Skip to main content

LightricksVideo model

LTX-2.5

Clip lengthn/a

n/alongest single cliphigher is better

PopularityRank 14 of 875

70out of 100+1 vs. last month

Summary

New in LTX-2.5 are open weights for video, audio, and world simulation. Lightricks describes a 22-billion-parameter diffusion core and a Gemma4-12B text encoder; video and audio are generated in sync across several local execution modes.

The API groups text-to-video, image-to-video, and audio-to-video under ltx-2-5-fast and ltx-2-5-pro. Fast reaches 4K, while Pro offers 720p and 1080p. Both are serving modes of one family.

What LTX-2.5 can do

“n/a” means the provider does not state it or we have not sourced it yet. That is not a “no”.

Text to videoyesSource
Image to videoyesSource
Referencesn/a
AudioyesSource
Runs locallyyesSource
Extendn/a
Start and end framen/a
Max. resolution4K (API Fast; Pro 720p oder 1080p)Source

Video sample from our test

This video exists as a documented run for the exact model version. Playback is muted.

Motion and physical consistency

This sample tests Anatomy, object consistency, camera tracking, rain, and reflections.

Open video at original sizeOpen source
Prompt and settings

Generate exactly one continuous five-second 16:9 cinematic video. A fictional bicycle courier in a mustard-yellow raincoat rides a red city bicycle from left to right across a rain-wet Hamburg street at blue hour for the entire clip. The camera tracks her smoothly at waist height. Both hands remain visible on the handlebars, the bicycle geometry stays physically correct, and a transparent umbrella fixed to the bicycle moves naturally in the wind. Wet-street reflections and raindrops follow realistic physics. No cuts, no slow motion, no text, no logos, no subtitles, and no audio.

Generated through fal.ai with model lightricks/ltx-2.5/text-to-video/fast. Archived settings were 16:9, one request, first output, prompt motion-and-physics, requested 6 s, off, seed provider random, request {"prompt":"Generate exactly one continuous five-second 16:9 cinematic video. A fictional bicycle courier in a mustard-yellow raincoat rides a red city bicycle from left to right across a rain-wet Hamburg street at blue hour for the entire clip. The camera tracks her smoothly at waist height. Both hands remain visible on the handlebars, the bicycle geometry stays physically correct, and a transparent umbrella fixed to the bicycle moves naturally in the wind. Wet-street reflections and raindrops follow realistic physics. No cuts, no slow motion, no text, no logos, no subtitles, and no audio.","resolution":"720p","duration":6,"aspect_ratio":"16:9","fps":24,"generate_audio":false}, published quote 0.54 USD, reserved estimate 0.63 USD, quote SHA-256 87984b917d2a51622bcd5f61aa0cdcbb7f621bdb8bec8f736fff696cc7c3abb7.

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 53 individual results from 6 sources

