# LTX-23-Video

> Lightricks LTX 2.3 distilled-1.1 FP8 on RTX 6000 Pro — cinematic T2V, I2V, and keyframe interpolation with 8-step inference, synchronized audio, and GPU prompt enhancement.

This file is intended for AI coding agents wiring an application to this specific Chutes model.

## Important URLs

- Model page: https://chutes.ai/app/chute/vonkaiser-ltx-23-video
- Model llms.txt: https://chutes.ai/app/chute/vonkaiser-ltx-23-video/llms.txt
- Model OpenAPI 3.1 spec (this chute, callable): https://chutes.ai/app/chute/vonkaiser-ltx-23-video/openapi.json
- Global Chutes llms.txt: https://chutes.ai/llms.txt
- Full Chutes docs export: https://chutes.ai/llms-full.txt
- Management API OpenAPI (account/billing/keys, NOT this model): https://api.chutes.ai/openapi.json

## Model Identifiers

- Name: `LTX-23-Video`
- Chute ID: `77711d2c-c0ec-5378-90a9-41823591058b`
- Slug: `vonkaiser-ltx-23-video`
- Owner: `vonkaiser`

## Authentication

- Use `Authorization: Bearer $CHUTES_API_KEY` for inference calls.
- Send JSON request bodies with `Content-Type: application/json` unless the endpoint documentation says otherwise.
- The request body is FLAT — send the request fields at the top level (no `input_args`/`args` wrapper).
- This model is served on its own host (`https://vonkaiser-ltx-23-video.chutes.ai`). Call the endpoint paths listed below directly.
- For base64 media fields, send raw base64 strings in API requests. The web playground may use temporary Blob upload references internally for large files before proxying to Chutes.

## Endpoints

### 1. POST /generate

- Base URL: `https://vonkaiser-ltx-23-video.chutes.ai`
- Output content type: `text/plain`

Request fields:

- `fps` (number, optional) - Output frame rate (8–50). Default 24. Default: null.
- `seed` (integer, optional) Default: null.
- `width` (integer, optional) Default: 768.
- `frames` (integer, optional) - Alias for num frames. Default: null.
- `height` (integer, optional) Default: 512.
- `prompt` (string, required)
- `duration` (number, optional) - Target clip length in seconds. With default 24 fps, derives frame count. If frames are set but fps omitted, fps is derived from duration for the output MP4. Default: null.
- `keyframes` (array, optional) - List of 2+ keyframes for multi-image interpolation. Each entry: {image b64, frame idx, strength?, crf?}. At least one keyframe must have frame idx=0. Takes precedence over input image b64 when provided. Default: null.
- `num_frames` (integer, optional) - Frame count (snapped to N×8+1). Omit to derive from duration×fps. Default: null.
- `return_json` (boolean, optional) Default: false.
- `video_format` (string, optional) Default: "mp4". Options: mp4, webm.
- `enhance_prompt` (boolean, optional) - Expand the prompt before generation with Llama 3.2 3B Instruct on GPU (SGLang); for I2V/keyframe also runs Florence-2 image captioning on the first image. Off by default — adds 1–3 s of latency. Default: false.
- `generate_audio` (boolean, optional) Default: true.
- `guidance_scale` (number, optional) Default: 3.1.
- `input_image_b64` (string, optional) - Base64 image for legacy single-image I2V (PNG/JPEG). Use keyframes instead for multi-image interpolation. Default: null.
- `negative_prompt` (string, optional) Default: "shaky, glitchy, low quality, worst quality, deformed, distorted".
- `use_random_seed` (boolean, optional) Default: true.
- `num_inference_steps` (integer, optional) - Ignored — always 8 for the pre-distilled model. Any value provided is silently overridden to 8. Default: 8.

Example call:

```bash
curl -X POST "https://vonkaiser-ltx-23-video.chutes.ai/generate" \
  -H "Authorization: Bearer $CHUTES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "seed": 42,
  "width": 768,
  "height": 512,
  "prompt": "a serene mountain lake at sunset",
  "return_json": false,
  "video_format": "mp4",
  "enhance_prompt": false,
  "generate_audio": true,
  "guidance_scale": 3.1,
  "negative_prompt": "shaky, glitchy, low quality, worst quality, deformed, distorted",
  "use_random_seed": true,
  "num_inference_steps": 8
}' \
  --output output.bin
```

### 2. GET /health

- Base URL: `https://vonkaiser-ltx-23-video.chutes.ai`
- Output content type: `application/json`

Request fields:


Example call:

```bash
curl -X GET "https://vonkaiser-ltx-23-video.chutes.ai/health" \
  -H "Authorization: Bearer $CHUTES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{}'
```

### 3. GET /models

- Base URL: `https://vonkaiser-ltx-23-video.chutes.ai`
- Output content type: `application/json`

Request fields:


Example call:

```bash
curl -X GET "https://vonkaiser-ltx-23-video.chutes.ai/models" \
  -H "Authorization: Bearer $CHUTES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{}'
```

