ComfyUI LTX-2 Workflow Pack: Best Nodes, Templates, and Settings (Download + Fixes)

2026/01/09

What You Will Get

Stop wasting time connecting random nodes. In this guide, we provide 3 battle-tested workflows for LTX-2:

  1. Photorealistic: For lifelike human motions and physical accuracy.
  2. Anime/Stylized: Optimized for 2D art consistency.
  3. Product Showcase: Smooth camera movements for cinematic advertising.

Compatibility Check:

  • ComfyUI Version: Latest (Updated after Jan 2026 recommended)
  • LTX-2 Weights: ltx-2-19b-distilled.safetensors (Recommended)

Part 1: Choose Your Workflow Template

Before you start, identify your goal. Different goals require different Guidance and Motion settings.

1. The "Realism" Template (Stability First)

Best for: Human subjects, nature documentaries, and footage that needs to obey physics.

  • Key Focus: Reducing "hallucinations" (morphing limbs).
  • Trade-off: Slightly less creative camera movement.

2. The "Anime" Template (Style First)

Best for: 2D animation, cel-shaded looks, and artistic experimentation.

  • Key Focus: Color saturation and line consistency.
  • Trade-off: May require higher steps to clean up artifacts.

3. The "Product Camera" Template (Motion First)

Best for: Perfume bottles, tech gadgets, and food commercials.

  • Key Focus: Smooth "Push-in" or "Orbit" camera moves.
  • Trade-off: Requires a static subject background.

Template A: The Stable Realism Workflow (For Beginners)

This is the "Hello World" of LTX-2. If you are new, start here.
My Workflow
(Download the JSON file at the bottom of the page)

Node Logic

  1. CheckpointLoaderSimple: Loads ltx-2-19b-distilled.
  2. EmptyLatentVideo: Sets the resolution (768x512 is the sweet spot).
  3. CLIPTextEncode: Uses "Positive" and "Negative" prompts.
  4. KSampler: The brain of the operation.
  • Resolution: 768x512 (Widescreen) or 512x768 (Mobile). Avoid 1024x1024 on consumer GPUs.
  • FPS: 24 (Standard Cinematic).
  • Steps: 30–35 (Going higher has diminishing returns).
  • Guidance (CFG): 3.0 – 4.0.
    • Too Low (<2.5): Blurry, ignores prompt.
    • Too High (>5.0): High contrast, "fried" look.
  • Seed: Fixed (for tuning) or Random (for exploration).

Troubleshooting Physics

  • Problem: Subject is drifting or warping?
  • Fix: Lower Motion Strength to 0.8 or increase Steps by +5.

Template B: Stylized & Anime (Consistency)

Anime needs strong lines and vibrant colors. LTX-2 can sometimes make anime look "washed out" or "3D-rendered" if not tuned correctly.

The "Style Block" Prompting

Connect a separate Style Node or add this to the start of your prompt:

"Anime style, cel shaded, flat color, high contrast, [Your Subject]"

Reducing "Jitter"

Anime lines tend to flicker. To fix this:

  1. Fix the Seed: Don't change the seed between tests.
  2. Use a "Smoother" Scheduler: Try sgm_uniform or simple.

Template C: Cinematic Product Looping

Want that Apple-style product reveal? This template focuses on Camera Control.

The Camera Dictionary

Use these keywords in your prompt for specific movements (LTX-2 understands these natively):

  • "Orbit right" / "Orbit left": Rotate around the object.
  • "Push in": Zoom slowly into the product.
  • "Dolly out": Pull back to reveal the scene.

How to Make it Look "Premium"

  1. Lighting: Add "studio lighting, rim light" to your text prompt.
  2. Materials: Specify "glass, messy texture, matte finish" to help the AI understand surface reflection.
  3. Slow Motion: Generate at 12 FPS and play back at 24 FPS for a 2x slow-mo effect.

Parameter Dictionary: The "Big 8"

You don't need to know every knob. Just master these 8 settings.

Parameter Meaning Recommended Range
Resolution Sharpness vs. VRAM 768x512 (Best), 1280x720 (High VRAM)
FPS Smoothness 24 (Standard)
Frame Count Duration (frames/FPS = seconds) 49 or 97 (Odd numbers often loop better)
Steps Detail level 30 - 50
Guidance How strictly it follows prompt 3.0 - 4.5
Seed Random noise pattern Keep fixed when debugging
Motion Scale How much movement occurs 1.0 (Normal), 1.2 (High Action)
Denoise Creativity vs. Input 1.0 (Text-to-Video), 0.6 (Image-to-Video)

Common Errors & Fixes

1. Red Nodes / "Missing Node" Error

  • Symptom: You load the JSON and boxes turn red.
  • Fix: Open ComfyUI Manager -> Click "Install Missing Custom Nodes". Restart ComfyUI.

2. Black Screen / Green Noise Output

  • Symptom: The video finishes but is solid black or static.
  • Fix: You are likely using bfloat16 on a GPU that doesn't support it (like older GTX cards). Force fp16 or fp32 in your command line arguments: --force-fp16.

3. OOM (Out of Memory)

  • Symptom: "CUDA out of memory".
  • Fix:
    1. Close Chrome/Photoshop.
    2. Lower output resolution.
    3. Add --lowvram to your ComfyUI bat/sh file.

When NOT to Use ComfyUI?

ComfyUI is powerful but messy. Sometimes, it's better to verify your idea first before building nodes.

  • Scenario A: You need to show a client 3 concepts now.
    👉 Use the Online Generator.
  • Scenario B: You want to integrate this into your own website.
    👉 Use our API.

FAQ

Q: I copied the workflow but the video flickers.
A: Check your CFG/Guidance. If it's too high (>5), LTX-2 tries too hard to "draw" every frame, causing flicker. Lower it to 3.5.

Q: How do I keep the same character consistent?
A: Use a specific seed and consider using "IPAdapter" nodes (coming soon for LTX-2) or very specific character descriptive tags (e.g., "blue tie, scar on left cheek").

Q: Which parameter saves the most VRAM?
A: Resolution. Dropping from 1024px to 768px saves ~40% VRAM memory.


Are you ready to move beyond downloading?

Alex Carter