Fine-tune vs straight fal call

Left column: our fine-tunes, full-precision bf16, native 345 frames, 384×672 portrait, ~$1/video on a B200. Right: hosted minimax/h3 text-to-video on fal, 768P, 10 s, 9:16, prompt_expansion_mode=fast, identical prompt text. Rendered: 25/25 ours, 25/25 fal. Click play for sound; double-click a video for fullscreen.

Round 3 — one photo → UGC video (i2v identity test)

Left: the single input photo. Second: our scratch100 LoRA rendering the performance from a prompt written around that person (text-only — the model never saw the photo). Third: PURE TRAINED MODEL i2v — the photo is a native first-keyframe anchor (encode_keyframes.py), no fal involved. Right: fal reference-to-video given the photo + our LoRA's audio, at 2K.

Weekly paper planner — office desk — photo identity

input photo
ours — LoRA performance (text-only)
ours — native i2v from the photo (no fal)
fal ref-to-video — photo + our audio (2K)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright modern-office daylight. A woman in her early thirties with shoulder-length wavy dark brown hair, wearing a cream knit sweater (S1) films a user-generated ad about a weekly paper planner in a modern office with dark wood shelving and plants behind her. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the planner flipping open to a weekly spread, a pen checking off boxes, and the elastic band snapping closed. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I run my entire week out of this planner now. Every Sunday I do a ten minute reset, and I have not missed a deadline since. The paper is thick enough for fountain pens, and the habit tracker on each page is honestly what keeps me consistent.</d>

**overall_soundscape:**

Quiet office room tone with a close phone-microphone voice. Page flips, a pen scratching, and the elastic snap are audible. No non-diegetic music.

Cold brew concentrate — office desk — photo identity

input photo
ours — LoRA performance (text-only)
ours — native i2v from the photo (no fal)
fal ref-to-video — photo + our audio (2K)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright modern-office daylight. A woman in her early thirties with shoulder-length wavy dark brown hair, wearing a cream knit sweater (S1) films a user-generated ad about a portable cold brew concentrate in a modern office with dark wood shelving and plants behind her. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a small concentrate bottle being uncapped, a dark pour swirling into a glass of iced water, and a stir with a metal straw. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I stopped waiting in coffee lines because of this. One pump into cold water and it tastes like the eight dollar cold brew from the shop downstairs. It lives in my desk drawer, no fridge needed, and one bottle lasts me almost a month.</d>

**overall_soundscape:**

Quiet office room tone with a close phone-microphone voice. A cap twist, liquid pouring over ice, and a metal straw clinking are audible. No non-diegetic music.

Hybrid — our render as fal reference (ref-to-video, 2K)

Left: our render (the reference source). Middle: fal conditioned on our audio track + one identity frame. Right: fal conditioned on our FULL VIDEO as the reference URL (motion + identity + audio in one input).

Claw clip — Latina woman, 20s, bathroom mirror — hybrid

ours — reference source
fal — audio + frame refs (2K)
fal — full video ref URL (2K)

Espresso machine — Indian man, 40s, kitchen — hybrid

ours — reference source
fal — audio + frame refs (2K)
fal — full video ref URL (2K)

Round 2 — OpenVDN base vs audio_scratch/step100 vs fal (15 characters)

Three-way: the released base model (left), our audio-only fine-tune of it (middle: same physics, trained UGC audio), and fal's hosted minimax/h3 (right). Same prompt text everywhere. Base vs ours isolates exactly what the LoRA adds — listen to the audio.

Beard oil — Black man, 30s, bathroom

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, warm bathroom lighting. A Black man in his thirties with a full beard and a fade haircut (S1) films a user-generated ad about a cedarwood beard oil in his bathroom, standing at the sink. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a few amber drops falling into his palm and being worked through the beard with both hands. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] My beard was itchy and patchy-looking until I started using this oil. Three drops, rub it in, takes ten seconds. No more flakes, no more itch, and my girl says it smells like a cabin in the woods. Been using it every morning for two months.</d>

**overall_soundscape:**

Bathroom room tone with a close phone-microphone voice. A dropper clink and palms rubbing together are audible. No non-diegetic music.

