🎧 Trigger Words<code>asmr ear lick</code> — main trigger, always includeAlso recommended in your prompt:<ul><li><code>binaural</code></li><li><code>3dio microphone</code></li></ul>⚙️ Recommended SettingsStrength: 1.0<hr />AboutThis LoRA produces surprisingly consistent, good-quality audio.The dataset primarily included clips a single actor performing alone but also included a few of two actors performing simultaneously, so dual/two-sided prompts will sometimes work. Be aware it's unreliable at syncing both sides at once at the moment - expect to try a few seeds if that's what you're going for.Heads up: part of the dataset contained NSFW themes, so outputs may lean that direction even on tamer prompts.🚧 RoadmapTraining is already underway on a larger dataset, possibly with a higher step count. Updated versions to follow. (Training ain't cheap, lol.)💬 FeedbackThis is my first LoRA, so any feedback, sample gens, or prompt tips are appreciated. <hr />🛠️ Bonus: LTX Training UII built a custom web UI specifically for this workflow and I'm considering releasing it. What it does:<ul><li>Browse and select videos straight from your local drive</li><li>Automatic clipping / segmenting</li><li>Automatic captioning via a model</li><li>Train directly from the interface</li><li>Optional RunPod integration — drop in your API key and it spins up a pod for training and serverless for captioning on its own</li></ul>If that sounds useful to you, let me know and I'll prioritize getting it out.

ASMR Ear Lick — LTX LoRA

ASMR-Ear-Lick-run12.safetensors

v1.0

<h1 id="asmr-ear-lick-ltx-2-video-+-audio-lora">ASMR Ear Lick — LTX-2 video + audio LoRA</h1><h2 id="plus-a-pony-lora-for-i2v-image-gen!">Plus a PONY LoRA for I2V image gen!</h2>Trigger: <code>asmr ear lick</code> — first words of the prompt, verbatim.Trained on 2,700 clip (3.8 hrs) of real 3dio binaural ear-licking, captioned with the audio described alongside the video. LTX-2 generates sound with the picture, and this LoRA trains the audio branch too — so it renders the wet, rhythmic binaural licking track locked to the tongue, not just the visual.Two builds: the Eros one (rank 64, audio layers trained, 9k steps — pair it with the 10-Eros distilled checkpoint) and the LTX-2 one (rank 32, stock base). Eros is the stronger of the two.Weight 1.0 (works 0.85–1.05). Tested at portrait 448×768, 481 frames (~20 s), t2v &amp; i2v, random seed.<h2 id="prompting">Prompting</h2>Words that work: <code>3dio microphone</code> · <code>aggressive ear licking</code> · <code>silicon ears</code> · <code>static close-up on her mouth and the microphone</code> · <code>licks</code> · <code>laps</code> · <code>slow wet licks</code> · <code>tongue drags</code> · <code>presses and slides along every curve</code> · <code>tracing the ear shapes repeatedly</code> · <code>mouth open, tongue visible as it makes contact</code> · <code>glistening saliva trails</code> · <code>wet mouth and licking sounds</code> · <code>wet and rhythmic</code> · <code>continuous</code> · <code>slow and deliberate</code>Always close with: <code>3dio microphone ear lick. Binaural audio.</code>The tricks:<ol><li>Give the audio its own sentence and tie it to the action — "The audio consists of the sounds of the woman licking the microphone, which are wet and rhythmic." Describing the room instead of the licking gives you vague, detached sound.</li><li>Never write "stroke"/"strokes" — the model reads it as hand stroking and puts her hands on the mic. Use licks, laps, tongue drags.</li><li>Always say the tongue is visible — "mouth open, tongue visible as it makes contact." A bare "licks the mic" renders a soft, undetailed tongue.</li><li>Say which side of her face the mic is on. Anchors the layout so the binaural image matches the picture.</li><li>Don't invent sound words. "Saliva clicks", "squelches", "suction pops" aren't in the captions and degrade the track — stick to wet/rhythmic/continuous. (Visual "glistening saliva trails" is fine.)</li><li>Go light on breathing — mentioning it pulls the mix away from the licks.</li><li>Negative prompts do nothing on the distilled checkpoints (CFG-free). Steer with positive phrasing only.</li><li>Two girls: side-by-side positional descriptors, one shared mic held between them, and a sequential join — "the woman on the left begins by licking… while the other watches, then joins in." Describe the audio globally; never map a sound to one girl or one ear.</li></ol>

ASMR Ear Licking

Base LTX 2.3 - not ErosThis one isn't really recommended as LTX 2.3 kinda struggles with tongues but its worth a shot and gives decent audio.

Files with SHA256: 28b17fbd21b4826b07e85caf590d19e96e5ddf56a98d4512f9990c947c606ea8