Page 1 of 5

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

WorldGuide Bench (Show measurement, test conditions, and source)
Source value
7.32
Score
7.32
Metric
Task Succ.
Unit
%
Benchmark version
arXiv:2610.12459v1 (2026-10-08)
Category
Procedural video-task completion
Direction
Higher is better
Source type
independent-evaluator
Evaluator
WorldGuide: Goal-Directed Video World Model for Procedural Task Execution
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. WorldGuide Bench Table 2, exact model row "LTX-2.5-22B"; unstarred baseline is given the complete ground-truth action plan and generates the procedure step by step. The score is the paper's video-judged task-success rate.
Context
These are end-to-end procedural video-task baseline results under the stated conditioning, not an autonomous agent-system result or an unconditional generic video-quality score. The paper gives model labels and relevant task conditions, but no immutable serving/checkpoint digest or confidence interval. A per-model valid denominator is omitted where the paper does not state it.
Source WorldGuide: Goal-Directed Video World Model for Procedural Task Execution
Video-CraftBench (Show measurement, test conditions, and source)
Source value
3.44
Score
3.44
Metric
Task Succ.
Unit
%
Benchmark version
arXiv:2610.12459v1 (2026-10-08)
Category
Procedural video-task completion
Direction
Higher is better
Source type
independent-evaluator
Evaluator
WorldGuide: Goal-Directed Video World Model for Procedural Task Execution
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Video-CraftBench Table 3, exact model row "LTX-2.5"; every model receives the initial image and task goal without reference actions. The score is the paper's video-judged task-success rate.
Context
These are end-to-end procedural video-task baseline results under the stated conditioning, not an autonomous agent-system result or an unconditional generic video-quality score. The paper gives model labels and relevant task conditions, but no immutable serving/checkpoint digest or confidence interval. A per-model valid denominator is omitted where the paper does not state it.
Source WorldGuide: Goal-Directed Video World Model for Procedural Task Execution
REMEDY: strict clinical success rate (PKA) (Show measurement, test conditions, and source)
Source value
12.11
Score
12.11
Metric
Task-level strict clinical success rate for PKA; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (TKA) (Show measurement, test conditions, and source)
Source value
6
Score
6
Metric
Task-level strict clinical success rate for TKA; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (MISS) (Show measurement, test conditions, and source)
Source value
10
Score
10
Metric
Task-level strict clinical success rate for MISS; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (US) (Show measurement, test conditions, and source)
Source value
4.71
Score
4.71
Metric
Task-level strict clinical success rate for US; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (Inj.) (Show measurement, test conditions, and source)
Source value
6
Score
6
Metric
Task-level strict clinical success rate for Inj.; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (Exam) (Show measurement, test conditions, and source)
Source value
50
Score
50
Metric
Task-level strict clinical success rate for Exam; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (Wound) (Show measurement, test conditions, and source)
Source value
16
Score
16
Metric
Task-level strict clinical success rate for Wound; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (Dent.) (Show measurement, test conditions, and source)
Source value
15
Score
15
Metric
Task-level strict clinical success rate for Dent.; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (IV) (Show measurement, test conditions, and source)
Source value
2.5
Score
2.5
Metric
Task-level strict clinical success rate for IV; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY: strict clinical success rate (Sut.) (Show measurement, test conditions, and source)
Source value
0
Score
0
Metric
Task-level strict clinical success rate for Sut.; every essential item must pass and no critical failure may occur
Unit
%
Benchmark version
REMEDY, arXiv:2609.32460 (2026-09-26)
Category
medical teaching demonstration video generation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
REMEDY: How Far Is Video Generation from Medical Education World Models?
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. Exact evaluated model label LTX-2.5, an open-source text-image-to-video (TI2V) model. The paper says it used official checkpoints and inference pipelines on NVIDIA H200 GPUs, with the same 900 real clinical first frames and task-specific prompts across models. Exact checkpoint revision and decoding/settings details are not stated in the captured results.
Source REMEDY: How Far Is Video Generation from Medical Education World Models?

Page 1 of 5

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

0 to 100, higher is more popular

20 of 98 models

LTX-2.570
MiniMax H366
Muse Video62
Sora 259
Wan 2.1 T2V 14B54
Wan 2.1 T2V 1.3B54
Runway Gen-4.553
Hy-Nomas53
Wan 2.1 I2V 14B53
Veo 3.150
Wan 2.2 TI2V 5B46
FLUX.3 Video45
LTX-2 19B45
Open-Sora 2.044
Wan 2.2 I2V A14B43
Gemini Omni Flash43
SkyReels V2 T2V 14B42
Grok Imagine Video42
Mochi 1 Preview39
Pika H3 Flex35

