Image to Video

This guide will help you generate video from a single source image using LTX-2 in ComfyUI.

When to Use

Image to Video workflows are ideal when you need to:

• Animate a specific frame - Bring a static image to life with motion
• Maintain identity consistency - Keep exact character appearance and framing
• Control composition precisely - Start from a known visual foundation

The LTX-2 workflow generates synchronized video and audio in a single pass.

Workflow Video Walkthrough

Before Starting

ComfyUI and the LTX-2 nodes need to be installed. See our Installation guide for full instructions.

You’ll also need:

• Source image to add to the workflow
• Sufficient VRAM - At least 32GB recommended - more is better
• Image-to-Video workflow from the LTX-2 repository

Step-by-Step Guide

1. Load the Workflow

1. Open ComfyUI
2. Load the Image-to-Video workflow JSON from the LTX-2 repository
3. The workflow will display as a node graph

2. Load Model Checkpoints

Locate the LTXVCheckpointLoader node and configure:

• Model selection - Choose between full or distilled checkpoint
  • Distilled model (default): Fast generation, optimized for iteration
  • Full model: Higher quality, slower generation (see Full Model Variant)
• VAE: Automatically loaded with the checkpoint
• Text encoder: Gemma CLIP for processing prompts

3. Load Your Source Image

Find the LoadImage node:

• Browse to select your image file
• Supported formats: PNG, JPG, WebP
• Recommended: Use images that match your target aspect ratio

The image will be automatically resized to match your configured resolution.

4. Configure Parameters

Find the LTXVImgToVideoConditioning node and set:

Resolution

Choose from standard aspect ratios or specify custom dimensions:

• 1920×1080 (16:9 landscape) - Full HD
• 1080×1920 (9:16 portrait) - Vertical video
• 1280×720 (16:9) - HD, lower VRAM (~720p)
• 768×512 (3:2) - Fast iteration
• 640×640 (1:1) - Square format

Frame Count

• Maximum: 257 frames (~10 seconds at 25fps)
• Recommended range: 121-161 frames for balanced quality and memory usage
• Shorter videos: Use 65-97 frames for quick iterations

Frame Rate

• 24 fps - Cinematic feel, film-like motion
• 25 fps - Standard (default), works for most content
• 30 fps - Smooth motion, better for fast action
• 48-60 fps - High frame rate for very fast or dynamic motion

5. Write Your Prompt

Describe the motion, action, and audio you want to generate from your source image.

Strong Prompts Include

• Motion description - How the scene should move or change
  • Example: “Camera slowly zooms in while the character turns their head to look at the sunset”
• Character actions - What subjects in the frame should do
  • Example: “The woman smiles and waves, her hair gently blowing in the breeze”
• Camera movement - Shot type and camera motion
  • Example: “Steady tracking shot following the subject from the side”
• Audio cues - Dialogue, music, ambient sound effects
  • Example: “Soft acoustic guitar playing, distant ocean waves, seagulls calling”

See our Prompting guide for additional tips on writing good prompts.

6. Configure Sampling

Find the KSampler or LTXVSampler node:

Steps

• Distilled model: 8 steps (represented by 9 manual sigma values in workflow)
• Full model: 20-50 steps (higher quality)
• Start with lower values and increase if quality is insufficient

CFG (Classifier-Free Guidance)

• Range: 2.0-5.0
• Lower values (2.0-3.0): More creative, less prompt adherence
• Higher values (4.0-5.0): Stronger prompt adherence, less variation
• Recommended: 3.0-4.0 for balanced results

Sampler Type

• euler: Fast, good for testing
• dpmpp_2m: Higher quality, slightly slower
• Res-2-S: Best for full model
• Experiment to find your preference

Seed

• Fixed seed: Reproducible results for iteration
• Random seed: Explore variations

7. Multi-Scale Generation

The workflow uses a two-stage approach for efficiency:

Stage 1: Base Generation

• Generates at half the target resolution (e.g., 1080p generates as 960×540 first)
• Fast iteration for testing prompts and parameters
• Establishes overall composition and motion

Stage 2: Upscale Pass

• Uses the LTXVUpscale node to double the resolution (2x)
• Refines details and enhances quality
• Frame rate must match Stage 1

This approach provides:

• Faster experimentation at lower resolution
• High-quality final output without generating everything at max resolution from the start
• Memory-efficient workflow

8. Sampling and Decoding

Image-to-Video uses LTX-2’s unified audio-video architecture:

1. Separate generation - Audio and video latents are created independently
2. Merge and sample - Latents are combined for joint processing
3. Split and re-merge - Streams are separated, then recombined for coherent output
4. Tiled decoding - Minimizes memory usage during final decode

The workflow uses LTXVDecoder nodes:

Audio Decoder

• Processes audio latents separately
• Outputs synchronized audio stream
• Supports dialogue, music, and ambient sound

Video Decoder

• Uses tiled decoding to minimize VRAM usage
• Processes video latents in manageable chunks
• Maintains quality while reducing memory requirements

Both distilled and full model variants are supported.

9. Save Output

Configure the Save Video node:

• Format: MP4 (default), MOV, or WebM
• Codec: H.264 (compatibility) or H.265 (smaller files)
• Audio: Automatically embedded from audio decoder
• Filename: Use descriptive names for organization

Advanced Techniques

Full Model Variant

The full model workflow provides higher quality at the cost of longer generation time.

Key Differences

• Uses the full LTX-2 checkpoint and specialized VAE
• Stage 1: 15-20 steps (up to 40 for experimentation)
• Uses LTXV Scheduler instead of manual sigmas
• Applies distilled LoRA in Stage 2 (recommended strength: 0.6)

Using LoRAs

Add LoRALoader nodes to customize:

• Style LoRAs: Apply artistic styles or visual aesthetics
• Motion LoRAs: Enhance specific types of movement
• Character LoRAs: Maintain consistent character appearance

See the LoRA guide for training and usage.