliant artist who's wearing a blindfold.

00:00:31.500 --> 00:00:33.119
That's really the core problem we're jumping

00:00:33.119 --> 00:00:35.820
into today. Welcome to the deep dive. We're focusing

00:00:35.820 --> 00:00:39.719
on a simple no -code way to basically take the

00:00:39.719 --> 00:00:41.439
blindfold off. Yeah, we're talking about combining

00:00:41.439 --> 00:00:44.039
Canva, which is super visual, with the power

00:00:44.039 --> 00:00:45.640
of nano -banana. That's what we call the image

00:00:45.640 --> 00:00:48.320
engine in Gemini 2 .5 Flash Image. It's fast.

00:00:48.840 --> 00:00:50.820
And our mission today is pretty straightforward.

00:00:51.179 --> 00:00:54.020
Turn that frustrating trial and error AI editing

00:00:54.020 --> 00:00:57.479
into something predictable, powerful, and honestly

00:00:57.479 --> 00:01:00.340
kind of a one -shot deal. So first up, we'll

00:01:00.340 --> 00:01:02.780
look at what NanoBanana does really well and,

00:01:02.820 --> 00:01:04.599
you know, where it kind of falls down that spatial

00:01:04.599 --> 00:01:06.819
guesswork. Then we'll reveal the simple fix using

00:01:06.819 --> 00:01:10.180
Canva's visual tools. After that, the step -by

00:01:10.180 --> 00:01:12.980
-step playbook. And finally, we peek under the

00:01:12.980 --> 00:01:14.920
hood a bit why this visual approach actually

00:01:14.920 --> 00:01:17.040
works so well on a technical level. Sound good?

00:01:17.359 --> 00:01:19.920
Sounds great. Okay, so let's start with the good

00:01:19.920 --> 00:01:22.859
stuff. What makes NanoBanana so promising, even

00:01:22.859 --> 00:01:25.439
with the spatial issue we mentioned? Well, the

00:01:25.439 --> 00:01:27.299
speed and accessibility are big ones, right?

00:01:27.400 --> 00:01:30.560
It's integrated, often free to use. But the tech

00:01:30.560 --> 00:01:34.239
itself, two things really jump out. First, its

00:01:34.239 --> 00:01:37.379
natural language understanding is excellent.

00:01:37.620 --> 00:01:40.000
Meaning you don't need weird code words. Exactly.

00:01:40.079 --> 00:01:41.939
You can talk to it pretty normally about what

00:01:41.939 --> 00:01:43.599
you want creatively, make this look more dramatic,

00:01:43.859 --> 00:01:45.829
stuff like that. Okay. And the second thing,

00:01:45.890 --> 00:01:47.709
you mentioned character consistency. I think

00:01:47.709 --> 00:01:49.750
this might be the real game changer. Totally

00:01:49.750 --> 00:01:53.230
agree. It's amazing at keeping a person looking

00:01:53.230 --> 00:01:55.530
like the same person across different edits.

00:01:55.750 --> 00:01:57.769
Change their clothes, change the background,

00:01:57.969 --> 00:02:00.909
change the pose. Wait, so the face stays consistent

00:02:00.909 --> 00:02:04.069
even if I change the hair or the lighting? That

00:02:04.069 --> 00:02:06.709
fixes so many headaches we've had with AI images.

00:02:06.790 --> 00:02:08.689
It really does help with that continuity. Yeah.