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 5, 2026

LTX-2.5MiniMax H3Muse VideoSora 2Runway Gen-4.5

LTX-2.5 compared

Compare with another model

The most popular comparisons

  1. 1Gemini Omni Flash vs. LTX-2.5
  2. 2FLUX.3 Video vs. LTX-2.5
  3. 3Dreamina Seedance 2.5 720p vs. LTX-2.5
  4. 4Dreamina Seedance 2.0 720p vs. LTX-2.5
  5. 5MiniMax H3 vs. LTX-2.5
  6. 6Sora 2 Pro vs. LTX-2.5
  7. 7Sora 2 vs. LTX-2.5
  8. 8Veo 3.1 vs. LTX-2.5
  9. 9Veo 3.1 Fast vs. LTX-2.5
Show all 47 comparisons
  1. 10Grok Imagine Video 1.5 vs. LTX-2.5
  2. 11Grok Imagine Video vs. LTX-2.5
  3. 12Wan 2.7 vs. LTX-2.5
  4. 13Wan 2.6 vs. LTX-2.5
  5. 14Runway Gen-4.5 vs. LTX-2.5
  6. 15Kling 3.0 Pro vs. LTX-2.5
  7. 16Kling 2.6 Pro vs. LTX-2.5
  8. 17Vidu Q3 Pro vs. LTX-2.5
  9. 18PixVerse V5.6 vs. LTX-2.5
  10. 19Hailuo 2.3 vs. LTX-2.5
  11. 20LTX-2 19B vs. LTX-2.5
  12. 21Wan 2.2 A14B vs. LTX-2.5
  13. 22HunyuanVideo 1.5 vs. LTX-2.5
  14. 23Pika 2.2 vs. LTX-2.5
  15. 24Gemini Omni 1.1 Flash vs. LTX-2.5
  16. 25Wan 3.0 vs. LTX-2.5
  17. 26Wan 3.0 Prime vs. LTX-2.5
  18. 27HappyHorse 1.0 vs. LTX-2.5
  19. 28HappyHorse 1.1 vs. LTX-2.5
  20. 29PixVerse V6 vs. LTX-2.5
  21. 30PixVerse C1 vs. LTX-2.5
  22. 31Seedance 2.0 Fast vs. LTX-2.5
  23. 32Seedance 2.0 Mini vs. LTX-2.5
  24. 33Vidu Q3 Turbo vs. LTX-2.5
  25. 34Vidu Q2 vs. LTX-2.5
  26. 35Vidu Q1 vs. LTX-2.5
  27. 36Kling 4.0 vs. LTX-2.5
  28. 37Kling 3.0 Standard API vs. LTX-2.5
  29. 38Kling Video O1 vs. LTX-2.5
  30. 39Hailuo 2.3 Fast vs. LTX-2.5
  31. 40Wan 2.5 Preview vs. LTX-2.5
  32. 41Wan 2.6 I2V Flash vs. LTX-2.5
  33. 42Wan 2.2 TI2V 5B vs. LTX-2.5
  34. 43Wan 2.2 I2V A14B vs. LTX-2.5
  35. 44HunyuanVideo 1.0 T2V vs. LTX-2.5
  36. 45HunyuanVideo I2V 1.0 vs. LTX-2.5
  37. 46Ray 3.2 vs. LTX-2.5
  38. 47Grok Imagine Video 1.5 Lite vs. LTX-2.5

All specifications

Architecture

Parameters
22B
Source: Parameters
Model class
Video, audio, and world simulation model
Source: Model class

Input and output

Maximum resolution
4K (API Fast; Pro 720p oder 1080p)
Source: Maximum resolution
Native audio
Yes
Source: Native audio
Modalities
Text-to-video, image-to-video, and audio-to-video with synchronized audio
Source: Modalities

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

Weights
Open weights
Source: Weights
Access
API, Web, Local
Source: Access
License
LTX-2.x Community License, with a separate agreement for larger companies
Source: License

Further details

API model IDs
ltx-2-5-fast and ltx-2-5-pro are separate API modes in the existing LTX-2.5 profile family.
Source: API model IDs
Weights repositories
Lightricks/LTX-2.5 and Lightricks/LTX-2.5-Diffusers. The Diffusers checkpoint contains distilled and full/SFT transformers in separate subfolders.
Source: Weights repositories
Input modes
Text-to-video, image-to-video, and audio-to-video with synchronized audio.
Source: Input modes
Generation limits
Fast API mode up to 4K; Pro API mode at 720p or 1080p. Audio-to-video outputs 1080p in both API modes.
Source: Generation limits
Weights license
LTX-2.x Community License; Lightricks specifies a separate agreement for larger enterprises.
Source: Weights license
API modes
ltx-2-5-fast up to 4K and ltx-2-5-pro at 720p or 1080p
Source: API modes
Audio
Audio-to-video generates 1080p on both API modes
Source: Audio

