Filmed without sound? No problem, LoRA Foley has you covered

The Foley LoRA generates a synchronized soundtrack for your silent videos. An open-weights adapter for LTX-2.3-22B that targets everyday sounds.

The LTX-2.3-22B model hosts a specialized LoRA that synthesizes sound effects synced to the visuals of silent clips, based on a prompt describing the visible action. The same adapter also accepts text-to-audio prompts.

The adapter targets the audio attention blocks and the video-to-audio cross-attention, keeping the input frames frozen and synthesizing only the soundtrack. It adds or regenerates sound effects for silent clips—footsteps, impacts, textures, ambiances, tools, cooking, weather—based on a prompt describing the action visible on screen. The training, conducted on approximately five thousand three hundred short clips, favors an intermediate checkpoint (step 500): pushing further improved music rejection but crushed the volume.

The scope remains intentionally narrow. No dialogue or speech synthesis, no music, no lip-sync, and the publisher specifies that the tool replaces neither a full production mix nor a licensed effects library when rights matter. A prompt that does not match the image produces a plausible but desynchronized sound.

The LoRA is available as open weights in safetensors format and loads into ComfyUI on top of the base model, via the dedicated video-to-audio workflow. Distribution is under the LTX-2 community license, with a portion of the dataset remaining subject to non-commercial terms.