ltx-2-video

What it does

Guides an AI agent to generate and edit synchronized audio-video using LTX-2.3, either via the official hosted API or a self-hosted local repo. It covers use cases like text-to-video, image-to-video, audio-to-video, retake, extend, HDR conversion, local inference, checkpoint selection, quantization, and LTX-specific production planning.

How it works

• Start with a plan: specify mode, duration, aspect ratio, resolution, frame rate, audio requirement, delivery codec, and whether inputs include people or sensitive data.
• Choose host: route between hosted or local via a routing table; avoid assumptions about cost or ease.
• Dry-run plan: before any paid request, asset upload, or model download, write a dry-run plan that lists exact endpoint/model or checkpoint filenames, estimated hosted charge, output paths, and validation steps.
• Approval gate: require explicit approval before first billable API submission or multi-gigabyte download; no auto top-up.
• Preserve provenance: capture prompt, seed, model/checkpoint, API request/job ID, source asset hashes, generation parameters, license review date, and output hash.
• Route selection: follow the routing table between hosted T2V/I2V, hosted A2V, hosted retake, hosted extend, hosted HDR, official local repo, or Node-based local workflow, choosing appropriate endpoints such as POST /v2/text-to-video, POST /v2/image-to-video, POST /v2/audio-to-video, POST /v2/retake, POST /v2/extend, or POST /v2/video-to-video-hdr depending on the task.
• Pricing and constraints: adhere to the hosted pricing and constraints by surface and resolution, and handle async vs sync behavior as documented.
• Input handling and validation: follow upload guidance, preserve input constraints, and validate outputs by URLs and hashes; download promptly and hash locally.
• Prompts and runs: write prompts for joint audio-video per guidelines and use a hosted dry-run client workflow for planning and publishing jobs.

When to use it

Use when a workflow needs jointly generated video and audio, first/last-frame conditioning, audio-driven video, partial retakes, extension, or an official self-hosted path. Prefer async V2 for production; sync V1 endpoints yield MP4 in a single request and are suitable only for short experiments.

What it can touch

• Endpoint interactions via hosted API: POST /v2/text-to-video, POST /v2/image-to-video, POST /v2/audio-to-video, POST /v2/retake, POST /v2/extend, POST /v2/video-to-video-hdr.
• Pricing constants and routing tables are included for decision logic in the client code.
• It references the hosted API base: https://api.ltx.video and requires Authorization: Bearer $LTXV_API_KEY.

Caveats

• Async operations require persisting the returned id and polling GET /v2/{operation}/{id} with allowed states; output URLs expire 24 hours after terminal state. HDR uses result.exr_frames_url.
• Old IDs (ltx-2-fast, ltx-2-pro) are deprecated and must be rejected in new work after 2026-07-15.
• Pricing is subject to change; re-fetch before production use.
• A due diligence process is required: explicit approval before billable actions and no auto top-ups.
• Data handling notes emphasize preserving provenance and compliance considerations for sensitive content.