Fashion with Imagen and Veo: Part 1

Creating Stunning Visuals in Vertex AI Media Studio

Written by Margaret Maynard-Reid, AI & Cloud GDE | 3D fashion designer

High-quality fashion photography is crucial throughout the fashion design process yet sourcing compelling stock photos for online businesses and social media remains a significant challenge. In this tutorial I will show you how to easily generate (and edit) product photos of visual consistency with Imagen 3, and then create stunning videos with Veo 2 on Google Cloud.

YouTube Shorts from the Google Cloud Tech official channel:

In-depth video tutorial posted on my YouTube channel:

Fashion with Imagen 3 and Veo 2, video by the author

First go to console.cloud.google.com, under Vertex AI Studio find the new Media Studio. Previously called Vision where there is only image generation / editing with Imagen, now Media Studio gives the option to generate images (Imagen 3), audio (Chirp 3), music (Lyria) and videos (Veo 2)!

Google Cloud Media Studio, screenshot by the author

Step 1. Generate images

First let’s create a photorealistic close-up portrait with Imagen 3 in a 1:1 aspect ratio.

Images generated with Imagen 3, image by the author

Step 2. Add a necklace

Then I use Inpaint (Insert) of Imagen 3 to add a pearl necklace.

Select the Inpaint (Insert) mode. Draw a rough shape of the necklace with the Brush mask, enter a prompt “A beautiful pearl necklace”, click on Generate. Then 4 images get generated, with the necklace added.

I downloaded one of the 4 images and continued to work on it.

Step 3. Close-up to wide shot

I would like to change from a close-up 1:1 image to a 9:16 full body image so that it shows not only the face but also the dress.

There are different ways to achieve this. One obvious choice is to use Outpainting which lets me change the aspect ratio of an image easily.

In the end, I decided to use Imagen 3’s Subject Reference feature (with the person as the reference) to get the full-body image that is more zoomed out than if I used outpainting.

9:16 portrait showing the dress and pose, Image by the author

Step 4. Change hand pose

Now the image is looking great but I want her right hand to look more relaxed.

Edit with box mask, Image by the author

Use the box mask to mark the right hand, type a simple prompt “Change her right hand to be relaxed”, click on Generate. We now have new images generated with the right hand looking much more relaxed.

Right hand more relaxed, image by the author

The hand pose change with Inpaint (Insert) is seamless, with the rest of the image intact.

Step 5. Change background

Fashion shoots in different locations can be very costly: from hiring a fashion model, stylist, make-up artist to studio / location rental. With GenAI, we can easily change the image background instead of the costly process of changing location.

First upload a starting image, then select “Product Background” mode, type a simple prompt such as “Outdoor setting of a botanical garden with a nature-inspired theme”, click on Generate. 4 images are created with a beautiful botanical garden backdrop in different variations.

Image by the author

Step 6. Change dress color

Virtual try-on or virtual dressing is a very useful feature for e-commerce. With Imagen 3, we can easily change the dress colors. I used the Person Reference feature again for changing the dress color.

One thing to note is to make this change, the prompt needs to be enhanced and simple prompts most likely will result in the images colors look bad.

Example with a simple prompt: “make the dress pale blue”. The color of the dress did change accordingly but the images have some sort of glare to the dress.

Result with a simple prompt, image by the author

Use the “help me write” button to enhance the prompt with Gemini. Notice that the dress colors with the enhanced prompt look much better.

Result with enhanced prompt, image by the author

Once I get a prompt that generates great results, I simply change the dress color in the prompt to generate new images. Now I have images of the dress in beautiful new colors.

Change dress color, image by the author

Step 7. Create a video with Veo 2

Google’s Veo 2 video generation model enables me to create stunning HD videos easily. I can either use “text-to-video” (a prompt as input) or “image-to-video” (a prompt plus a reference image) to create the videos.

I used one of the images created with Imagen 3 from the previous steps as reference, and created a short video that is 8 seconds long.

8-second video clip created with Veo 2, by the author

Interested in creating with Veo 2? Read my blog post Getting Started with Veo 2 to see the various options for you to start exploring Veo 2.

Making changes with Code

So far I showed you how to use Google Cloud’s Media Studio to make changes with one input image at at time. You can also make these changes at scale with code. In fact, being able to work with images and videos programmatically at scale, is the power of using these GenAI models on Google Cloud.

I hope you enjoyed reading my story of how to use Imagen 3 and Veo 2 for generating and editing fashion stock images. While my examples in this tutorial are about fashion, you can use the same techniques for any branded designs or content creation.

This is tutorial part 1 focuses mostly on Imagen 3 with a bit of Veo 2. Stay tuned for a step-by-step in-depth tutorial on Veo 2.

Watch the detailed video tutorial on YouTube: