​
Getting Started Privately

​
Initial Setup

  1. Type /model to pick a model that suits your needs
  2. Enter the newly created private active Thread (If no System Prompt was entered previously, the thread name will be a numerical string)
  3. Type / to access media generation models and help commands:
    • text-to-video
    • text-to-image
    • image-to-text
  4. Prompt the model with your request

​
Understanding Private Threads

  • Private thread creation messages are ephemeral (only visible to the user creating the ModelChat thread)
  • Your active threads can be found by checking ⁠model-chat
  • The channel menu will show a blinking dot as ⁠model-chat threads are private by default
  • Access channel names under active threads via right-click (context menu)
  • Additional users must be invited using the @ symbol once the model loads in the channel

​
System Prompts

Select System Prompt for text/vision models and type the system_prompt to define the model’s purpose and/or Role. This action will name the thread. Optional

​
Media Generation Guidelines

​
Image and Video Input

We currently support the following multimodal models: Nova Lite/Pro, Reels & Canvas, Claude Sonnet, Haiku 3 & Llama 11b/90b

​
Image and Video Input

For Nova Lite/Pro (video-to-text):

  • First, upload the video
  • Provide a brief description of the video content on the prompt
  • Prompt the model with the desired output

For Reels (text-to-video):

  • Use /text-to-video to generate 6 seconds of video
  • Use /image-to-video to convert images to videos
  • Videos might take 5-6 minutes to generate

For Canvas (text-to-image):

  • Use /text-to-image to generate images from text descriptions
  • Depending on the amount of Images, generations might take 1-2 minutes to generate

For Claude and Llama (multimodal vision models):

  • First, upload the image
  • Provide a brief description of the image content on the prompt
  • Prompt the model with the desired output

​
Character Limits

  • Image generation: Maximum 1024 characters
  • Video generation: Maximum 512 characters

​
Model-Specific Notes

  • For video understanding, use Nova Lite or Nova Pro Models
  • For media generation workflows, we recommend pre-prompting a multimodal model (Claude, Llama) with image or text context

​
Important Notes

  • Mediagen processes requests in an orderly queue which include cool down periods
  • Reference help commands under / commands for detailed guidance

​
Need Support?

  • DM @StreamPod AI
  • Ask questions in ⁠general-support
ModelchatAudiogen