00:02:08.810 --> 00:02:12.490
But, and here's the catch. Even with all that

00:02:12.490 --> 00:02:14.810
smarts, if you've got three people in a photo

00:02:14.810 --> 00:02:17.889
and say, change the shirt, it still doesn't inherently

00:02:17.889 --> 00:02:20.069
know which shirt. Right. It's back to the guessing

00:02:20.069 --> 00:02:22.610
game. That's the spatial ambiguity. It has to

00:02:22.610 --> 00:02:24.909
statistically figure out, you know, which pixels

00:02:24.909 --> 00:02:28.090
are shirt versus arm versus background that looks

00:02:28.090 --> 00:02:29.569
kind of like a shirt. And that's where we all

00:02:29.569 --> 00:02:31.750
fall into that trap of writing these incredibly

00:02:31.750 --> 00:02:35.389
long, specific prompts. Oh, yeah. The legal document

00:02:35.389 --> 00:02:38.310
prompts. Change only the blue cotton T -shirt.

00:02:38.669 --> 00:02:41.310
Short sleeved, worn by the person standing second

00:02:41.310 --> 00:02:43.310
from the left, slightly behind the oak tree,

00:02:43.409 --> 00:02:46.210
ignoring the logo. Size slightly. And I still

00:02:46.210 --> 00:02:47.949
wrestle with that myself. Try and get spatial

00:02:47.949 --> 00:02:51.250
precision just with words. It often just drifts

00:02:51.250 --> 00:02:54.370
or fails completely. I wasted like half an hour

00:02:54.370 --> 00:02:56.409
last week trying to tell it just the blue stripe

00:02:56.409 --> 00:02:58.750
on the bag, not the blue background. Never got

00:02:58.750 --> 00:03:01.610
it. It's just a mismatch. Fundamentally, we think

00:03:01.610 --> 00:03:04.310
visually, we point, we say this thing right here.

00:03:04.710 --> 00:03:08.740
The AI thinks in text. in probabilities based

00:03:08.740 --> 00:03:11.419
on that text. So describing where with words

00:03:11.419 --> 00:03:15.180
is just the wrong language. Pretty much. It's

00:03:15.180 --> 00:03:18.340
inefficient, often ineffective. Okay, so how

00:03:18.340 --> 00:03:20.580
do humans typically try to solve this spatial

00:03:20.580 --> 00:03:23.280
guessing game right now before this visual trick?

00:03:23.639 --> 00:03:25.659
By writing those incredibly specific complex

00:03:25.659 --> 00:03:28.759
prompts, which, as we said, usually fail to guarantee

00:03:28.759 --> 00:03:31.819
precision anyway. Right. So if text fails, how

00:03:31.819 --> 00:03:34.639
do we actually talk to the AI in a way it understands

00:03:34.639 --> 00:03:37.099
spatially? We stop describing the location and

00:03:37.099 --> 00:03:39.460
start showing it the location, visually. Ah,

00:03:39.719 --> 00:03:42.280
moving from description to demonstration. Okay,

00:03:42.360 --> 00:03:44.539
that leads us to the breakthrough. Using visual

00:03:44.539 --> 00:03:47.259
markers. It sounds simple. It really is. We use

00:03:47.259 --> 00:03:49.960
NanoBanana's other big strength, multimodal editing.

00:03:50.120 --> 00:03:53.039
It can understand both text and images. So we

00:03:53.039 --> 00:03:54.800
give it an image with visual instructions drawn

00:03:54.800 --> 00:03:57.180
right on it. And Canva is perfect for that because

00:03:57.180 --> 00:03:59.120
it's easy, accessible. You don't need to be a

00:03:59.120 --> 00:04:01.900
graphic designer. Exactly. No complex software

00:04:01.900 --> 00:04:05.490
needed. You literally just draw. say, a bright

00:04:05.490 --> 00:04:07.710
pink rectangle around the exact area you want

00:04:07.710 --> 00:04:09.969
to change, a really high contrast marker. Like

00:04:09.969 --> 00:04:12.229
a big visual flag. Precisely. And then you put

00:04:12.229 --> 00:04:14.229
the text instruction, change to blue, remove

00:04:14.229 --> 00:04:17.500
this. right next to or sometimes inside that

00:04:17.500 --> 00:04:19.879
box. So going back to the analogy, we're not

00:04:19.879 --> 00:04:22.879
asking it to guess which shirt concept it should

00:04:22.879 --> 00:04:26.860
use from its massive internal library, its latent

00:04:26.860 --> 00:04:29.439
space. Right. You're drawing a giant pink box

00:04:29.439 --> 00:04:32.040
around the actual pixels on the image and saying,

00:04:32.060 --> 00:04:35.540
loud and clear, this shirt at this specific thing.

