Video

Configure video generation capabilities in IntraLLM AI. This page explains how video prompts, generation engines, workflows, and default video parameters are configured.

Video Settings

This page configures video generation capabilities in IntraLLM AI. Video generation allows models and workflows to create short video outputs from text prompts using a selected generation engine and workflow configuration.

Video Prompt Generation

Video prompt generation controls how text prompts are prepared and passed to the video generation engine. Prompts may be provided directly by users or generated by models as part of a workflow.

Both positive and negative prompts can be used to guide video quality, style, and motion.

Video Generation Engine

The Video Generation Engine can be selected from the available engines list.

ComfyUI

When ComfyUI is selected:

  • Video generation requests are sent to a connected ComfyUI service
  • The engine executes a predefined workflow to generate video frames and encode output

Engine Configuration

  • ComfyUI Base URL: specifies the address of the ComfyUI service
  • ComfyUI API Key: used to authenticate requests to the ComfyUI service

The selected engine must be reachable from the IntraLLM AI backend.

ComfyUI Workflow

Workflow Definition

The ComfyUI Workflow defines the video generation pipeline in JSON format. The workflow controls:

  • Model loading
  • Prompt and negative prompt encoding
  • Sampling configuration
  • Frame generation
  • Video encoding and output format

You can:

  • Paste a workflow definition directly
  • Upload a workflow.json file exported from ComfyUI in API format

Workflow Nodes

Workflow nodes map video generation parameters to ComfyUI components.

Common node mappings include:

  • Prompt node (CLIPTextEncode) – required
  • Negative prompt node (CLIPTextEncode) – required
  • Model loader (UNETLoader)
  • Sampler (KSampler)
  • Latent video configuration (EmptyHunyuanLatentVideo)
  • VAE decode and video save nodes

Prompt and negative prompt node IDs must be correctly defined for video generation to function.

Default Video Generation Settings

Default settings are applied when not explicitly overridden by a workflow or template.

  • Set Default Model: default text-to-video model used for generation
  • Set Video Size: output resolution (width × height)
  • Set Video Frames: number of frames generated per video
  • Set Steps: number of sampling steps
  • Set CFG Scale: controls prompt guidance strength
  • Set Sampler: sampling algorithm
  • Set Scheduler: sampling scheduler
  • Set Batch Size: number of videos generated per request
  • Set Model Shift: model-specific sampling adjustment

These defaults provide a baseline for quality, motion smoothness, and performance.

How it works

When video generation is triggered:

  • A video prompt and negative prompt are prepared
  • The selected generation engine executes the configured workflow
  • Frames are generated, decoded, and encoded into a video format
  • The resulting video is returned and displayed in the platform

Usage considerations

  • Video generation is resource-intensive and requires sufficient GPU capacity
  • Higher frame counts, resolution, and step values increase generation time
  • Prompt quality strongly influences motion consistency and visual quality
  • Workflow configuration directly affects output format and performance

Quick checklist

  • Video generation engine selected
  • Engine connection configured correctly
  • Workflow loaded and validated
  • Prompt and negative prompt node IDs defined
  • Default model, size, frames, and steps configured
  • Adequate GPU resources available