AI Ads Commercial Tutorial: How to Make Ads Training Product LoRAs

Introduction

Creating a professional-looking advertisement for brands or businesses has never been easier, thanks to advancements in AI technology. In this article, we will walk through the process of creating an AI-powered commercial, using a Chinese electric car model—the Huawei Avatar 11—as our example. This step-by-step guide works for advertisements across various platforms, including TV and YouTube.

Step 1: Collecting Images

To start, we need high-quality images of the Huawei Avatar 11. I recommend using Google Images for this purpose. By searching "Huawei Avatar 11," I was able to download approximately 40 images using a Chrome extension called Image Downloader.

Step 2: Training the AI Model

Next, we will train an AI model using the images collected. For this, we will use Flux Gym in conjunction with Pinocchio. Here’s a brief overview of the training process:

Open Flux Gym: Upload all the images of the electric car.
Name the Model: I labeled my model as "Huawei Avatar 11" and set a unique trigger word—"ha_a11_yWQ_KB."
Configure Settings: I set the VRAM to 12 GB and reduced the training cycles to five per image while keeping the other settings at their defaults.
Add Captions: Using Florence 2, I added captions to the images, ensuring to replace any incorrect references, such as car names like Tesla or Jaguar, with the term “electric EV.”
Begin Training: After everything was set, I initiated the training process, which took around 10 hours to complete.

Once the training process was done, I accessed the output folder to find my newly trained models.

Step 3: Generating Images

The next step involves using my trained models to generate new images. I switched over to Comfy UI and selected both the newly trained models and the Flux One Dev Bas model. Remember to include the trigger word in the prompt when generating images—this time, I wanted to create pink cars.

Once I clicked "Q prompt," I eagerly awaited the results. The generated images were remarkably accurate compared to the original dataset.

Step 4: Upscaling Images

Having created the images, it was time to enhance them. Using Comfy UI again, I employed a workflow involving the Flux upscaler. After uploading the generated images, I initiated the upscaling process. The results were truly impressive, with sharper images containing greater detail, especially in areas like the car rims.

Step 5: Creating the Video

With the images ready, I moved to Runway ML to convert the images into a video. The "Images to Video" feature allowed me to upload my upscaled pictures and generate a 10-second video. I sought ChatGPT's assistance to generate specific prompts for my pink SUV advertisement, yielding around 20 prompts which I adjusted based on the uploaded images.

Following the video generation, I downloaded the clips and used Topaz Video AI to upscale them to 4K resolution.

Step 6: Final Editing

For the final step, I opened CapCut (or any preferred editing software) and imported all the assets into the project. Music selection is crucial, as it sets the project's tone. I will explore the topic of music selection based on ad types in a future video, so let me know if you’re interested in that!

With everything in place, I edited the project into a cohesive commercial, which I will now speed ramp to share the final result. Additionally, I will include snippets of other recent ads created using the same AI processes.

Conclusion

Creating AI-powered commercials is a fulfilling process involving image collection, model training, image generation, upscaling, video creation, and final editing. If you found this tutorial helpful, please like and subscribe for more in-depth tutorials on using AI to create real-world job opportunities.

Happy creating!

Keywords

AI-powered commercial
Huawei Avatar 11
Flux Gym
Pinocchio
Image Downloader
Comfy UI
Video creation
Upscaling
ChatGPT
CapCut

FAQ

Q1: What AI model do you recommend for creating commercials?
A1: I recommend using the Flux Gym in conjunction with Pinocchio for training AI models.

Q2: How long does the training process take?
A2: The training process can take up to 10 hours, depending on the number of images and the performance of the GPU used.

Q3: What type of images can I use for my commercial?
A3: You can use any high-quality images of the product you're promoting, downloaded from sources like Google Images.

Q4: Can I create videos from AI-generated images?
A4: Yes, platforms like Runway ML allow you to convert images into videos easily.

Q5: What software tools can I use for editing my final video?
A5: You can use various editing software tools. For this project, I used CapCut, but others like Adobe Premiere Pro or Final Cut Pro could work as well.