2 posts tagged with "Alibaba Cloud"

What if you could create a complete, polished advertisement — visuals, voiceover, and full video — all for under $12?

That's exactly what I set out to prove in this project. In this blog post, I'll walk you through:

1. The tools I used and why
2. A cost breakdown
3. Comparisons with OpenAI Sora and WanAI
4. How I ultimately built the ad using Vertex AI VEO

Let's get started.

🧠 Why AI-Powered Ads Matter​

AI-powered ads are revolutionizing how we create and scale marketing content. Here's why:

1. ⚡ Speed: Full video ads can be created in minutes
2. 💰 Affordability: Cost is a fraction of traditional production
3. 🔄 Scalability: Generate multiple ad variants quickly
4. 🎥 No actors or equipment needed

With just a text prompt, these tools handle:

1. Script generation
2. Voice narration
3. Visual animation

🧪 Tools Compared: Sora, WanAI, and Vertex AI​

To benchmark what works best, I tested three AI platforms:

🔷 1. OpenAI Sora​

Sora delivered a high-quality 8-second cinematic video based on my prompt.

1. ✅ Smooth transitions
2. ✅ Excellent lighting and motion
3. ❌ Limited access
4. ❌ No built-in voice-over

Sora can be ideal for short cinematic storytelling, but not suitable for complete end-to-end ad creation unless you combine tools.

🔶 2. WanAI​

WanAI provided an easy-to-use interface, but I hit a limitation quickly.

1. ❌ Video generation failed
2. ❌ No output, even after retries
3. ⚠️ Likely due to the free-tier restriction

If you're using WanAI's free version, it may not be reliable for actual ad production.

🌐 Accessing WanAI via Alibaba Model Studio​

WanAI is part of Alibaba Cloud's generative AI offerings and can be accessed through the Model Studio interface. While not as frictionless as some Western platforms, it's still worth exploring — especially for experimentation.

🪪 Step 1: Create an Alibaba Cloud Account​

1. Visit https://www.alibabacloud.com
2. Register for a new account (email and phone verification required)
3. You may be prompted to complete identity verification depending on your region

🧠 Step 2: Navigate to Model Studio​

1. After logging in, search for Model Studio from the Alibaba Cloud Console or access it directly at: https://modelstudio.console.aliyun.com
2. Agree to terms and enable the service for your account

🖼️ Step 3: Find WanAI​

1. Inside Model Studio, explore the generative AI models section
2. Locate WanAI or its equivalent under video generation or multi-modal AI
3. Note: the UI is partially in Chinese — use browser translation if needed

🎬 Step 4: Provide Prompt and Generate​

1. Use a descriptive text prompt to generate your video
2. Wait for processing (this may take 1-2 minutes)
3. Note: in the free tier, results may be throttled or not generated at all

⚠️ In my case, WanAI did not produce a video under the free plan — likely due to quota restrictions or runtime limits.

If you're interested in evaluating WanAI for business use, consider upgrading to a paid Alibaba Cloud subscription to unlock full capabilities.

✅ 3. Vertex AI VEO​

This is where the magic happened. Vertex AI VEO allowed me to:

1. 🎞️ Generate a high-quality 24 1. 30 second video
2. 🗣️ Add professional voice narration
3. 🧱 Use slide-based or text-prompt-based generation
4. ✅ Fully control the visuals, timing, and tone

Best of all, the total cost stayed under $12 for the complete video with voice.

📊 AI Platform Comparison​

Below is a detailed comparison of the AI tools used during this ad creation journey:

🎬 How I Built the Ad in Vertex AI VEO​

🧰 Media Studio: A Unified Interface for Generative Creativity​

One of the most powerful components of Google's AI content ecosystem is the Media Studio — a clean, intuitive interface that brings together all major generative modalities.

From a single dashboard, you can:

1. Imagen 1. Generate stunning images using natural language descriptions
2. Chirp 1. Produce voice-over narration with lifelike clarity
3. Lyria 1. Compose custom music tracks based on mood, tone, and genre
4. Veo 1. Generate high-quality, dynamic video scenes

1️⃣ Start with a Prompt or Slide Structure​

In my case, I used a text prompt generated from the Model Studio prompt generator to define a narrative sequence of four emotional scenes:

Generate a series of four images depicting a business user, a middle-aged Asian man with short, dark hair, showcasing a range of emotions. In the first image, he is depicted feeling frustrated, his brow furrowed, and his lips pursed in a grimace. He is wearing a crisp, dark blue suit and a white dress shirt, conveying a sense of professionalism. The office is a typical corporate setting, with a large window behind him overlooking a bustling city landscape, with a cool color scheme. In the second image, he is intrigued, his eyes widened slightly as he leans forward in his chair, studying some documents on his desk. The office is the same, but a warm, muted orange tones color scheme is implemented, with soft light filtering in. In the third image, he is captured in a moment of excited, his arms raised in a gesture of triumph and a wide grin on his face. The office is the same, with high-contrast color scheme and dramatic lighting. In the fourth and final image, the user is portrayed as relieved and relaxed, his shoulders slumped in a comfortable posture, and a gentle smile playing on his lips. He is wearing the same suit as in the first three images, but now is sitting on a comfy couch in a modern, minimalist living room with a calm, pastel color scheme and soft, warm ambient lighting.

You can either:

1. Enter your enter a prompt similar to above, or
2. use Create Prompt from Model Studio prompt generator

2️⃣ 🖼️ Create or Upload an Image:​

If you already have an image of your character or scene, you can upload it to enhance, animate, or extend it using AI tools inside Media Studio.

Alternatively, you can generate a new image using Imagen, Google's generative image model.

For example, try a prompt like:

Photorealistic depiction of a middle-aged Asian business man sitting in a modern office, looking frustrated, with a city skyline visible through the window behind him

This will help you visually establish the first emotional scene for your AI-generated ad.

With links to documentation and API references at the top, Media Studio is designed for both no-code users and developers — enabling fast experimentation and production across image, voice, music, and video workflows.

Whether you're creating an ad, explainer video, music-backed clip, or voice-over narration — Media Studio serves as your creative AI command center.

3️⃣ Generate Video​

⚙️ VEO 2 Configuration and Output Setup​

To generate the final video outputs, I used VEO 2 — the latest iteration of Vertex AI's video generation model. Here's how I configured it for optimal results:

1. Model: VEO 2
2. Aspect Ratio: 16:9 (landscape, suitable for YouTube and web ads)
3. Number of Results: 4 (to get multiple variations for creative flexibility)
4. Video Length: 8 seconds per scene (ideal for short ad segments)

I also specified a Google Cloud Storage (GCS) output path for storing the generated videos:

gs://[your-bucket-name]/ads/

Storing outputs in GCS helped ensure persistence, easy access, and safe backup of all video assets during editing and review.

Finally, I enabled the Prompt Enhancement feature, which uses an LLM to automatically rewrite and enrich prompts for better video quality and fidelity. This dramatically improved the expressiveness and alignment of the output with my original creative intent.

🧩 Prompt-to-Image: From Concept to VEO Input​

After generating the composite image showing the four emotional expressions using the earlier prompt, I downloaded it and used a basic image editor to split it into four separate images — each representing one distinct emotion:

1. Frustrated
2. Intrigued
3. Excited
4. Relieved

I uploaded these images one by one into Vertex AI VEO, using them as input slides. For each image, I crafted a specific narrative prompt to guide the video generation:

1. Frustrated:

A small business user is frustrated at the speed, availability of his website — and that too after spending an exorbitant amount.

1. Intrigued:

The same user is now intrigued after discovering a new cloud-based website platform that promises better speed, uptime, and affordability.

1. Excited:

He is excited and overjoyed after seeing his website go live instantly, with blazing fast performance and zero downtime.

1. Relieved:

Now relaxed and smiling, he's enjoying peace of mind, knowing that his online business is running smoothly and cost-effectively.

This scene-by-scene approach allowed VEO to generate a seamless, emotionally resonant ad narrative — visual storytelling powered entirely by AI.

For each image, I provided a targeted contextual prompt. For example, for the first image:

A small business user is frustrated at the speed, availability of his website — and that too after spending an exorbitant amount.

This approach helped VEO sequence the emotional narrative visually and thematically, delivering a more dynamic and relatable ad experience.

Choose:

1. Aspect ratio: 16:9
2. Duration: 24-30 seconds
3. Resolution: 720p or 1080p

Within 1-2 minutes, VEO generated a fluid video matching my structure.

4️⃣ Add Voice-Over​

Now that the video was created, had to add a voice where I used Vertex AI's built-in voice generation:

1. English narration or multilingual depending on your ad.
2. Voice: Choose the voice. Ensure you are choosing male or female based on the character
3. Output: Clean, clear, and professional

You can review and re-generate multiple options, then export the audio.

5️⃣ 🎵 Background Music Generation (Optional)​

To enhance the emotional tone of the ad, I also generated background music using AI.