00:04:35.620 --> 00:04:38.240
It bypasses the guessing. Completely. Yeah. Which

00:04:38.240 --> 00:04:41.019
leads to a key question someone might have. Does

00:04:41.019 --> 00:04:43.319
this mean we need complicated drawing skills

00:04:43.319 --> 00:04:45.420
or expensive software to make these markers?

00:04:45.600 --> 00:04:47.939
Not at all. It relies entirely on Canva's simplest

00:04:47.939 --> 00:04:50.379
tools, drawing basic shapes and adding plain

00:04:50.379 --> 00:04:52.699
text. Super easy. Okay, let's get practical.

00:04:52.839 --> 00:04:54.620
Walk us through the playbook. How does this actually

00:04:54.620 --> 00:04:56.220
work step by step? It sounds like it could be

00:04:56.220 --> 00:04:58.139
really fast once you know the shortcuts. Oh,

00:04:58.160 --> 00:05:00.079
definitely. Once you get it down, you can mark

00:05:00.079 --> 00:05:02.680
up even a complex image in like under a minute.

00:05:03.019 --> 00:05:05.839
So steps one to four are all in Canva. Upload

00:05:05.839 --> 00:05:08.899
your image. Easy. Got it. Then hit the R key.

00:05:09.209 --> 00:05:11.730
Shortcut for rectangle. Draw your box around

00:05:11.730 --> 00:05:14.589
the specific bit you want to edit. Okay. R for

00:05:14.589 --> 00:05:17.870
rectangle. Simple enough. What's next? You mentioned

00:05:17.870 --> 00:05:20.730
formatting it as an AI signal. Right. This part's

00:05:20.730 --> 00:05:23.350
key. You need to make the box speak the AI's

00:05:23.350 --> 00:05:26.149
language. So first, change the fill color to

00:05:26.149 --> 00:05:29.069
transparent. No fill. Why transparent? So the

00:05:29.069 --> 00:05:31.149
AI can still see the image underneath the box

00:05:31.149 --> 00:05:34.029
clearly. The box is just a boundary marker. Then

00:05:34.029 --> 00:05:36.629
set the border color. Use something really bright,

00:05:36.709 --> 00:05:39.949
high contrast. We recommend obnoxious pink. Chuckle

00:05:39.949 --> 00:05:43.350
slightly. Obnoxious pink? Why pink? It just stands

00:05:43.350 --> 00:05:46.350
out. It's rarely the main color in a photo, so

00:05:46.350 --> 00:05:48.649
the AI sees it as an instruction, not part of

00:05:48.649 --> 00:05:51.449
the scene. Make the border, say, three to five

00:05:51.449 --> 00:05:54.230
pixels wide so it's really obvious. Okay. Transparent

00:05:54.230 --> 00:05:56.389
fill, bright pink border, couple pixels wide.

00:05:56.490 --> 00:05:58.550
Got it. Then the instruction. Hit the T key.

00:05:58.790 --> 00:06:01.230
Shortcut for text tool. Type your clear, simple

00:06:01.230 --> 00:06:03.439
instruction right next to the pink box. Remove

00:06:03.439 --> 00:06:05.920
this car. Make sure dark green. Keep it concise.

00:06:06.240 --> 00:06:09.000
R for rectangle. T for text. Pink box. Clear

00:06:09.000 --> 00:06:12.160
instruction. Done. Now what? Now the export part.