Robot vacuum — East Asian woman, 40s, living room

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright afternoon living-room light. An East Asian woman in her forties with a short bob and cardigan (S1) films a user-generated ad about a robot vacuum in her hardwood-floor living room with a cat on the sofa. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the vacuum gliding under the sofa, the cat watching it pass, and the dustbin being emptied into a trash can. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I have two cats and a toddler, so my floors were a disaster every single day. Now I start this from my phone before school pickup and come home to clean floors. It gets under the couch where the cat hair lives. I have not touched my old vacuum in a month.</d>

**overall_soundscape:**

Living room tone with a close phone-microphone voice. The vacuum's soft whir and a faint cat meow are audible. No non-diegetic music.

Matte lipstick — Middle Eastern woman in hijab, 20s

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, soft window light. A Middle Eastern woman in her twenties wearing a beige hijab (S1) films a user-generated ad about a matte liquid lipstick in her bedroom, sitting by the window. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the wand being pulled from the tube, one smooth swipe across the lips, and a coffee cup with no lipstick mark on the rim. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] This lipstick survived my wedding, and I cried at my wedding. One coat, it dries down in a minute, and it does not transfer. I drank coffee all morning, nothing on the cup. It is the only matte that does not dry my lips out either.</d>

**overall_soundscape:**

Quiet bedroom room tone with a close phone-microphone voice. A soft click of the tube cap is audible. No non-diegetic music.

Cordless drill — white man, 50s, garage

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, garage work-light lighting. A white man in his fifties with a gray mustache wearing a work jacket (S1) films a user-generated ad about a compact cordless drill in his garage workbench. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a screw driving flush into a two-by-four, the clutch dial being turned, and the battery clicking into place. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I have owned every drill brand you can name, and this little twelve-volt is what I actually reach for. Light enough for overhead work, drives a three-inch screw no problem. Battery lasted a whole deck repair. Do not let the size fool you.</d>

**overall_soundscape:**

Garage room tone with a close phone-microphone voice. The drill's motor whine, a screw seating, and a battery click are audible. No non-diegetic music.

Meal prep containers — Latina woman, 30s, kitchen

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright kitchen lighting. A Latina woman in her thirties with hoop earrings and hair in a bun (S1) films a user-generated ad about glass meal prep containers in her kitchen counter with five containers lined up. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing rice being portioned into containers, lids snapping on, and a stack going into the fridge. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] Sunday meal prep changed my whole week, and these containers are why. The lids actually snap tight, so no soggy salads, no leaks in my work bag. They go from freezer to microwave to dishwasher. I meal prep five lunches in under an hour now.</d>

**overall_soundscape:**

Kitchen room tone with a close phone-microphone voice. Rice scooping, lids snapping, and the fridge door are audible. No non-diegetic music.

Silk pillowcase — Black woman, 40s, bedroom

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, cozy evening bedroom lighting. A Black woman in her forties with natural curly hair (S1) films a user-generated ad about a mulberry silk pillowcase in her bedroom, sitting on the bed. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a hand smoothing across the silk, hair gliding over the surface, and the pillowcase being zipped on. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] If you have curly hair, listen to me. I switched to this silk pillowcase six months ago and I wake up with zero frizz. No more creases on my face either. I bought the cheap satin one first, trust me, the real silk is different. Best sleep upgrade I have made.</d>

**overall_soundscape:**

Quiet bedroom room tone with a close phone-microphone voice. Soft fabric sounds and a gentle zip are audible. No non-diegetic music.

Energy powder — white college guy, dorm kitchen

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, ordinary dorm-kitchen lighting. A white college-age man with shaggy brown hair in a hoodie (S1) films a user-generated ad about a sugar-free energy drink powder in his apartment kitchen before class. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a scoop dissolving into a glass of water with a swirl, and the glass being downed. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I quit my four dollar energy drink habit for this powder. One scoop is like half the price, zero sugar, and I do not get the two PM crash anymore. It tastes like green apple. I keep a bag in my backpack for the library. My wallet is so happy.</d>

**overall_soundscape:**

Kitchen room tone with a close phone-microphone voice. Powder fizzing into water and a glass clink are audible. No non-diegetic music.

