LTX 2.3 Video Generation Showcase: What's Possible in 2026

Editor's Note: This showcase highlights what's achievable with LTX 2.3 in ComfyUI, based on community results and official demos. The new official FP8 models and spatial upscalers open up high-quality video generation on consumer GPUs.

LTX 2.3 Video Generation Showcase

LTX 2.3 represents a significant leap in open-source video generation. With 22 billion parameters and a distilled variant that runs in just 8 steps, it produces cinematic-quality videos that were previously only possible with proprietary models.

What Makes LTX 2.3 Stand Out

Speed: The distilled model generates 5-second videos in under 30 seconds on a 32GB GPU. With FP8 quantization, 16GB VRAM users can achieve similar results.

Quality: Motion coherence, temporal consistency, and prompt adherence are all dramatically improved over previous versions.

Flexibility: The dev model supports LoRA fine-tuning, enabling custom styles and subjects.

Showcase Examples

Text-to-Video (T2V)

The distilled model excels at T2V generation. Community members report impressive results with:

• Cinematic scenes: "A lone astronaut walks across a red desert at sunset, dust swirling around their boots" — smooth camera movement, realistic lighting
• Nature footage: "Ocean waves crashing against rocky cliffs in slow motion" — detailed water simulation, consistent motion
• Abstract visuals: "Colorful paint drops falling into water in macro photography" — precise detail, vivid colors

Settings: 8 steps, CFG=1, 768×512 resolution, 25 frames

Image-to-Video (I2V)

LTX 2.3 brings static images to life with natural motion:

• Portrait photos with subtle breathing and eye movement
• Landscape photos with moving clouds and water
• Product shots with gentle rotation

Two-Stage Pipeline with Spatial Upscaler

The new spatial upscaler x2 enables a powerful two-stage workflow:

1. Generate at 512×320 (fast, low VRAM)
2. Upscale to 1024×640 with the spatial upscaler

Result: Full HD quality video at a fraction of the VRAM cost.

Hardware Results

Getting Started

1. Download ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors (16GB) or ltx-2.3-22b-distilled-1.1.safetensors (32GB)
2. Download taeltx2_3.safetensors (VAE, required)
3. Use the workflow templates at ltxworkflow.com/workflows

Sources

• Lightricks/LTX-2.3 on HuggingFace
• ComfyUI-LTXVideo GitHub