00:06:12.399 --> 00:06:14.680
Select everything. The original image. All the

00:06:14.680 --> 00:06:17.720
pink boxes you drew. All the text labels. Everything

00:06:17.720 --> 00:06:19.920
together. Clubbed all. Okay. And this is important.

00:06:20.079 --> 00:06:23.379
Use Canva's download selection option. Not download

00:06:23.379 --> 00:06:26.519
page or download all. Just the selection. The

00:06:26.519 --> 00:06:28.800
download selection. Why is that specific? It

00:06:28.800 --> 00:06:30.839
ensures you only get the image with the markup

00:06:30.839 --> 00:06:33.449
perfectly aligned without... Any extra white

00:06:33.449 --> 00:06:35.730
space from the Canva Canvas. Save it as a PNG.

00:06:35.930 --> 00:06:39.610
High quality. Got it. Marked up PNG, downloaded

00:06:39.610 --> 00:06:42.589
via download selection. Then we head over to

00:06:42.589 --> 00:06:45.490
NanoBanana. Exactly. Open Gemini or wherever

00:06:45.490 --> 00:06:47.949
you access NanoBanana. Upload that marked up

00:06:47.949 --> 00:06:50.170
PNG file you just saved. Okay. Image uploaded.

00:06:50.250 --> 00:06:53.550
Now the prompt. Is it complicated? Nope. This

00:06:53.550 --> 00:06:56.029
is the beauty of it. You use one simple universal

00:06:56.029 --> 00:06:57.990
prompt for almost everything. Universal prompt.

00:06:58.129 --> 00:07:01.509
What is it? It's simply read the pink text in

00:07:01.509 --> 00:07:04.420
the image. And make the modifications. Remove

00:07:04.420 --> 00:07:08.439
the pink text and boxes. That's it. Huh. Okay,

00:07:08.540 --> 00:07:11.199
let me unpack that. What is the exact purpose

00:07:11.199 --> 00:07:13.680
of that single -sentence universal prompt at

00:07:13.680 --> 00:07:17.439
the end? It does two crucial things. Tells the

00:07:17.439 --> 00:07:19.680
AI what to look for the visual instructions marked

00:07:19.680 --> 00:07:21.920
in pink, and then tells it to clean up after

00:07:21.920 --> 00:07:24.420
itself, removing the guides for the final image.

00:07:24.839 --> 00:07:26.879
Wow. So it reads the instructions on the edges,

00:07:26.899 --> 00:07:28.860
does the edits, and erases the instructions.

00:07:28.920 --> 00:07:30.779
So that's incredibly efficient. It really is.

00:07:30.879 --> 00:07:33.560
One prompt, precise edits. And you're saying

00:07:33.560 --> 00:07:36.040
this isn't just for fixing one small thing. It

00:07:36.040 --> 00:07:38.680
can handle more complex stuff. Absolutely. This

00:07:38.680 --> 00:07:40.639
scales really well. You can use some more advanced

00:07:40.639 --> 00:07:43.329
techniques. For starters... Multiple simultaneous

00:07:43.329 --> 00:07:46.110
edits. Meaning? Just draw more boxes, put a pink

00:07:46.110 --> 00:07:48.709
box around the sky that says, make vibrant blue.

00:07:49.009 --> 00:07:51.170
Another around a person saying, remove this person.

00:07:51.269 --> 00:07:53.970
Another on a building saying, add Ivy. One image

00:07:53.970 --> 00:07:57.089
upload, one universal prompt. And NanoBanana

00:07:57.089 --> 00:07:59.209
understands each separate instruction applies

00:07:59.209 --> 00:08:02.509
only to its specific pink box region. Exactly.

00:08:02.670 --> 00:08:04.850
The spatial guidance is locked in for each one.

00:08:04.949 --> 00:08:07.579
It executes them all in one go. Okay, that's

00:08:07.579 --> 00:08:10.180
powerful. What about really complicated edits

00:08:10.180 --> 00:08:13.339
like major architectural changes or something?