Prices

fal.ai (lightricks/ltx-2.5/text-to-video/fast)
$0.63 per request (estimated)Source

Sources and data date

Show all 20 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

EvaluatorTest seriesComposite indicesRecorded active measurements
Artificial Analysis007

Without vendor claims and historical or disputed results; those stay labelled in the single values. 7 measurements are not yet assigned to a test series.

TypeEvidence and data date
Research statusResearch date October 4, 2026. 17 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources

family-level-benchmark

Original wording

AA reports Fast at 945 Elo and Pro at 943 Elo in separate audio-enabled text-to-video cohorts. Neither result is transferred to the family profile.

Not published

ltx-2-5-unqualified-family-score

Note

AA publishes exact Fast and Pro rows separately for T2V and I2V; no unqualified LTX-2.5 family aggregate is asserted.

Not found in the inspected sources

wiro-video-models-exact-row

Note

The inspected Wiro leaderboard has no exact model-version route. Nearby variants were not transferred.

Not found in the inspected sources

arena-video-edit-exact-row

Note

The inspected Video Edit table has no exact model-version row. T2V and I2V values were not reused.

Not found in the inspected sources

vabench-exact-model-row

Note

VABench has no exact model row. Sora 2 was not transferred to Sora 2 Pro or other models.

Not found in the inspected sources

arena-t2v-exact-row

Note

No exact Arena text-to-video observation was found in the checked data.

Not found in the inspected sources

arena-i2v-exact-row

Note

No exact Arena image-to-video observation was found in the checked data.

Unresolved

vbench-current-public-leaderboard

Note

The current VBench leaderboard is embedded in a dynamic Hugging Face iframe and was not retrievable; current scores remain unresolved.

Not found in the inspected sources

vidbenchmark-exact-row

Note

No exact model id is present in the 2026-08-04 dataset; similarly named variants were not transferred.

Unresolved

Artificial Analysis text-to-video exact current result

Note

The opened leaderboard response exposed only its heading and column header, not an exact LTX-2.5 result cell. This does not affect the separately sourced WorldGuide human-evaluation row.

Additional sourceLightricks (retrieved October 3, 2026; October 15, 2026; November 1, 2026) · LTX-2.5 model card · Editorial description reviewed October 3, 2026
Additional sourceLightricks (retrieved November 1, 2026)12
Additional sourcefal.ai
Additional sourceLTX-2.5 API changelog (retrieved October 3, 2026) · LTX-2.5 API changelog · Editorial description reviewed October 3, 2026
Additional sourceArtificial Analysis AA-Video-T2V v2.1 (retrieved October 15, 2026; October 16, 2026; October 18, 2026)
Additional sourceArtificial Analysis AA-Video-I2V v1.1 (retrieved October 15, 2026; October 16, 2026; October 18, 2026)
Additional sourceWorldGuide: Goal-Directed Video World Model for Procedural Task Execution (retrieved October 17, 2026; October 19, 2026)
Additional sourceREMEDY: How Far Is Video Generation from Medical Education World Models? (retrieved October 18, 2026)
Additional sourceCOEY LTX 2.5 Apple Silicon speed and memory study (retrieved October 20, 2026)
Additional sourceSpicyAPI LTX 2.5 test report (retrieved October 21, 2026)
Additional sourceModel metadata (retrieved October 16, 2026)123
Additional sourceOpenRouter: Popularity index (derived), CC BY 4.0
Additional sourceVercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional sourceOllama: cumulative downloads for the popularity index
Additional sourceReplicate: cumulative runs for the popularity index
Additional sourceGoogle Trending Now: limited feed selection for Germany and the US
Additional sourceCreative Commons Attribution 4.0