Editor's Note: This tutorial teaches you how to create precisely controlled animations using LTX 2.3's First-Middle-Last frame workflow in ComfyUI.

LTX 2.3 First Middle Last Frame Video Workflow in ComfyUI

Introduction

LTX 2.3's First-Middle-Last (FML) frame workflow gives you precise control over video animation by defining three keyframes: where your animation starts, a midpoint pose, and where it ends. The model generates smooth motion between these anchor points.

System Requirements

• GPU: 16GB+ VRAM (FP8 workflow)
• ComfyUI with LTX 2.3 nodes installed
• LTX 2.3 model files (FP8 quantized recommended for 16GB)

Step 1: Prepare Your Anchor Frame Images

Start by preparing three keyframe images:

• First Frame: Your opening image
• Middle Frame: The pose or moment at the midpoint
• Last Frame: Where your animation ends

💡 Consistency Tips:

• Keep the same overall composition across all frames
• Maintain consistent subject scale and lighting direction
• Avoid large jumps in framing or perspective
• The model works best when motion between keyframes is believable in a single continuous shot

Pro Tip: Use the Qwen Multi Angle Workflow to generate consistent multi-pose images ready for this pipeline.

Step 2: Write Your Animation Prompt

Your text prompt steers the motion style, atmosphere, and character of the generated footage. Since the model knows the start and end points, focus on the quality of the transition.

Example Prompt for 360° Rotation:

Uninterrupted 360° counterclockwise rotation of character in elegant gown. Begins relaxed frontal pose, smooth graceful full turn at even speed, displays details from all sides. Perfect consistency, realistic motion, subtle movement. Clean background, soft lighting creating gentle glow. Ultra-detailed 24fps smooth shallow DoF.

Key Guidelines:

• Keep camera description static
• Avoid prompting sudden or large movements
• Focus on motion smoothness, lighting consistency, and detail quality
• Use negative prompts: "jump cut, teleport, morph, dissolve, glitch, stutter, duplicate face, warped limbs"

Step 3: Configure Video Settings

Set your output parameters:

• Width & Height: Must be divisible by 32 (e.g., 736×1280 for 9:16)
• Frame Count: Must be divisible by 8, plus 1
• FPS: 24 fps recommended
• Length: Set in seconds

Common Resolutions:

• 720p Portrait (9:16): 736×1280
• 1080p Portrait: 1088×1920 (requires RTX 5090 or better)
• Landscape: 1280×736

⚠️ Important: Invalid parameters will be silently rounded to valid values.

Step 4: Run the Generation

Once configured:

1. Click RUN in ComfyUI
2. LTX 2.3 runs two sampling passes:
   • Fast distilled pass
   • Refinement pass
3. Decodes video latents
4. Applies spatial upscaler
5. Renders final output

The result is a continuous video clip that smoothly transitions through all three keyframes.

What Makes This Workflow Powerful

Creative Control: Unlike standard image-to-video that generates organic motion, FML workflow lets you compose your animation. You define the start, turning point, and destination.

Use Cases:

• Product showcases ending on specific hero shots
• Character animations hitting defined poses
• Environmental transitions (day to dusk)
• 360° character rotations
• Directed motion sequences

Conclusion

The First-Middle-Last frame workflow hands creative control back to you. If you can supply the keyframes, LTX 2.3 can connect them with smooth, coherent motion.

Now pick your anchor images, write your prompt, and create precisely directed animations!

Sources

• LTX 2.3 FML Workflow Tutorial
• LTX 2.3 Official Documentation