00:08:13.579 --> 00:08:15.879
For that, you can use layer refinement. Think

00:08:15.879 --> 00:08:20.699
of it like working in stages. How so? So, generation

00:08:20.699 --> 00:08:24.779
one. You upload the original image, mark it up

00:08:24.779 --> 00:08:26.899
for the big structural changes, maybe removing

00:08:26.899 --> 00:08:29.339
some ugly scaffolding from a building. You run

00:08:29.339 --> 00:08:32.220
it, get the result. Okay, scaffold's gone. Then

00:08:32.220 --> 00:08:34.600
you take that resulting image. Upload it again

00:08:34.600 --> 00:08:38.519
and add new pink boxes for Generation 2. This

00:08:38.519 --> 00:08:40.720
time maybe focusing on details like fixing a

00:08:40.720 --> 00:08:42.879
crack in the wall or changing a reflection in

00:08:42.879 --> 00:08:45.399
a window. Ah, so you break down complex tasks

00:08:45.399 --> 00:08:48.519
into smaller management chunks for the AI? Precisely.

00:08:48.639 --> 00:08:50.899
It prevents overwhelming the model and gives

00:08:50.899 --> 00:08:53.019
you more control over each stage. You could even

00:08:53.019 --> 00:08:55.860
do a Generation 3 for final polish. Makes sense.

00:08:55.899 --> 00:08:58.240
Can you combine this with reference images? If

00:08:58.240 --> 00:08:59.799
I want the sky to look like a specific photo

00:08:59.799 --> 00:09:02.190
I have. Yep. That's another great technique.

00:09:02.309 --> 00:09:04.730
Draw your pink box around the sky in your main

00:09:04.730 --> 00:09:07.350
image. In the text next to it, write something

00:09:07.350 --> 00:09:10.309
like, match the style and colors of the reference

00:09:10.309 --> 00:09:12.649
image for this sky area. And then you upload

00:09:12.649 --> 00:09:14.850
both images, the marked up one and the reference

00:09:14.850 --> 00:09:18.470
sky photo. Exactly. Upload both. The AI uses

00:09:18.470 --> 00:09:20.730
the pink box to know where to apply the style

00:09:20.730 --> 00:09:23.210
and the reference image to know what style to

00:09:23.210 --> 00:09:26.259
apply. Spatial accuracy plus aesthetic matching.

00:09:26.440 --> 00:09:28.460
That's really versatile. What about for people

00:09:28.460 --> 00:09:31.159
doing lots of similar edits, like product photos

00:09:31.159 --> 00:09:33.960
for e -commerce? Template reuse is your friend

00:09:33.960 --> 00:09:36.899
there. In Canva... Create a template with your

00:09:36.899 --> 00:09:39.279
standard image size and maybe some pre -placed,

00:09:39.299 --> 00:09:42.240
pre -formatted pink boxes for common edits like,

00:09:42.340 --> 00:09:44.620
say, always cleaning up the background. So you

00:09:44.620 --> 00:09:46.279
just drop in the new product photo, maybe adjust

00:09:46.279 --> 00:09:48.240
the box slightly, type the instruction, and boom.

00:09:48.379 --> 00:09:50.340
Pretty much. Super fast for high volumes. You

00:09:50.340 --> 00:09:52.480
can even color code your boxes if you get really

00:09:52.480 --> 00:09:55.340
fancy. I tower code. Yeah. Maybe pink means modify,

00:09:55.620 --> 00:09:57.820
red means remove, blue means change lighting.

00:09:58.000 --> 00:09:59.940
You just add a little text legend somewhere on

00:09:59.940 --> 00:10:02.000
the template like AI, pink modify, red remove.

00:10:02.340 --> 00:10:05.120
Whoa. Okay. Imagine scaling that. Templates?

