Amir Mushich builds product rotations with the LTX-2.5 API

• Real product photos became rotating previews through LTX-2.5 in AmirMushich's demo, with the interface coded using Astra-6 in Codex.
• Each preview links to a real brand’s store, as he explained in a follow-up.
• Generation took roughly 10-12 seconds, according to his recollection.

The concept came from @cambreedesigns, whom Mushich credited for the inspiration. LTX’s model reference also exposes final-frame control and a switch for silent video.

Amir Mushich described a three-step build, a tidy bit of storefront hacking:

1. Photography: Real product photos supplied the source material.
2. Motion: LTX-2.5 generated a rotation for each photo through the LTX API.
3. Interface: Astra-6 in Codex handled the UI code.

The product-rotation demo

LTX’s image-to-video documentation describes an asynchronous route for generating these kinds of assets:

• Submission: POST /v2/image-to-video accepts a reference image and a prompt describing the desired motion. The image becomes the first frame.
• Status: The response contains a job ID. GET /v2/image-to-video/{id} reports progress until the status is completed.
• Output: The finished clip is available through result.video_url.

The LTX-2.5 model reference distinguishes ltx-2-5-fast, designed for speed and low cost, from ltx-2-5-pro, designed for higher fidelity.

Mushich’s 10-12-second estimate came from memory; he offered to recheck his logs for a precise figure.

The shopping link leads to Molimao, a Sarajevo clothing storefront selling products such as the Grow oversized T-shirt and TAG tote bag.

Mushich also floated landing pages for new collections, saying the concept could work well for click-through rates.

Mushich has also teased a Blender experiment, naming “GPT-6.1 Sol” in a separate post.

In the replies, youraipulse suggested Blender MCP plus LTX and said they were testing it.

Mushich said he planned to publish a full retrospective within five to seven days, with preparation still under way. In a Korean-language reply, he also said he intended to open-source a detailed implementation breakdown on GitHub.