Avid AI Image Creator - Imagen 4 - The Challenge

byMrDon1

1.2K followers

3 months ago

Another weird 3 am, and I am still walking this road that exists for me.

I signed into Whisk Google Labs to create an image for another project I have started.

Are you familiar with Imagen 4?

Do you create your own images?

IF you are not, here are three interesting facts about Google's Labs FX, Whisk, and related AI tools that you might not know:

1. Whisk uses descriptive text generated from image inputs. The "image-to-image" workflow in Whisk uses a hidden process. When photos for the subject, scene, and style are uploaded, Google's Gemini model analyzes them and creates text captions. These captions are then used by the Imagen 4 model to generate the final image. A great prompt is just as good as an image.
   
2. All images created with Labs FX tools have an invisible watermark. Images generated with tools like ImageFX have a digital watermark from SynthID, a tool by Google DeepMind. This watermark can be detected by computers, allowing verification of the AI-generated content.
   
3. ImageFX’s "expressive chips" are for more than just fun variations. Clicking the "chips" (small, clickable text tags) in ImageFX helps users write better, more detailed AI prompts. The tool suggests specific words for moods, styles, and dimensions, helping users see how subtle changes in a text prompt can create different visual results. This process enhances the user's creative vocabulary and prompt-engineering skills.

Nice to Know

Imagen 4 and FreePik are two platforms that create lifelike images that are hard to beat.

Well, Google is currently going bananas (Flash 2.5), and is looking for some free advertising for the rest of its toys.

I am doing my part to fill in your image creation blanks. Whisk lets you set the Subject (Person, Cat, Dog, etc), the Scene, where the person is, as well as the Style. You can create some great art just by swapping in one of the three placeholders.

I changed the style box from the original and am passing the challenge on to you. You can pass it on to as many people as you would like. This is their introduction to Whisk.

The Challenge link is here.

https://labs.google/fx/tools/whisk/share/3srlssvh00000

If you prefer, go to https://labs.google/fx/ for the Google Creatives that people ARE NOT talking about. You will be met with the same challenge.

For those interested in my new best friend, FreePik, the link is here

https://www.freepik.com/

You know me well. What all have you been watching, and what can you tell me that may surprise me?

The only reason I asked this question is that I am always suspicious of Google watching everything I do. I went to Google and was going to get some background information on Whisk and IMAGEN 4 for this post (after I took the challenge), when I clicked on the Google AI tab on the right.

This was a fresh tab. I now have 24 tabs open on this browser. I have yet to crush my new copilot laptop. I put extra strain on it by having CapCut and Notion open as well as 3 other browsers, and Copilot.

It has not died yet! Enough rambling on about my new 17" HP laptop.

I just knew Google was watching me; the Google AI suggested prompts were about Google Image Generation. Did I want to compare it, try it, learn more about it?

So I asked.

What might be surprising about Google's AI's "knowledge"

He promised that he was not watching; he would never do that! This is what he thought would surprise me:

• Inference, not memory: If enough conversational details are provided within a single chat, it can infer things about your context. For example, if you say, "I just got back from seeing the new 'Inside Out' movie," it will understand you're likely talking about Inside Out 2, but this understanding is temporary and is not stored for later use.
• • In other words, you need to remind Google AI what you were talking about... constantly!
    
• The scale of training data: The volume of data used to train large language models is immense. If all the text this AI was trained on were printed out, it would fill numerous libraries. This is how it can reference an incredible breadth of knowledge on a huge variety of topics, from advanced physics to specific movie trivia. - AI - The Master of Trivia. "Think You're Smarter Than A 5th Grader?"
  
• Predictive, not cognitive: Answers are based on predicting the most statistically likely next word or phrase in a sequence, not on a cognitive understanding of the world. The responses may seem human-like, but they are the result of a complex mathematical process.
  
  This is why all of your AI answers come straight down the middle and seem so lifeless.

There is no edge, no life, no sense of being. This is why you need to "breathe life" into everything that AI does for you. It was taught to walk in the middle of the road. Not too far left (yes, it does lean left) nor too far right. It wants to be "Just Right" Goldilocks.

I will leave you with this question:

When was the last time you were asked to produce your ID? What was it for?

Show Me Your ID (Even for Netflix) - https://youtu.be/wQvsSGyTfLI

Created by yours truly. Enjoy!