I used the following prompt to create the audio:

A melodious soft music required for advertisement where a light music plays.

The result was a gentle, unobtrusive melody that blended perfectly with the voice-over and visuals — helping to elevate the overall ad experience without overpowering the narration.

🎞️ Editing the Final Ad in CapCut​

Once all the individual video clips were generated by VEO and downloaded, I moved to the final phase of production: editing and assembling the ad in CapCut.

CapCut is a free and user-friendly video editor that offers a wide range of tools to polish raw footage into a professional-quality video.

🪜 Step-by-Step: Editing AI-Generated Videos in CapCut​

1️⃣ Launch CapCut and Start a New Project​

1. Open CapCut on your desktop or mobile device
2. Click “New Project”
3. Drag and drop all your downloaded video clips into the timeline

2️⃣ Arrange the Video Sequence​

1. Organize the clips in the intended narrative flow:
2. Frustrated scene
3. Intrigued scene
4. Excited scene
5. Relieved scene
6. Trim any silent lead-in or fade-out to keep the pacing tight

3️⃣ Add Transitions​

1. Insert subtle transitions (e.g., fade, slide, or zoom) between clips
2. This creates smoother scene shifts and a more cohesive visual flow

4️⃣ Import and Sync Voice-Over​

1. Import the AI-generated voice-over audio file
2. Drag it to the audio track in the timeline
3. Adjust timing and sync it with the visuals so that emotional cues align

5️⃣ Add Background Music​

1. Import the AI-generated background music file or you can choose from Canva or Capcut assets to have a background music
2. Lower the background music volume (e.g., 30-40%) so it complements the voice-over
3. Apply fade-in and fade-out to ensure the music blends smoothly

6️⃣ Overlay Text or Branding (Optional)​

1. Add on-screen text, brand logo, or a call-to-action (CTA) at the end:
2. “Visit now”
3. “Start your website in minutes”
4. “Powered by AI”

7️⃣ Export the Final Video​

1. Set export resolution to 1080p for best quality
2. Click Export
3. Save your finished ad for distribution or upload to your preferred platform

Using CapCut gave me complete control over how the scenes, voice, and music came together — transforming AI-generated assets into a polished, emotionally compelling advertisement ready for publishing.

💵 Cost Estimation: How I Calculated the $12 Budget​

One of the key takeaways from this experiment was demonstrating that a professional-quality ad can be produced for under $12, using only AI-powered tools — no actors, no cameras, no editing studio.

Here's how I calculated the total cost:

🔍 Reference Sources​

To estimate costs accurately, I referred to:

1. Yahoo Finance coverage of Google's VEO pricing
2. Google Cloud's official pricing calculator

📊 Breakdown of Costs​

🧠 Note: Costs are based on prompt complexity, video duration, resolution, and number of generated outputs. For most short-form ads under 30 seconds, this range is a realistic budget using cloud-based generative AI.

📈 Final Thoughts​

Creating professional ads used to take:

1. Studios
2. Actors
3. Editing software
4. Weeks of effort

Now, with tools like Vertex AI VEO you can create powerful, production-ready ads in minutes and under $12.

Ready to create your own AI ad? Here's what to do:

1. Sign up for Vertex AI

2. Use my prompt templates above

3. Experiment with different voices and music

4. Share your results!

For the product featured in this ad, check out cloudmysite.com - the affordable cloud hosting solution that inspired this project.​

📎 Key Takeaways​

1. 🎯 Use Sora if you have access and need cinematic shots
2. ⚠️ Avoid relying on WanAI's free tier for production work
3. ✅ Vertex AI VEO is the best choice for full, affordable ad creation with voice-over and visuals

Ready to create your own ad? Let me know, and I'll share a downloadable template and workflow to get you started!

Call to Action​

Choosing the right platform depends on your organizations needs. For more insights, subscribe to our newsletter for insights on cloud computing, tips, and the latest trends in technology. or follow our video series on cloud comparisons.

Interested in creating ads without the high agency cost? If yes, please contact us and we'll be more than glad to help you embark on not only building professional ads but also marketing on all platforms.

Deploying large language models (LLMs) like DeepSeek-V3 and DeepSeek-R1 has become more accessible with Alibaba Cloud's Platform for AI (PAI). This guide provides a step-by-step walkthrough to help you set up your Alibaba Cloud account, configure necessary resources, and deploy DeepSeek models efficiently.

