This is single-model genuine WAN 2.2 14B, extracted from WAN 2.2 14B S2V. No merges or strange injections inside, just clean model.

Note 1. You should generate latent with WanSoundImageToVideo node, not WanImageToVideo. Connect source image to ref_image input to make model act as Image-to-video, or leave this input empty to achieve Text-to-video output.

Note 2. You should use T2V High Noise WAN2.2 Lightning LoRAs, not I2V or Low Noise. Don't ask, it's by design of source S2V model.

Note 3. Model will generate some warnings about audio encoder during loading, it's okay. Don't try to connect audio to model, audio is totally disabled.

I didn't test any LoRAs with this model except of lightning. Hope you will give me feedback about your experience and share your generated artworks on this page.