LTX-2.5
Lightricks
Lightricks' open-weight audio-video model, published on Hugging Face as a split, ComfyUI-aligned pack with one safetensors file per component rather than a single monolith. A 22B-parameter diffusion transformer paired with a custom Gemma 4 12B text encoder generates synchronized video and audio from text, image and video inputs, supported by separate video and audio VAEs, a vocoder, x2 spatial and temporal latent upscalers, and an optional duration head that infers clip length from the prompt. Both a distilled checkpoint on a fixed 8-step schedule at CFG 1 and a full trainable transformer are released.
Strengths
- Generates synchronized video and audio in one model, with a dedicated audio VAE plus vocoder
- Native multishot generation holds character identity, environment, lighting, voice and visual style across cuts
- 22B diffusion transformer with a custom Gemma 4 12B text encoder for long, multi-element prompts
- Ships both a distilled 8-step checkpoint at CFG 1 and a fully trainable dev transformer for LoRA work
- Optional duration head predicts clip length from the prompt instead of a fixed frame-count parameter
- Free for commercial and production use by entities under USD 10M annual revenue, self-hosted with no per-generation billing
- Runs through ltx-pipelines, ComfyUI or Diffusers, with int8, fp8-cast and NVFP4 builds plus CPU offload for low-VRAM machines
Caveats
- The Hugging Face repository is gated: "You need to agree to share your contact information to access this model" and to "consent to receive offers and updates including targeted and personalized advertisements"
- Not OSI open source - entities over USD 10M annual revenue need a paid commercial agreement, and revenue is measured across all subsidiaries and affiliates under common control
- Lightricks' own limitations: "This model is not intended or able to provide factual information", it "may amplify existing societal biases", "Prompt following is heavily influenced by prompting style" and it "may generate content that is inappropriate or offensive"
- No benchmark scores appear on the model card, and the paper it cites (arXiv 2601.03233) is about LTX-2, not LTX-2.5
- Rigid generation constraints and packaging friction: frame count must satisfy num_frames % 8 == 1, width and height must be divisible by 32, the int8-convrot files are ComfyUI-only, and "LTX-2.5 support is not in a diffusers release yet"
- Vendor
- Lightricks
- License
- LTX-2.x Community License Agreement
- Modalities
- videoaudio
Capabilities
Resources
Pricing
Open weights: "Commercial and production use at no cost under the LTX-2.x Community License" for entities under USD 10M annual revenue, where revenue "is measured across the whole entity, including subsidiaries and affiliates under common control". Entities over USD 10M need a "Paid Commercial Use Agreement for LTX-2.x with full weights, engineering support, LoRAs, and flexible deployment options", and "Transfer of fine-tunes may require a paid license". Lightricks also sells hosted API access and LTX Studio; no per-generation rate is stated on the model card.