whatmodel.sh / video / LTX-2.3
LTX-2.3
↗ share
+ contribute
LTX-2.3 is an open-weights video model from Lightricks, released March 2026. Its published size is 22B parameters. Licence: ltx-2-community.
Model information sources
· Weights licence ↗ · Parameter count ↗
Maker's own site. Consistent with checkpoint names on the official card (https://huggingface.co/Lightricks/LTX-2.3/blob/3c6a4e66e5d0a684231950b9c74dd4ded7b6fadc/README.md): 'ltx-2.3-22b-dev', 'ltx-2.3-22b-distilled'. Covers the audio-video DiT (text encoder separate).
$ how it compares
The video models ranked either side of LTX-2.3, out of 139 with a score, then more from Lightricks.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 110 |  Runway Gen-4 Turbo | 37±3 | api $0.050/sec | ||
| 111 |  MAGI-1 4.5B | 37±5 | comparison only | ||
| 112 | LTX-2.3 | 36±3 | api $0.040/sec +local | ||
| 113 |  CogVideoX1.5-5B | 36±2 | local | ||
| 114 | CogVideoX-2B | 36±2 | local | ||
| more from Lightricks | |||||
| 53 | LTX-2.5 | 53±6 | api $0.12/sec +local | ||
| 70 | LTX-2 | 49±2 | local | ||
| 83 | LTX-2 19B Distilled | 47±8 | api $0.020/sec +local | ||
$ what it's good at
text to video▆▄▃▄▆▅█▃▂▂▂▄ #47 of 68 * · 7.0% of its score · 1 test
Beats Mochi 1, HunyuanVideo, T2V-Turbo (VC2) here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Sci-VBench testmini (expert human rating) | 38.4% ✎ | ▅·▃█▃·▅▃▅▃▃▃ beats 0/14 | 7.0% |
image to video▁▃·▂▅█▅▄▅▆▃▄ #48 of 50 * · 76% of its score · 1 test
Beats Runway Gen-4 Turbo here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| WorldReasonBench human Elo (image+text-to-video world-state prediction) | 587.0 ✎ | ▃▃··█··▃▃▃·▃ beats 0/8 | 76% |
other tests2 not in a task · 14% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MindWorldBench intention accuracy (I2V mental-state reasoning) | 11.9% ✎ | ▅·▃▅·█·▃···▃ beats 0/9 | 8.1% |
| PhyGround physics human evaluation (Overall, TI2V) | 53.8% ✎ | █▄▄··▄···▄·█ beats 3/7 | 6.2% |
$ where to run it
Through an API
api · Lightricks (LTX API) · $0.040/sec · docs.ltx.io ↗ · price sourceUSD per second of generated video, LTX API text-to-video, LTX-2.3 pro model at 1280x720. Product: LTX API v2/text-to-video. API model: ltx-2-3-pro. Settings: resolution=1280x720, tier=pro (same tier as the ltx-2-5 row). Price basis: published cost per second of output video. ltx-2-3-fast $0.03/s at 720p; pro 1080p $0.08/s, 1440p $0.16/s, 2160p $0.32/s (image-to-video the same; audio-to-video pro 720p $0.06/s). fal's partner endpoint fal-ai/ltx-2.3/text-to-video starts at 1080p ($0.08/s) and is not used. Currency: USD. LTX API pricing documentation, observed 2026-09-26.
Other ways to run it
- Run it as published · needs ~34.2 GB · from github.com/leejet
sd-cli -M vid_gen --diffusion-model models/ltx-2.3-22b-dev-UD-Q4_K_M.gguf --vae models/ltx-2.3-22b-dev_video_vae.safetensors --audio-vae models/ltx-2.3-22b-dev_audio_vae.safetensors --llm models/gemma-3-12b-it-qat-UD-Q4_K_XL.gguf --embeddings-connectors models/ltx-2.3-22b-dev_embeddings_connectors.safetensors -p "a lovely cat" --cfg-scale 6.0 --sampling-method euler -v -n "worst quality, low quality, blurry, distorted, artifacts" -W 1280 -H 720 --diffusion-fa --offload-to-cpu --video-frames 33 --fps 24 -o t2v.webm
Official weights and documentation: huggingface.co/Lightricks ↗
$ how the score was worked out · WorldReasonBench: Human-Align… 76%, MindWorldBench: Evaluating Me… 8%, Sci-VBench: Evaluating Knowle… 7% and 1 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (arXiv v1) | 1 | 76% |
| MindWorldBench: Evaluating Mental-State-to-Behavior Reasoning in Image-to-Video Generation (arXiv:2609.39147v1) | 1 | 8% |
| Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains | 1 | 7% |
| PhyGround: Benchmarking Physical Reasoning in Generative World Models (arXiv:2605.10806v1) | 1 | 6% |
| prior · pull toward the middle | 3% | |
| weighted average | = 36.3 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json