00:10:05.549 --> 00:10:07.830
Color coding, you could process hundreds, thousands

00:10:07.830 --> 00:10:10.629
of images with that level of precision driven

00:10:10.629 --> 00:10:14.529
by simple visual cues. That's serious leverage.

00:10:14.809 --> 00:10:16.889
It really opens things up. So let's say I need

00:10:16.889 --> 00:10:19.190
to maintain really consistent product branding

00:10:19.190 --> 00:10:21.370
across like all my seasonal marketing images.

00:10:22.230 --> 00:10:25.149
Which advanced technique should I lean on most?

00:10:25.850 --> 00:10:27.950
Template reuse combined with that color coding

00:10:27.950 --> 00:10:31.289
idea is probably best for repeatable, consistent,

00:10:31.429 --> 00:10:34.049
high volume edits where you want minimal variation.

00:10:34.470 --> 00:10:37.100
Got it. Okay, let's dive a bit deeper. Why does

00:10:37.100 --> 00:10:39.919
this work so well? Why is a simple pink box so

00:10:39.919 --> 00:10:41.720
much better than that thousand word prompt we

00:10:41.720 --> 00:10:44.509
talked about? It gets down to how these AI models

00:10:44.509 --> 00:10:48.570
actually see or, well, process images. You're

00:10:48.570 --> 00:10:50.110
essentially guiding the attention mechanisms

00:10:50.110 --> 00:10:52.870
directly. Attention mechanisms, like where the

00:10:52.870 --> 00:10:55.629
AI focuses its processing power. Exactly. That

00:10:55.629 --> 00:10:57.789
bright pink box is like a giant flashing neon

00:10:57.789 --> 00:11:01.169
sign yelling, hey, AI, pay attention to these

00:11:01.169 --> 00:11:03.210
specific pixels right here. You're telling it

00:11:03.210 --> 00:11:05.490
exactly where to concentrate. So it's not just

00:11:05.490 --> 00:11:07.669
analyzing the whole image vaguely based on the

00:11:07.669 --> 00:11:10.580
text anymore? Right. You're solving what's sometimes

00:11:10.580 --> 00:11:12.580
called the latent space problem more efficiently.

00:11:12.860 --> 00:11:15.419
Think of the AI's mind, its latent space, as

00:11:15.419 --> 00:11:19.399
this huge abstract library of every visual concept

00:11:19.399 --> 00:11:22.340
it knows. Typing shirt makes it wander through

00:11:22.340 --> 00:11:24.799
the entire shirt section of the library, trying

00:11:24.799 --> 00:11:27.259
to guess which one you mean. The pink box is

00:11:27.259 --> 00:11:29.480
like giving it the exact page number and paragraph.

00:11:29.860 --> 00:11:31.799
You're massively narrowing down the search space.

00:11:32.000 --> 00:11:35.179
Hugely. From potentially millions of possibilities

00:11:35.179 --> 00:11:38.570
down to just the pixels inside that box. This

00:11:38.570 --> 00:11:41.389
computational localization saves processing,

00:11:41.710 --> 00:11:45.190
reduces errors. It takes the success rate from

00:11:45.190 --> 00:11:49.289
maybe 50 -50. Or worse, up to like 99 .9%. And

00:11:49.289 --> 00:11:51.529
it perfectly uses the AI's ability to handle

00:11:51.529 --> 00:11:54.710
multiple types of input. Precisely. It's multimodal

00:11:54.710 --> 00:11:57.389
information fusion at its best. Visual data,

00:11:57.509 --> 00:11:59.490
the pink box telling it where, plus text data,

00:11:59.549 --> 00:12:01.730
the label telling it what to do. They combine

00:12:01.730 --> 00:12:04.350
for precise action. You're finally speaking its

00:12:04.350 --> 00:12:06.529
most effective language. How does this compare

00:12:06.529 --> 00:12:09.110
to, say, traditional methods like using masking

00:12:09.110 --> 00:12:11.870
tools in Photoshop? Well, Photoshop masks or

00:12:11.870 --> 00:12:14.110
in -painting masks are pixel perfect. You can

00:12:14.110 --> 00:12:17.419
get absolute precision. learning curve, expensive

00:12:17.419 --> 00:12:20.700
software, time consuming. All of the above. Mastering