Creating an Alibaba Cloud Account​

Before leveraging Alibaba Cloud services, you need to create an account:

1. Visit the Official Website: Navigate to Alibaba Cloud

2. Initiate Sign-Up: Click on the Free Trial button located at the top-right corner.

3. Choose Account Type: Select between Personal or Enterprise account types based on your need.

4. Enter Account Information:

   1. Email Address Provide a valid email address.
   2. Password: Set a secure password.

5. Verify Your Account:

   1. Email Verification: Click on "Send" to receive a verification code in your email. Enter the code to verify.
   2. Phone Verification: Alternatively, provide your mobile number to receive a verification code via SMS.

6. Agree to Terms: Read and accept the Membership Agreement, Privacy Policy, Product Terms, and Terms of Use.

7. Complete Sign-Up: Click on Sign Up to finalize the registration.

8. Add Billing Information: Before purchasing services, add your billing and payment details. Note that if you intend to use services in mainland China, real-name verification is required.

Selecting the Appropriate Region​

Choosing the right region is crucial for performance and compliance:

1. Geographical Proximity: Select a region closer to your target audience to reduce latency.

2. Service Availability: Ensure the desired services are available in the selected region.

3. Pricing Considerations: Be aware that resource pricing may vary between region.

4. Internal Communication: If your services need to communicate over an internal network, deploy them within the same region.

Note: Once a resource is created in a region, its region cannot be changed.

Understanding DeepSeek Models​

Alibaba Cloud's PAI Model Gallery offers access to advanced LLM:

1. DeepSeek-V3: A 671B parameter mixture of experts (MoE) model designed for complex language tasks.

2. DeepSeek-R1: Built upon DeepSeek-V3-Base, this model excels in reasoning and inference tasks.

3. Distilled Versions: For resource-constrained environments, distilled models like DeepSeek-R1-Distill-Qwen-32B offer a balance between performance and resource usage.

Deployment Methods on PAI​

Alibaba Cloud PAI provides multiple deployment option:

1. BladeLLM Accelerated Deployment: High-performance inference framework optimized for large-scale models.

2. SGLang Accelerated Deployment: Fast service framework tailored for large and visual language models, compatible with OpenAI APIs.

3. vLLM Accelerated Deployment: Widely-used library for LLM inference acceleration, offering efficient deployment capabilities.

4. Transformers Standard Deployment: Standard deployment method without inference acceleration, supporting both API and WebUI interfaces.

Step-by-Step Deployment Guide​

1. Access the PAI Console: Log in to your Alibaba Cloud account and navigate to the PAI Console.

2. Select a Workspace: In the upper-left corner, choose a region that aligns with your business requirement. From the left-side navigation pane, select Workspaces and click on your desired workspace.

3. Navigate to Model Gallery: Within your workspace, go to QuickStart > *Model Gallery.

4. Choose a Model: Browse the available models and select one that fits your needs, such as DeepSeek-R1-Distill-Qwen-32B.

5. Configure Deployment: **Click Deploy in the upper right corner. Choose your preferred deployment method and adjust resource settings as necessary.

6. Monitor Deployment: After initiating deployment, monitor the progress via Model Gallery > Job Management > *Deployment Jobs. Here, you can view deployment status and access call information.

Usage Recommendations​

To optimize model performance:

1. Temperature Settings: Set the temperature between 0.5 and 0.7, ideally at 0.6. This controls the randomness of responses — lower values produce more predictable output, higher values are more creative.
2. Prompt Formatting: Always include the full instruction within the user prompt. Do not use the system prompt field, as it is ignored by most DeepSeek deployments.
3. Math & Reasoning Tasks: For structured tasks like solving math problems or logical questions, append this phrase to your prompt:

Accessing the Model Interface​

After deployment, interacting with your model can be done via WebUI or API, depending on the method:

✅ For Transformers Standard Deployment​

1. Supports both API and WebUI.
2. Once deployed, click View Web App on the service page in the PAI console to open a simple chatbot-style interface.

⚡ For Accelerated Deployments (BladeLLM, vLLM, SGLang)​

1. These support API access only for higher performance.
2. Alibaba Cloud provides Gradio-based WebUI code in the deployment page, which you can use to launch a custom interface locally.

Resource Considerations​

Deploying full-scale models requires substantial cloud resources:

1. DeepSeek-V3 or full R1 models may require:

2. 8 GPU cores

3. 96 GB GPU memory

4. Up to 2 TB of system memory for optimal inference

5. Distilled models like DeepSeek-R1-Distill-Qwen-32B offer:

6. Lower resource consumption

7. Faster response times

8. Reduced costs

📌 You can choose the instance size in the deployment UI. For most users, we recommend starting with a distilled model and upgrading based on need.

