Pushing LTX 2.3 I2V: Moving gears, leg pistons, and glossy porcelain reflections (ComfyUI / RTX 4090)
The AI model creates 5-second clips of robotic automatons with rigid gears and glossy porcelain textures in ~200 seconds.
A detailed test of the LTX 2.3 (ltx-2.3-22b-dev) Image-to-Video model demonstrates a significant leap in generating precise, non-organic animations. Running in ComfyUI on high-end hardware (NVIDIA RTX 4090, AMD Ryzen 9 9950X), the model was tasked with animating base images—created with FLUX1-dev and a custom LoRA stack—featuring intricate robotic automatons. The key achievement was the model's adherence to "strictly mechanical movement," producing rigid actuation of leg pistons and precise twitches in transparent wings, completely avoiding the organic, chaotic motion that plagues many video AI systems.
Beyond motion, the LTX 2.3 model excelled at complex material rendering. It accurately calculated and maintained dynamic lighting reflections on glossy porcelain surfaces as internal gold gears turned, preserving crisp contrasts between translucent wings, white ceramic, and metallic components without color bleeding. The test output, consisting of six native 1080p vertical clips, took approximately 200 seconds each to render and even included procedurally generated mechanical ASMR audio. This performance indicates the model's advanced understanding of 3D geometry and physics, moving beyond simple texture application to simulate how light interacts with moving, reflective objects in a coherent scene.
- Generated six 5-second, 1088x1920 video clips in about 200 seconds each using an NVIDIA RTX 4090 GPU.
- Achieved precise, rigid mechanical animation (gears, pistons) without devolving into organic 'melting' motion common in other models.
- Accurately rendered dynamic lighting and reflections on complex materials like glossy porcelain and metal, maintaining scene integrity.
Why It Matters
This shows AI video generation is advancing beyond organic shapes to reliably create complex mechanical and material-specific animations for design and prototyping.