PainterLongVideo ComfyUI Tutorial for Wan2.2 Long Video

Local or hosted? PainterLongVideo is a local ComfyUI workflow for people who want node-level control. If you first need to validate the reference image, prompt, and motion direction, run a hosted image-to-video test before committing to the model downloads; see credit costs and plans.

Prefer not to install anything? The generator below is the same hosted LongCat generator as on the home page (image-to-video and text-to-video) — you can run it right here:

What is PainterLongVideo?

PainterLongVideo is a revolutionary ComfyUI custom node that fundamentally changes how creators approach long-form video generation with Wan2.2. Unlike traditional approaches that struggle with consistency decay and unnatural slow-motion artifacts, PainterLongVideo maintains coherent motion, global scene consistency, and flawless transitions across unlimited video length.

Developed by the PainterX team, this breakthrough node takes reference from previous video segments while maintaining consistency through anchor images - enabling creators to construct 30-second, 60-second, or even multi-minute videos that feel like single continuous shots.

The Slow Motion Problem Solved

Traditional Wan2.2 suffers from a frustrating limitation: generated videos feel unnaturally slow, with movements appearing jerky or delayed despite looking visually realistic. PainterLongVideo fixes this through the motion_amplitude parameter, which injects realistic, natural movement back into the generation.

Real-World Impact:

• Default motion feels 30-40% slower than natural
• motion_amplitude adjustment brings movement to 95%+ natural speed
• Difference is immediately obvious when viewing before/after videos

Key Features and Benefits

1. Seamless Video Continuation

Unlike frame-by-frame stitching, PainterLongVideo analyzes the previous video segment and understands its motion patterns, then continues the action naturally into the next segment.

How It Works:

• Analyzes last frames of previous clip
• Extracts motion information
• Applies similar motion to new segment
• Result: Zero visible transitions, continuous action

2. Global Consistency Maintenance

The anchor image (your initial reference image) is referenced throughout the entire workflow, maintaining identity consistency across 30+ seconds of video generation.

Benefits:

• Character face stays consistent
• Clothing/outfit remains the same
• Environmental elements don't shift
• Color palette maintains consistency
• Lighting direction stays constant

3. Slow Motion Fix

The critical motion_amplitude parameter controls motion speed:

• Default (1.15): Smooth but feels slightly slowed
• 1.2-1.25: Natural motion, recommended for most
• 1.3+: Accelerated motion, use for dynamic scenes

This single adjustment often makes the difference between amateur-looking and professional-looking videos.

4. Multi-Shot Workflow Support

Chain multiple shots together in a single workflow:

• Shot 1: Initialize with reference image
• Shot 2-6: Continue from previous shot's latent
• Total: 30-second coherent narrative

System Requirements

Minimum GPU VRAM for PainterLongVideo

Computation Time Per 5-Second Shot

Using the Distiller LoRA reduces times by 50-60%.

Installation and Setup

Step 1: Install ComfyUI (If Not Already Done)

If you already have ComfyUI installed, skip to Step 2.

On Windows:

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txt
python main.py

On macOS:

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txt
python main.py

On Linux:

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txt
python main.py

Step 2: Install ComfyUI-PainterLongVideo Custom Node

Navigate to your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/princepainter/ComfyUI-PainterLongVideo.git

Alternatively, use the ComfyUI Manager if installed:

1. Open ComfyUI interface
2. Click "Manager" in the top menu
3. Search for "PainterLongVideo"
4. Click "Install"
5. Restart ComfyUI

Step 3: Download Model Files

You'll need these models for the workflow. Download from Hugging Face:

Main Wan2.2 Model (Choose One):

• wan-2.2-base-bf16 (~27GB) - Highest quality, requires RTX 4080+
• wan-2.2-smooth-mix-uncensored (~27GB) - Recommended, excellent quality, unrestricted content
• wan-2.2-remix (~27GB) - Alternative quality option

Optional Speed-Up Models:

• wan-2.2-distiller-lora (~2GB) - Reduces steps from 50 to 10-12, significant speedup
• wan-2.2-4step-lora (~500MB) - Ultra-fast, acceptable quality

Required Supporting Models:

• VAE: wan-2.1-vae.safetensors (~5GB)
• Text Encoder: t5-xxl-clip-bf16.safetensors (~4GB)

Installation Locations:

ComfyUI/models/diffusion_models/          # Main models here
ComfyUI/models/loras/                      # LoRA files here
ComfyUI/models/vae/                        # VAE models here
ComfyUI/models/clip/                       # Text encoders here

Understanding the PainterLongVideo Workflow

Node Structure Overview

A typical PainterLongVideo workflow consists of:

1. User Input Section (Left side)

   • Initial anchor image upload
   • Resolution and prompt settings
   • Motion control parameters

2. Shot Groups (Center, repeated 6 times)

   • Group 1: Uses PainterI2V node (initial generation)
   • Groups 2-6: Use PainterLongVideo node (continuation)

3. Output Section (Right side)

   • Video merge and combine
   • Final video export

Key Nodes Explained

PainterI2V (First Shot Only)

Generates the initial video segment from your anchor image.

