← All models

LTX-2.3

Video Available Full comparison ↗

lightricks/ltx-2-3 · by Lightricks · Diffusion transformer (DiT)

Pricing — 1 offering(s)

fal.ai

View on fal.ai ↗

Fast · 1080p

  • $0.06 / second Current 2026-07-15 → present Correction
    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the…

    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the wrong row is no longer served.

    In effect since at least 2026-07-15, per an archived copy of fal's model page. Recorded here on 2026-10-18; an earlier entry showed the lower rate from fal's out-of-date price table.

Fast · 1440p

  • $0.12 / second Current 2026-07-15 → present Correction
    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the…

    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the wrong row is no longer served.

    In effect since at least 2026-07-15, per an archived copy of fal's model page. Recorded here on 2026-10-18; an earlier entry showed the lower rate from fal's out-of-date price table.

Fast · 2160p (4K)

  • $0.24 / second Current 2026-07-15 → present Correction
    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the…

    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the wrong row is no longer served.

    In effect since at least 2026-07-15, per an archived copy of fal's model page. Recorded here on 2026-10-18; an earlier entry showed the lower rate from fal's out-of-date price table.

Pro · 1080p

  • $0.08 / second Current 2026-07-12 → present Correction
    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the…

    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the wrong row is no longer served.

    In effect since at least 2026-07-12, per an archived copy of fal's model page. Recorded here on 2026-10-18; an earlier entry showed the lower rate from fal's out-of-date price table.

Pro · 1440p

  • $0.16 / second Current 2026-07-12 → present Correction
    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the…

    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the wrong row is no longer served.

    In effect since at least 2026-07-12, per an archived copy of fal's model page. Recorded here on 2026-10-18; an earlier entry showed the lower rate from fal's out-of-date price table.

Pro · 2160p (4K)

  • $0.32 / second Current 2026-07-12 → present Correction
    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the…

    Corrected 2026-10-18: Modelglass first recorded a wrong price for these dates; this row replaces it, and the wrong row is no longer served.

    In effect since at least 2026-07-12, per an archived copy of fal's model page. Recorded here on 2026-10-18; an earlier entry showed the lower rate from fal's out-of-date price table.

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how LTX-2.3 fits into a cost-aware routing setup

See how →

Capability profile

How LTX-2.3 rates across core capability dimensions, with the task-level evidence behind each rating.

native audio Strong

Genuine architectural capability, not marketing copy — confirmed by Lightricks' own model card, and independently corroborated by Replicate's model page (generate_audio parameter, on by default) and fal.ai's release notes ("Cleaner native audio generated alongside video"). First native-audio model in this registry's LTX line.

Resolution ceiling Strong

Up to 2160p (4K) at 24-48fps per fal.ai's own model pages.

Inference speed Variable

Fast tier trades resolution/quality for 25% lower per-second cost than Pro at the same resolution across fal.ai's pricing tiers.

cost efficiency Moderate

$0.06-0.32/s depending on tier/resolution — pricier than Wan 2.5 or LTX Video 0.9.7, but adds native audio and up to 4K output neither of those offers.

temporal consistency Unknown
motion quality Unknown

Same reasoning as temporal-consistency — composite Arena Elo doesn't isolate this dimension. Confirmed still genuinely unavailable, 2026-08-19.

Prompt adherence Moderate
clip duration ceiling Moderate

Up to 20 seconds per fal.ai's text-to-video page; some secondary sources cite up to 30s for other endpoints, not confirmed on a Lightricks-owned page.

video editing Moderate

Supports extend and retake (re-generate specific sections of an existing clip) per Replicate's model listing, beyond plain T2V/I2V.

Ratings are estimated — limited independent data is available for this model.

Operator guidance

Route here over LTX Video 0.9.7 or Wan 2.5 when native synchronized audio or higher (1440p/4K) resolution is required — both are gaps in those cheaper open-source-lineage options. Within LTX-2.3, Fast is the cost-optimized default; Pro trades ~33% higher per-second cost for better fidelity. Still meaningfully cheaper than closed premium audio-video models like Veo 3.1 ($0.40-0.60/s) at the cost of being a smaller, less-polished model. dev/distilled open-weight checkpoints exist on Hugging Face for self-hosting, not covered by this hosted-inference doc.

Use cases

Limitations

Citations