00:12:20.700 --> 00:12:23.740
manual masking takes time and skill. This Canva

00:12:23.740 --> 00:12:27.279
workflow, it gives you maybe 95 % of that pixel

00:12:27.279 --> 00:12:30.340
level precision, but for, I don't know, 10 %

00:12:30.340 --> 00:12:32.919
of the effort and cost. So for the average professional

00:12:32.919 --> 00:12:35.700
using AI for content creation, not necessarily

00:12:35.700 --> 00:12:38.399
high -end retouching, what's the key advantage

00:12:38.399 --> 00:12:41.399
here over shelling out for expensive pro software

00:12:41.399 --> 00:12:44.539
and training? It offers really high spatial precision.

00:12:44.970 --> 00:12:47.110
really quickly without needing the budget or

00:12:47.110 --> 00:12:49.250
the time investment for deep manual masking skills.

00:12:49.529 --> 00:12:52.269
It's democratizing precise editing. The applications

00:12:52.269 --> 00:12:53.870
seem pretty widespread then. Oh, absolutely.

00:12:54.110 --> 00:12:56.289
E -commerce is huge, like we said. Changing product

00:12:56.289 --> 00:12:58.509
colors, standardizing backgrounds to pure white

00:12:58.509 --> 00:13:01.269
for catalogs. Perfect use case. I can see it

00:13:01.269 --> 00:13:04.389
for real estate too. Turning a drab gray sky

00:13:04.389 --> 00:13:07.009
blue and listing photos. Big impact. Definitely.

00:13:07.029 --> 00:13:09.529
Or virtual staging. Adding furniture realistically.

00:13:09.909 --> 00:13:11.690
Removing distracting stuff like, you know, a

00:13:11.690 --> 00:13:13.429
trash can on the curb or a car in the driveway.

00:13:13.629 --> 00:13:15.710
All pinpoint accurate. Social media managers

00:13:15.710 --> 00:13:18.129
must love this. Creating variations for A -B

00:13:18.129 --> 00:13:21.679
testing ads or posts. Super fast. Generate five

00:13:21.679 --> 00:13:23.720
versions of an image with slightly different

00:13:23.720 --> 00:13:26.320
elements in minutes, all controlled. Even just

00:13:26.320 --> 00:13:28.500
for personal photos, right? Finally removing

00:13:28.500 --> 00:13:31.200
that random person who photobombed your perfect

00:13:31.200 --> 00:13:34.259
vacation shot. Yeah. Or making a sunset just

00:13:34.259 --> 00:13:36.500
a little more dramatic. It makes those kinds

00:13:36.500 --> 00:13:38.820
of edits reliable, not a frustrating gamble.

00:13:39.059 --> 00:13:41.419
But we should be clear, it's not magic. What

00:13:41.419 --> 00:13:43.559
are the limitations? Right. It's important to

00:13:43.559 --> 00:13:45.960
set expectations. This is not a full Photoshop

00:13:45.960 --> 00:13:48.820
replacement for, say, high -resolution billboard

00:13:48.820 --> 00:13:51.620
ads or complex magazine cover retouching where

00:13:51.620 --> 00:13:54.379
every single pixel needs manual finessing. It's

00:13:54.379 --> 00:13:57.139
for that 95 % zone, not the absolute highest

00:13:57.139 --> 00:13:59.919
end. Exactly. And crucially, the results still

00:13:59.919 --> 00:14:03.019
depend entirely on Nano Banana's underlying abilities.