Inputs:

• Image: Your reference/anchor image
• Text Prompt: Description of the motion and scene
• Motion Frames: Number of frames to analyze (recommend 10)

Outputs:

• Video latent (feeds to next shot)
• Preview frames

PainterLongVideo (Shots 2+)

Continues video generation based on previous segment.

Inputs:

• Previous Video Latent: Output from prior shot's sampling
• Anchor Latent: Initial image encoded (maintains consistency)
• Motion Frames: Frames from previous segment to reference (3-7 recommended)
• Motion Amplitude: Speed adjustment (default 1.15, increase for faster motion)
• Prompt: New segment's scene description

Key Parameters:

Workflow Configuration

For Standard Setup (6 Shots = 30 Seconds):

Shot 1 (0-5s):    Use PainterI2V + Anchor Image
Shot 2 (5-10s):   Use PainterLongVideo + Shot1 Latent
Shot 3 (10-15s):  Use PainterLongVideo + Shot2 Latent
Shot 4 (15-20s):  Use PainterLongVideo + Shot3 Latent
Shot 5 (20-25s):  Use PainterLongVideo + Shot4 Latent
Shot 6 (25-30s):  Use PainterLongVideo + Shot5 Latent

For Extended Setup (10 Shots = 50 Seconds):

Simply copy the Shot 6 group, paste after it, and increment the shot number. Chain the previous shot's latent output to the new shot's input.

Detailed Setup Guide

Configuring Model Loading

1. Model Loader Node Settings:

   • Model: Select your Wan2.2 variant
   • Block Swap: Set based on VRAM (0-10 for high VRAM, 30-50 for low VRAM)
   • Attention: sage_attention if available, else sdpa
   • Torch Compile: enabled only if you have Visual Studio C++ tools installed

2. LoRA Setup (Optional but Recommended):

   • If using Distiller LoRA:
     • Load through "Load LoRA" node
     • Strength: 1.0 (full effect)
     • Reduces sampling steps by 80%

Setting Video Parameters

Resolution Considerations:

Frame Count Per Shot:

• 81 frames (5 seconds at 16fps) - Standard
• 93 frames (6 seconds at 16fps) - Slightly longer
• 48 frames (3 seconds at 16fps) - Quick tests

Overlap Frames:

• Default: 13 frames
• For smoother transitions: 15-20 frames
• Trade-off: Higher = better consistency, slower generation

Writing Effective Prompts for Each Shot

Unlike single-shot generation, multi-shot requires prompts that continue the narrative.

Shot 1 Prompt (Establish Scene):

A woman sits at a wooden desk in a modern apartment,
looking at a laptop screen with focused expression.
Natural window lighting from the left, afternoon sunlight.
She begins typing. Subtle camera push-in over 5 seconds.

Shot 2 Prompt (Continue Action):

The woman continues typing at her desk, eyes concentrated on screen,
afternoon sunlight illuminates her face with warm tones.
The camera slowly pushes in from medium shot to close-up of hands on keyboard.

Shot 3 Prompt (Advance Narrative):

Woman finishes typing, sits back in her chair with satisfied expression,
hands leaving the keyboard. She smiles slightly and looks out the window.
Same apartment lighting, wider camera angle showing more of the room.

Why This Structure Works:

• Shot 1: Establishes who/what/where
• Shot 2: Continues with motion and detail
• Shot 3: Moves story forward
• Each references previous shot's lighting and environment
• Prompts avoid contradicting each other

Advanced Technique: Motion Amplitude Tuning

The motion_amplitude parameter is the secret to professional-looking videos.

Understanding Motion Amplitude Values

1.0 (Minimal Motion):

• Extremely subtle movement
• Useful for: Static interviews, still compositions
• Drawback: Feels artificial and boring

1.15 (Default, Slightly Slow):

• Natural but perceptibly slower than real-world
• Useful for: Calm, meditative content
• Result: Good consistency but motion feels delayed

1.2-1.25 (Recommended, Natural):

• Matches real-world motion speed
• Useful for: General content, cinematic videos
• Result: Professional appearance, natural movement

1.3 (Fast Motion):

• Noticeably accelerated
• Useful for: Action sequences, dynamic scenes
• Drawback: Can look artificial if too high

1.5+ (Very Fast):

• Only use for specific effects
• Useful for: Time-lapse, comedic effects
• Drawback: Usually looks unnatural

How to Apply Motion Amplitude Adjustments

In the PainterLongVideo node:

1. Locate the "Motion Amplitude" slider/input field
2. Set to 1.2 as baseline
3. Generate a test segment
4. If motion feels slow: increase to 1.25 or 1.3
5. If motion feels fast: decrease to 1.15
6. Fine-tune based on footage

Motion Amplitude by Content Type

Troubleshooting Common Issues

Issue 1: Character Face Changes Between Shots

Problem: Anchor image isn't maintaining identity consistency.

