00:00Character consistency is no longer a challenge with Nano Banana 2 and Pro.
00:03You simply use a reference image of your character, describe the image you want to get,
00:07and you can be 95% sure that the person on the generated image will look consistent.
00:12However, many creators still struggle with the fine details,
00:15capturing the perfect pose, facial expressions, or lighting.
00:18Sometimes images look like amateur photos, or they can be too sterile.
00:22Often the issue is not the AI, it's the prompt.
00:24But there's a way to fix this.
00:26Even if you aren't a professional photographer,
00:27familiar with technical terminology, you can still get great results.
00:31You can find an image that inspires you.
00:33Many people use Pinterest, for example.
00:35Once you find one you like, you can ask Gemini or ChatGPT
00:39to create an extremely detailed description of the uploaded image in JSON format.
00:44Don't let the term JSON, JavaScript object notation, intimidate you.
00:48No coding skills are required.
00:50In this case, JSON is used purely to keep the prompt clearly structured.
00:53You can copy the generated prompt and use it anywhere.
00:55I'm going to create images in Google's Flow this time.
00:58So now you can upload a reference image of your character and paste the prompt.
01:02In my case, the prompt is really long and very detailed.
01:06You can use either the Nano Banana 2 or the Pro version.
01:09Nano Banana 2 allows you to create a few hundred images per day,
01:12while Nano Banana Pro lets you generate dozens of images.
01:15The result is flawless.
01:16As you can see, the person remains remarkably consistent,
01:20while the lighting, composition, and overall vibe perfectly mirror the image I used for the prompt generation.
01:26Looking closer, you'll see each image aspect is defined separately.
01:30JSON format makes it easy to edit individual parts without rewriting the whole prompt each time.
01:35I changed night to evening.
01:37Pose and all the other elements will remain the same.
01:39I'm going to create two images this time.
01:41Here's how this scene looks in the evening.
01:43Here is the overall workflow.
01:44You can generate a full description of some image you like,
01:47or focus on specific elements like lighting and colors.
01:50And then you can use it together with the reference image of your character.
01:54And here's a little bonus.
01:55There's a pretty simple trick to give your images a very casual, cinematic, or professional look.
02:00First, I'm going to place my character in a room, wearing a delicate pink sundress.
02:05The output is fairly standard and predictable.
02:07The images aren't bad, but they don't go beyond the typical Nano Banana default style.
02:11I won't change anything, but I'll add another attribute that can be named in different ways.
02:15I'll call it general style and purpose of the image.
02:18For example, let's go with a very casual, low-quality photo.
02:22As you can see, everything from the lighting and composition to the pose looks completely different now,
02:27matching my description perfectly.
02:28I can even push this effect further by just specifying that the shot is very, very casual,
02:33and the quality is very, very low.
02:35The new couple of images look even more amateur.
02:38Now, let's try the opposite.
02:40This time, this should be dramatic, professional shot for the fashion magazine.
02:44Just look at this.
02:45I didn't mention anything like camera lens types, lighting, composition, or character's pose.
02:50This can be shot for the advertising of this delicate pink sundress.
02:54The fact that it's a casual shot or a product advertising automatically implies certain characteristics
02:59typical for that specific context.
03:01Just a few words at the beginning of the prompt can change the output significantly.
03:05But, of course, if you need something extremely accurate and specific,
03:09the first method with AI-generated prompt will help you to create something much more predictable.
03:14That's all for today.
03:15Thanks for watching.
Comments