00:14:03.360 --> 00:14:06.980
If the base AI model is just... Bad at generating

00:14:06.980 --> 00:14:08.960
realistic hands, for example. This technique

00:14:08.960 --> 00:14:11.299
won't magically fix that. Nope. It will help

00:14:11.299 --> 00:14:13.960
you tell the AI exactly where to try and generate

00:14:13.960 --> 00:14:16.899
those potentially wonky hands, but it can improve

00:14:16.899 --> 00:14:19.700
the AI's fundamental drawing skills, so to speak.

00:14:20.059 --> 00:14:22.899
That makes sense. So just to confirm, since this

00:14:22.899 --> 00:14:25.919
relies on the underlying AI model, will this

00:14:25.919 --> 00:14:29.320
technique fix a universally known AI problem,

00:14:29.480 --> 00:14:31.679
like rendering realistic hands consistently?

00:14:32.100 --> 00:14:34.960
No, unfortunately not. The technique gives you

00:14:34.960 --> 00:14:37.980
pinpoint spatial control, but it can't overcome

00:14:37.980 --> 00:14:40.600
the core creative or representational limitations

00:14:40.600 --> 00:14:43.500
of the AI model itself. Better hands require

00:14:43.500 --> 00:14:46.620
a better base model. Got it. So stepping back,

00:14:46.740 --> 00:14:49.200
the big picture here. We were struggling, trying

00:14:49.200 --> 00:14:51.940
to bend our visual way of thinking into the AI's

00:14:51.940 --> 00:14:54.899
text -only input. Lots of friction, lots of failure.

00:14:55.179 --> 00:14:57.059
Yeah, it was like trying to hammer a screw. We

00:14:57.059 --> 00:14:59.379
were using the wrong tool. The solution was actually

00:14:59.379 --> 00:15:02.039
simple. Use visual instructions for visual tasks.

00:15:02.340 --> 00:15:05.080
Speak the AI's multimodal language. And doing

00:15:05.080 --> 00:15:07.639
that transforms Nanobanana from something powerful

00:15:07.639 --> 00:15:10.320
but kind of erratic into the precise, reliable,

00:15:10.539 --> 00:15:12.840
creative partner we were hoping for. It makes

00:15:12.840 --> 00:15:15.039
the AI adapt to us. It really does feel like

00:15:15.039 --> 00:15:17.639
unlocking its potential. Which leads to a fascinating

00:15:17.639 --> 00:15:21.000
final thought. If combining two really simple,

00:15:21.080 --> 00:15:24.019
accessible, no -code tools like Canva and Anobanana

00:15:24.019 --> 00:15:26.659
creates this level of precision and control,

00:15:27.080 --> 00:15:29.740
what does that imply about the future of how

00:15:29.740 --> 00:15:32.779
we interact with all AI? Right. Will the most

00:15:32.779 --> 00:15:36.139
powerful, most complex AI systems actually be

00:15:36.139 --> 00:15:38.659
hidden behind the simplest, most intuitive visual

00:15:38.659 --> 00:15:41.279
interfaces? Maybe the command line gives way

00:15:41.279 --> 00:15:44.559
to the pink rectangle. Something to ponder. But

00:15:44.559 --> 00:15:46.779
for now, the takeaway for you listening is try

00:15:46.779 --> 00:15:49.279
this. Seriously. Open Canva. Upload an image.

00:15:49.399 --> 00:15:52.220
Hit R. Draw a pink box. Hit T. Type an instruction.

00:15:52.500 --> 00:15:54.799
Download the selection. Upload it to Gemini or

00:15:54.799 --> 00:15:57.659
Nano Banana. Use that universal prompt. Read

00:15:57.659 --> 00:15:59.700
the pink text in the image and make the modifications.

00:15:59.919 --> 00:16:02.360
Remove the pink text in boxes. You'll likely

00:16:02.360 --> 00:16:04.679
be amazed at how accurately it follows your visual

00:16:04.679 --> 00:16:07.179
lead. Get out there and start editing with precision.