Solutions:

1. Verify anchor latent connects to all PainterLongVideo nodes (pink connection)
2. Check that Shot 1 uses high-quality reference image
3. Increase motion_frames in PainterLongVideo nodes (try 7-10)
4. Add character description to every prompt (e.g., "same woman as before")
5. Keep lighting descriptions identical across all prompts

Issue 2: Slow Motion Despite Configuration

Problem: Motion still looks unnaturally slow.

Solutions:

1. Increase motion_amplitude to 1.3 (if not already)
2. Verify you're not using extremely low sampling steps (min 12-15)
3. Check that motion_frames isn't set too low (use 5-7 minimum)
4. Test with Smooth Mix model instead of base model
5. Increase overlap_frames to 15-20

Issue 3: Color Shifts Between Shots

Problem: Colors gradually shift or drift across shots.

Solutions:

1. Add color correction node after video merge
2. Include specific color descriptions in prompts ("maintain warm tones")
3. Ensure lighting descriptions match across all prompts
4. Use color-match node if available in workflow
5. Reduce number of shots and keep each longer

Issue 4: Out of Memory Errors

Problem: Generation crashes with CUDA out of memory error.

Solutions:

1. Increase block_swap value (try 40-50)
2. Reduce resolution (try 854x480 instead of 1280x720)
3. Lower sampling steps (minimum 10 with Distiller LoRA)
4. Reduce frames per shot (try 81 instead of 93)
5. Disable unused shots (right-click and "Set to Never")
6. Close other GPU applications

Issue 5: Blurry Output Quality

Problem: Generated video quality is soft/blurry.

Solutions:

1. Increase sampling steps to 20-25 (minimum with Distiller)
2. Improve prompt specificity (more detailed descriptions)
3. Increase resolution to 1280x720
4. Ensure VAE is correct version (wan-2.1-vae)
5. Add sharpening in post-production

Real-World Example: Creating a 30-Second Product Demo

Let's create a professional product demo video using PainterLongVideo.

Example: Phone Unboxing Sequence

Anchor Image: Product photo of new smartphone in packaging

Shot 1 (0-5s) - The Box:

A brand new smartphone in sleek black box, product shot style lighting,
hand reaches into frame from top, beginning to open the box.
Camera fixed, professional studio lighting, shallow depth of field.

Shot 2 (5-10s) - Device Reveal:

Box opens completely, revealing pristine smartphone with screen protector,
hand slowly lifts the phone out of the box, rotating to show all sides.
Gentle overhead lighting highlights the device's curves and edges.

Shot 3 (5-10s) - Close-up Detail:

Hand holds the phone at eye level, camera pushes in slowly to extreme close-up.
Sunlight reflects off the glossy back, showing the camera module and branding.
Focus remains sharp on device while background blurs.

Shot 4 (15-20s) - Screen Power:

Thumb presses the power button, screen illuminates with startup animation.
Vibrant display shows colorful welcome screen, user's face reflected in glass.
Soft glow from phone illuminates the hand holding it.

Shot 5 (20-25s) - Interface:

Finger swipes across screen navigating through home screen,
app icons are visible and crisp, icons glow slightly with depth.
Same hand and lighting as previous shot, continuous action.

Shot 6 (25-30s) - Confident Hold:

Phone held confidently in both hands in portrait orientation,
user looks at screen with satisfied expression, professional photography lighting,
slow camera pulls back revealing the full device and hands.

Generation Parameters:

• Resolution: 1280x720
• Sampling Steps: 20 (with Distiller LoRA)
• Motion Amplitude: 1.2
• Motion Frames: 7
• Overlap Frames: 13

Expected Result: Seamless 30-second product demo with natural motion, consistent lighting, and professional appearance.

