From Setup to Success — All in One Place

Vertex AI

1. Content Generation
2. Video Generation

Peeklogic AI Orchestrator integrates with Google’s Vertex AI to bring advanced enterprise-grade generative models to Salesforce. This integration requires a valid Google Cloud Project setup.

Content Generation:

Use Vertex AI’s text models (including Gemini via Vertex) to generate text, code, or structured content.

Parameter Configuration Constraints The Vertex AI API is strict regarding parameter values.

• Avoid Zero Values: Do not set fields like Temperature, Top K, Candidate Count, or Max Output Tokens to 0. This will cause an “Invalid JSON payload” or “INVALID_ARGUMENT” error.
• Defaults: If you do not wish to customize a parameter, leave the field empty to use the Google default values. * Top K: If you encounter errors regarding topK, ensure the field is empty rather than 0.

Required fields:

• Name: Friendly name for your configuration.
• Region: The Google Cloud region where your resources are hosted (e.g., asia-south2, us-central1).
• Model: Select a supported Generative AI model (e.g., Gemini Pro, Gemini Flash).

Warning: Selecting non-generative models (such as imageclassification or legacy predictive models) from the dropdown list will result in an INVALID_ARGUMENT error.

Important Note on Model Selection: The dropdown menu may display a comprehensive list of models, including legacy versions preserved by Google. Some of these may no longer be supported, resulting in an “INVALID_ARGUMENT” or “400” error. For the most up-to-date list of stable and available models, please refer to the official documentation: Google Vertex AI Model Versions

• Project Id: Your Google Cloud Project ID.

Generation Configuration:
These parameters control how the model generates responses:

• Candidate Count: The number of response variations to return for a single prompt.
• Max Output Tokens: The maximum number of tokens to generate in the response.
• Stop Sequences: A list of strings that tell the model to stop generating text if encountered.
• Temperature: Controls randomness. Higher values produce more creative output; lower values produce more deterministic output.
• Top K: Limits the model to the top K most probable tokens at each step.
  • Category: Defines the type of content to filter.
    • Harassment: Negative or harmful comments targeting identity.
    • Hate speech: Content that is rude, disrespectful, or profane.
    • Sexually explicit: References to sexual acts.
    • Dangerous: Promotes harmful acts.
    • Civic integrity: Election-related queries.
  • Threshold: Defines how strict the filter should be.
    • Block none: Always show content regardless of probability.
    • Block few: Blocks only high-probability unsafe content.
    • Block some: Blocks medium or high-probability unsafe content (Standard).
    • Block most: Strict filtering; blocks low, medium, and high probability.Top P: Filters tokens based on cumulative probability (Nucleus sampling).
      Vertex Safety Settings:
      You can configure safety filters to block harmful content. You can add multiple categories by clicking the add button.

Video Generation:

Use Vertex AI’s Veo models to generate high-quality videos from text prompts.

Asynchronous Generation Process Generating high-quality video with Veo 3 is a resource-intensive task and does not happen instantly.

• Job Scheduling: Upon sending a request, the system schedules a background job. The immediate response will confirm: “Job has been scheduled” and provide a Job ID.
• Processing Time: Generation may take significant time (from several minutes to over an hour) depending on the complexity and current load.
• Retrieving the Video: The final video file will be saved automatically to the Storage Uri (Google Cloud Storage bucket) you configured. Please check your bucket for the result using the Job ID or timestamp.

Required fields:

• Name: Friendly name for your configuration.
• Project Id: Your Google Cloud Project ID where Vertex AI is enabled.
• Region: The Google Cloud region where your resources are hosted (e.g., asia-south2).
• durationSeconds: The length of the generated video in seconds (e.g., 8).
• Storage Uri: Enter the path in the following format: gs://your-bucket-name/your-folder-name (Example: gs://peeklogic-video-output/Veo3)

• Format: gs://your-bucket-name/folder/
• Note: Do not use https:// links.

Optional fields:

• Model: The specific video generation model to use (e.g., veo-2.0-generate-001).
• Aspect Ratio: The format of the video frame (e.g., 16:9 for widescreen, 9:16 for mobile).
• Compression Quality: Controls the visual quality versus file size (e.g., optimized).
• Generate Audio: If checked, the model will attempt to generate synchronized audio for the video.
• Negative Prompt: Specify what you do not want to see in the video (e.g., “blur”, “distortion”).
• Person Generation: Controls the policy for generating images of people (e.g., allow_adult).
• Resolution: The output resolution of the video (e.g., 720p).
• Sample Count: The number of video variations to generate for a single prompt.

Video Preview Limitations The AI Record Component may not be able to play the video directly if the output format is a raw Google Storage URI (gs://). To view the generated video, navigate to your Google Cloud Console and access the file directly from the Storage Bucket.