LTX-2.5 Fast & Pro: Lightricks' Open Video Foundation

At a Glance

• IDs: catalog IDs ltx-2.5-fast / ltx-2.5-pro; live API slugs ltx-2-5-fast / ltx-2-5-pro
• Org: Lightricks (LTX)
• HF OW: Lightricks/LTX-2.5 (split Comfy-aligned pack; also Lightricks/LTX-2.5-Diffusers)
• Stack (vendor): ~22B DiT family (ltx-2.5-22b-* distilled / dev transformers); Gemma 4 12B TE; DiffVAE + audio VAE
• Modalities: Text / image / video in; synchronized video + audio (world-model pitch)
• License: LTX-2.x Community (free under $10M ARR; paid above)
• Min VRAM claim (product): 16GB class for self-host
• Prior catalog peers: ltx-2.3-fast, ltx-2.3-pro

Fast vs Pro (API)

Duration (API docs): Fast up to 20s at 720p/1080p @24/25 fps; 10s at 48/50 fps and at 1440p/4K. Pro: 6 / 8 / 10s only. Auto duration (duration: null) bills for the length produced.

Source: https://docs.ltx.io/models/ltx-2-5

Product FAQ API numbers ($0.15/1080p, $0.37/4K) disagree with live docs. Prefer docs.ltx.video/pricing below.

API pricing (live docs; per second of output)

Text-to-video and image-to-video (same rates):

Audio-to-video (1080p only on 2.5): Fast $0.13/s, Pro $0.17/s (billed on input audio duration). Auto-duration charges for the length the model actually produces. LTX-2 (ltx-2-fast / ltx-2-pro) is deprecated; removed after Aug 15, 2026.

Source: https://docs.ltx.video/pricing

What's New vs LTX-2.3

1. Native multi-shot in one pass (character / env / lighting / voice continuity across cuts).
2. New diffusion video decoder + dynamic compute by scene complexity.
3. Gemma 4 12B text encoder + prompt enhancer; optional duration predictor.
4. Stronger distilled checkpoint (API Fast path) with fewer retries claimed.
5. Native 4K HDR / RAW workflow pitch on the product page (self-host / finishing story; check which path your API job actually returns).

Open weights

HF pack is component-split (transformer, TE, VAEs, duration head, upscalers). Distilled DiT uses a fixed 8-step schedule (CFG=1). Dev transformer is the fine-tune base. Most 2.3 LoRAs / IC-LoRAs run on 2.5 per vendor testing (validate before production). Runtimes: ltx-pipelines, ComfyUI, Diffusers (main).

When to Use It

Good fit: Self-host / fine-tune video+audio stacks under the Community license; API Fast for cheap iteration up to 4K; API Pro when artifact rate matters more than resolution headroom.

Cost note: 2.5 Fast/Pro list above 2.3 at the same resolution. You are paying for the new foundation quality, not a price cut.

Not automatic: Retake / extend / HDR upscale / reframe endpoints on the pricing page still list ltx-2.3-pro only at fetch time. Confirm 2.5 coverage before promising those workflows.

Caveats

• Prefer pricing docs over the product FAQ when they conflict.
• Artifact / latency leaderboards on ltx.io are vendor-measured; not LLM Stats verified.
• License is Community, not Apache: ARR is entity-wide (subsidiaries/affiliates); fine-tune transfer may need a paid license.

Sources

• https://ltx.io/model/ltx-2-5
• https://docs.ltx.io/models/ltx-2-5
• https://huggingface.co/Lightricks/LTX-2.5
• https://docs.ltx.video/pricing