🧠 Deploying DeepSeek-R1-Distill-Qwen-32B on Alibaba Cloud​

Once you're inside the Alibaba Cloud PAI Model Gallery, follow the steps below to deploy the DeepSeek-R1-Distill-Qwen-32B model — a powerful yet resource-efficient distilled version designed for high-performance inference at lower cost.

🔍 Step 1: Locate the Model​

1. Navigate to your PAI workspace.
2. Click on Quick Start > Model Gallery.
3. In the search bar, type Qwen-32B or DeepSeek.
4. Locate the model: deepseek-r1-distill-qwen-32b.
5. Click on the model card to open its detail page.

⚙️ Step 2: Click “Deploy”​

On the model detail page:

1. Click the blue “Deploy” button in the top-right.

2. Choose your deployment method:

   1. For standard usage with a WebUI, choose Transformers Standard Deployment.
   2. For faster performance, choose BladeLLM, vLLM, or SGLang depending on your preference and API-only usage.

3. In the deployment configuration panel:

   1. Service Name: Auto-generated or give it a custom name like deepseek-qwen32b-service.
   2. Resource Type: Select your desired GPU/CPU resource. For distilled models, fewer resources like ecs.gn7i-c8g1.2xlarge may be sufficient.
   3. Click Next and then Deploy.

Within a few minutes, your service will be provisioned.

🌐 Step 3: Get the API Endpoint​

Once deployed:

1. Go to Job Management > Deployment Jobs.
2. Click on the deployed service.
3. Copy the API endpoint URL provided — you'll use this in Postman or any HTTP client.

📬 Step 4: Test the Model in Postman​

Here's how to use Postman to make your first inference call:

🧾 Request Type: POST​

📍 URL:​

Use the endpoint you copied, for example:

https://your-service-name.pai-eas.aliyuncs.com/api/predict

🧠 Headers:​

Content-Type: application/json
Authorization: Bearer <your-access-token>  # Optional depending on your setup

📦 Body (raw JSON):​

For deepseek-r1-distill-qwen-32b, use the following format:

{
  "model": "deepseek-r1-distill-qwen-32b",
  "inputs": [
    {
      "role": "user",
      "content": "What is the capital of France?"
    }
  ],
  "parameters": {
    "temperature": 0.6,
    "max_tokens": 512
  }
}

📝 You can modify:

1. "content": to ask any question.
2. "temperature": to control randomness (0.6 is ideal).
3. "max_tokens": controls how long the output will be.

▶️ Click “Send”​

You'll receive a JSON response containing the model's answer under content.

✅ Tips for Better Results​

1. For math or reasoning problems, include:

"content": "Solve this step by step: 248 + 382. Put the final answer in \\boxed{}"

1. Use "stream": false unless your deployment supports streaming output.

🧠 What Happens Behind the Scenes​

By deploying Qwen-32B, you're spinning up an inference-optimized container hosted on Alibaba Cloud's GPU infrastructure, preloaded with model weights and optimized runtime. It's production-grade and ready to scale with API Gateway, logging, and access control if needed.

Integrating with Alibaba Cloud API Gateway​

For secure and scalable integration, expose your model's inference endpoint through Cloud-Native API Gateway:

Step 1: Create an AI Service​

1. Go to API Gateway Console > AI Services > Create AI Service
2. Fill in:
3. Service Source: PAI-EAS
4. Model Supplier: (select appropriate deployment method)
5. Endpoint URL: Found in your model's service page

Step 2: Publish an AI API​

1. Navigate to APIs > Create API
2. Attach the AI Service you created
3. Customize method (POST), headers, and body parameters

Step 3: Monitor & Secure​

1. Use the API Gateway's tools to:
2. Throttle requests
3. Apply authentication
4. View logs and response times

This ensures stable public access while protecting backend models from abuse or excessive load.

Conclusion​

Alibaba Cloud's Platform for AI (PAI) has made advanced LLM deployment accessible to everyone — from developers to enterprises. With just a few clicks, you can spin up state-of-the-art models like DeepSeek-R1 and DeepSeek-V3 and integrate them into your apps or websites securely and efficiently.

By combining one-click deployment, high-performance inference frameworks, and secure API exposure, Alibaba Cloud offers a best-in-class experience for anyone looking to harness the power of large language models.

Want help building your first chatbot, virtual assistant, or generative app with DeepSeek? Reach out to us at Arina Technologies — we specialize in AI-based cloud solutions.