Comparing: PainterLongVideo vs Traditional Approaches

Traditional Approach (Frame-by-Frame)

Process:

1. Generate Shot 1 (5 seconds)
2. Manually trim last frame
3. Use last frame as reference for Shot 2
4. Repeat for each additional shot
5. Manually stitch in video editor

Problems:

• Visible seams between shots
• Color/lighting mismatch at cut points
• Character appearance inconsistencies
• Slow-motion issues accumulate
• Very time-consuming workflow

PainterLongVideo Approach

Process:

1. Set up 6-shot workflow once
2. Adjust motion_amplitude for natural speed
3. Write 6 continuous prompts
4. Hit "Generate" and wait
5. Download seamless result

Advantages:

• Zero visible transitions
• Perfect motion continuity
• Character consistency maintained
• Natural motion throughout
• Single workflow handles everything

Time Comparison:

• Traditional: 60-90 minutes manual work + generation
• PainterLongVideo: 15 minutes setup + generation

Best Practices for Professional Results

1. Pre-Generate Test Shots

Before committing to a full 30-second generation:

1. Create 1-2 shot test workflow
2. Test your prompts at 854x480 resolution
3. Verify character/motion/colors work
4. Use motion_amplitude values to find perfect speed
5. Only then scale up to full production

Time Saved: Avoid 30-60 minute wasted generations

2. Prompt Engineering Framework

Use this structure for each shot's prompt:

[SUBJECT AND ACTION] doing [SPECIFIC ACTION],
in [DETAILED ENVIRONMENT],
[CAMERA MOVEMENT] camera,
[LIGHTING DESCRIPTION],
[MOOD AND STYLE],
[TECHNICAL SETTINGS FOR QUALITY]

Example:

Professional woman presenting quarterly earnings in modern conference room,
discussing metrics on large screen behind her, discussing metrics on large screen behind her,
steady camera angle with slight zoom,
bright overhead and side lighting from professional fixtures,
corporate documentary style, crisp and clear,
4K quality with sharp focus on presenter

3. Lighting Consistency

Critical for seamless shots:

• Describe lighting in every prompt
• Use consistent lighting direction (e.g., "light from left" throughout)
• Match color temperature ("warm afternoon sunlight" vs "cool fluorescent")
• Maintain shadow direction across shots
• Include ambient light sources if visible

4. Character Consistency Tips

To keep characters identical across shots:

• Include detailed character description in first prompt only
• Reference appearance in subsequent prompts ("same woman as before")
• Keep outfit description identical
• Maintain same makeup/hair styling throughout
• Use high-quality anchor image

5. Motion Continuity Checklist

Before generating full sequence:

• Shot prompts tell continuous story
• Motion type consistent (no sudden direction changes)
• Camera movement planned (not random direction swaps)
• Lighting stays consistent throughout
• Character position/direction flows logically
• No contradictory actions between shots

Post-Production Enhancement

Recommended Tools

Free Options:

• DaVinci Resolve - Professional color grading and effects
• FFmpeg - Command-line video processing
• CapCut - User-friendly editing and effects

Suggested Enhancements:

1. Color grading pass for consistency
2. Audio track with background music (-20dB volume)
3. Smooth fade in/out at beginning/end
4. Subtle sharpening filter
5. Noise reduction if needed

Audio Integration

Best practices for adding audio:

# Combine video with audio using FFmpeg
ffmpeg -i video.mp4 -i audio.mp3 -c:v copy -c:a aac -shortest output.mp4

• Keep background music subtle (-20dB to -25dB)
• Center voice/dialogue at -5dB to -10dB
• Add 100ms fade-in/out to avoid pops
• Use royalty-free music (Epidemic Sound, Artlist, etc.)

Frequently Asked Questions

Q: Can I generate longer than 30 seconds? A: Yes. Copy shot groups and extend indefinitely. Most users limit to 10-12 shots (~60 seconds) due to face drift beyond that.

Q: What's the best motion_amplitude value? A: Start at 1.2 (natural speed). Adjust based on your footage - slower content uses 1.15, dynamic scenes use 1.25-1.3.

Q: How do I prevent character face changing? A: Ensure anchor latent connects to all PainterLongVideo nodes. Include character description in prompts. Use high-quality reference image.

Q: Can I use different models in the same workflow? A: No. All shots must use the same model for consistency. Mixing models causes quality/style mismatches.

Q: What if I only have 8GB VRAM? A: Use FP8 model + Distiller LoRA + 640x360 resolution. Reduce steps to 10. Use RunningHub cloud option for better results.

Q: How long does a 30-second video take to generate? A: RTX 4090: 15-20 minutes. RTX 4080: 25-35 minutes. RTX 4070 Ti: 35-45 minutes. With Distiller LoRA: 50% faster.

Related Resources

1. LongCat Video Complete Guide - Master the fundamentals of LongCat video generation
2. LongCat Video Extra Long Generation - Create extended videos with concatenation techniques
3. WAN 2.1 Installation Guide - Complete setup guide for WAN models
4. Stable Video Infinity WAN 2.1 Tutorial - Advanced long-form video generation techniques