ASMR Ear Lick — LTX-2 video + audio LoRA

Plus a PONY LoRA for I2V image gen!

Trigger: asmr ear lick — 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 & i2v, random seed.

Prompting

Words that work: 3dio microphone · aggressive ear licking · silicon ears · static close-up on her mouth and the microphone · licks · laps · slow wet licks · tongue drags · presses and slides along every curve · tracing the ear shapes repeatedly · mouth open, tongue visible as it makes contact · glistening saliva trails · wet mouth and licking sounds · wet and rhythmic · continuous · slow and deliberate

Always close with: 3dio microphone ear lick. Binaural audio.

The tricks:

1. 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.

2. Never write "stroke"/"strokes" — the model reads it as hand stroking and puts her hands on the mic. Use licks, laps, tongue drags.

3. Always say the tongue is visible — "mouth open, tongue visible as it makes contact." A bare "licks the mic" renders a soft, undetailed tongue.

4. Say which side of her face the mic is on. Anchors the layout so the binaural image matches the picture.

5. 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.)

6. Go light on breathing — mentioning it pulls the mix away from the licks.

7. Negative prompts do nothing on the distilled checkpoints (CFG-free). Steer with positive phrasing only.

8. 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.