You will find here all the LTX-2 Video models used with WanGP.

LTX-2.5

WanGP provides LTX-2.5 Dev and Distilled in BF16 and INT8 ConvRot, plus the official Distilled NVFP4 variant. Large upstream bundles are split into independently offloadable transformer, Gemma 4, text-projection, video/audio connector, VAE, vocoder, and upscaler files. Dev and Distilled share connector files only where their tensors were verified identical; NVFP4 has checkpoint-specific connectors.

See LTX-2.5-MANIFEST.md for the runtime file map and LICENSE-LTX-2.5.txt for the upstream license. The split files preserve the original tensor values; their metadata records the WanGP packaging modification.

WanGP by DeepBeepMeep makes leading open-source video generative models accessible to the GPU poor, with low VRAM modes, support for older GPUs, automatic model downloads, queues, integrated media tools, and LoRA workflows.

Discord: https://discord.gg/g7efUW9jGV

Follow DeepBeepMeep on Twitter/X: https://x.com/deepbeepmeep

Downloads last month: 262,935

GGUF

Model size

19B params

Architecture

ltxv

Hardware compatibility

4-bit

6-bit

8-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for DeepBeepMeep/LTX-2

Base model

Lightricks/LTX-2

Quantized

(9)

this model

DeepBeepMeep
/

LTX-2

LTX-2.5

Model tree for DeepBeepMeep/LTX-2

Spaces using DeepBeepMeep/LTX-2 13