## Model Facts

- Parameters: 22B (per Lightricks checkpoint naming, ltx-2.3-22b)
- Architecture: DiT-based joint audio-video foundation model; this deployment runs the distilled-1.1 checkpoint (8 steps, CFG=1) with an x2 spatial upscaler and a Gemma text encoder
- Modalities: text, image in → video, audio out
- Precision: FP8 transformer + BF16 VAE/text-projection/upscaler (as served on Chutes)
- License: LTX-2 Community License Agreement
- Released: 2026-03
- Modes: Text-to-video, image-to-video, multi-keyframe interpolation
- Synchronized audio: Yes, generated jointly with video (generate_audio, default true)
- Inference steps: Fixed at 8 (distilled); any requested value is overridden
- Max resolution: Up to 3840x2160; width/height snapped to multiples of 64 on this chute
- Max duration: ~20 s at 480p-1080p, ~10 s at 1440p, ~5 s at 4K width
- Frame rate: 8-50 fps, default 24; frame count snapped to Nx8+1
- Prompt enhancement: Optional on-GPU Llama 3.2 3B rewrite (+Florence-2 captioning for I2V), off by default
- Hugging Face: https://huggingface.co/Lightricks/LTX-2.3

## Recommended Parameters

- `guidance_scale`: 3.1 (chute default; chute readme documents range 1.0-5.1)
- `num_inference_steps`: 8 (fixed for the distilled-1.1 checkpoint; the endpoint silently overrides other values)
- `width/height`: multiples of 64 (the chute snaps dimensions (e.g. 1920x1080 becomes 1920x1088); upstream card requires divisibility by 32)

## Best For

- Cinematic text-to-video clips with synchronized sound in one call
- Image-to-video: animating a single frame with matching audio
- Multi-keyframe interpolation: smooth transitions through 2+ anchor images
- Short high-resolution renders up to 4K width
- Fast iteration: 8-step distilled inference keeps latency low for a 22B video model

Not ideal for:

- Clips longer than ~20 seconds (hard duration caps that shrink with resolution)
- Fine step-count or CFG experimentation: steps are fixed at 8 on this deployment
- Factual or text-heavy content; like all video diffusion models it does not render reliable information

## FAQ

### Does LTX 2.3 really generate audio with the video?

Yes. LTX-2 is a joint audio-visual foundation model, and this chute has generate_audio: true by default, so the returned MP4 includes a synchronized soundtrack. Lightricks notes audio without speech can be lower quality. Set generate_audio: false for silent output.

### Can I change the number of inference steps or use CFG like the full model?

No. This deployment runs the distilled-1.1 checkpoint, which is trained for exactly 8 steps at CFG=1; the endpoint silently overrides any num_inference_steps you send. You can still tune guidance_scale (default 3.1, range 1.0-5.1 per the chute readme).

### How does keyframe interpolation work?

Pass a keyframes array of 2 or more entries, each with a base64 image and a frame index (optionally strength and crf); at least one must have frame index 0. The model generates coherent motion through the anchors. keyframes takes precedence over input_image_b64.

### What resolutions and durations are supported?

Width and height snap to multiples of 64, from 256x256 up to 3840x2160. Duration caps scale with resolution: roughly 20 seconds at 480p-1080p, 10 seconds at 1440p, and 5 seconds at 4K width. Frame counts snap to Nx8+1 and fps runs 8-50 (default 24).

### How do I call the API?

POST a flat JSON body to https://vonkaiser-ltx-23-video.chutes.ai/generate with Authorization: Bearer $CHUTES_API_KEY. Only prompt is required. The response is raw video bytes (mp4 by default, webm optional); set return_json: true to get JSON instead.

### Can I use the output commercially?

The weights are under the LTX-2 Community License Agreement, not a standard OSI license. Review the license at github.com/Lightricks/LTX-2/blob/main/LICENSE for the permitted-use terms that apply to your case.

### What does enhance_prompt do and should I use it?

It expands your prompt on-GPU with Llama 3.2 3B Instruct before generation, and for I2V/keyframe jobs also captions the first image with Florence-2 so the rewrite matches your input. It is off by default and adds 1-3 seconds. Use it for terse prompts; skip it when you have already written a detailed prompt.

### Why did my 1920x1080 request come back as 1920x1088?

The chute snaps dimensions to multiples of 64, so 1080 rounds to 1088. Either accept the snap or request dimensions that are already multiples of 64.


## Model Guide & Sources

- Full model guide: https://chutes.ai/docs/models/vonkaiser-ltx-23-video
- Source: https://huggingface.co/Lightricks/LTX-2.3
- Source: https://huggingface.co/Kijai/LTX2.3_comfy
- Source: https://huggingface.co/Lightricks/LTX-2
- Source: https://chutes.ai/app/chute/vonkaiser-ltx-23-video/llms.txt

## Agent Integration Checklist

- Pick the endpoint path that matches the desired task.
- Set `CHUTES_API_KEY` in the server-side environment only.
- Validate required fields before sending requests.
- For media models, keep file upload, base64 conversion, and output preview/download handling explicit in the app UI.