CoolFace
Modelpublic

rzgar/LTX-2.3-Motion-Enhancer-n4w

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
8likes2.8kdownloads
Model Card

LTX-2.3 Motion Enhancer (N54W / General Purpose)

<video controls width="100%" height="80%"> <source src="https://huggingface.co/rzgar/LTX-2.3-Motion-Enhancer-n4w/resolve/main/demo/ComfyUI_00011-audio.mp4" type="video/mp4"> Your browser does not support the video tag. </video>

<Gallery />

Model description

📖 About

Similar to &quot;Bernini-R Motion Enhancer I2V&quot;, this LoRA has been specifically tuned to serve as a general-purpose N54W motion enhancer.

It understands the majority of explicit and complex prompts without being overly specialized in any specific domain, action, or animation style. Because of its generalized nature, it is highly recommended to use this as a companion LoRA alongside other specialized models. When stacked, it significantly enhances the fluidity, coherence, and motion of domain-specific LoRAs.

It has been extensively tested and successfully paired with the majority of the top-rated and most downloaded LoRAs hosted on Civitai. ___

Files

FileDirect Link
LTX-2.3-Motion-Enhancer-n4w.safetensorsDownload

⚙️ Usage &amp; Recommended Strengths

  • —Stacked with other LoRAs: 0.65 (Recommended)
  • —Standalon:0.75 to 1.0

📊 Performance &amp; Error Rate

(Testing methodology: Based on 15 unique prompts, 3 runs per prompt, random seeds, all Image-to-Video )

Scenario 1: Portrait Input &#x2F; Complex Prompting

  • —Input: Portrait.
  • —Prompt: 90% of the events&#x2F;objects described are not present in the input image.
  • —Result: Error rates vary heavily depending on prompt quality, and a second iteration (or re-roll) is often needed. Mild body deformities, missing fingers, or broken animations occur more frequently if the input image lacks the necessary visual context for the base model to work with.

Scenario 2: Detailed Full-Body Input &#x2F; Aligned Prompting

  • —Input: Detailed, full-body shot where all persons and objects mentioned in the prompt are already visible in the image.
  • —Result: Error rates decrease significantly. About 3 out of 5 generations are highly usable on the first run, else only slight or mild animation artifacts.

💡 Key

While a good input image provides the foundation, your text prompt describing the action has a much higher impact on the final generation quality. Detailed motion prompting is highly recommended!

Download model

Download them in the Files & versions tab.