Mechanical keyboard — East Asian man, 20s, gaming desk

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, desk RGB-accent lighting at night. An East Asian man in his early twenties with glasses and a black tee (S1) films a user-generated ad about a mechanical keyboard in his gaming desk setup. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing fingers typing fast in close-up, a keycap being pulled off to show the switch, and the volume knob being rolled. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] Okay the sound test is coming, but first: hot-swappable switches, gasket mount, and this knob controls volume. It sounds like rain on a window when I type. My old board sounded like a typewriter having a breakdown. Here, listen to this.</d>

**overall_soundscape:**

Quiet room tone with a close phone-microphone voice. Crisp thocky keystrokes and a knob's soft detents are audible. No non-diegetic music.

Espresso machine — Indian man, 40s, kitchen

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, morning kitchen light. An Indian man in his forties with salt-and-pepper hair (S1) films a user-generated ad about a compact espresso machine in his kitchen counter. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a portafilter being locked in, espresso streaming into a small cup with crema, and milk steaming. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I did the math and my coffee shop habit was costing me two thousand a year. This machine paid for itself in four months. Look at that crema. It heats up in three minutes, and the milk wand makes a better flat white than the café near my office. My mornings are transformed.</d>

**overall_soundscape:**

Kitchen room tone with a close phone-microphone voice. The pump's hum, espresso trickling, and milk steaming hiss are audible. No non-diegetic music.

Gym duffel — Middle Eastern man, 20s, entryway

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, locker-room style lighting. A Middle Eastern man in his twenties with short dark hair and a athletic build (S1) films a user-generated ad about a gym duffel bag with a shoe compartment in his apartment entryway bench. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing sneakers sliding into the ventilated bottom pocket, a wet towel going into a sealed side pocket, and the bag zipping shut. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] This bag ended the era of my gym shoes touching my clean clothes. Separate shoe garage on the bottom, waterproof pocket for the wet towel, and a little sleeve for my padlock. It fits in the work locker too. Should have bought this years ago.</d>

**overall_soundscape:**

Entryway room tone with a close phone-microphone voice. Zippers, shoes sliding on fabric, and a strap clip are audible. No non-diegetic music.

Claw clip — Latina woman, 20s, bathroom mirror

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright bathroom vanity lighting. A Latina woman in her early twenties with long thick wavy hair (S1) films a user-generated ad about an oversized claw clip in her bathroom mirror. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing hair being twisted up and clipped in one motion, a head shake to prove it holds, and the clip releasing a cascade of hair. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] Girls with thick hair, this is the only claw clip that has never snapped on me. Watch, one twist, clip, done. I can drive, work out, whatever, it does not budge and it does not give me a headache. I have one in every color now.</d>

**overall_soundscape:**

Bathroom room tone with a close phone-microphone voice. The clip's spring click and hair rustling are audible. No non-diegetic music.

Running shoes — Black man, late 20s, porch

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, early morning outdoor light on a porch. A Black man in his late twenties in a running jacket (S1) films a user-generated ad about cushioned running shoes in his front porch before a run. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the shoe flexing in his hands, the foam midsole being pressed with a thumb, and laces being tied. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I run about thirty miles a week and my knees used to hate me for it. Four hundred miles on this pair and the cushion still feels brand new. No more shin splints, no more numb toes. I just bought my second pair before they change the design. Runners know.</d>

**overall_soundscape:**

Outdoor morning ambience with birds and a close phone-microphone voice. Foam squishing and laces snapping tight are audible. No non-diegetic music.

Wood-wick candle — tattooed white woman, 30s

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, warm dim living-room lighting in the evening. A white woman in her thirties with visible arm tattoos and a septum ring (S1) films a user-generated ad about a wooden-wick soy candle in her living room coffee table. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a match lighting the flat wooden wick, the flame settling, and the label being turned toward the camera. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] This candle is why my apartment smells like a bookstore and a fireplace had a baby. The wooden wick actually crackles, listen. It burns clean for like sixty hours and no headache like the mall candles give me. I am on my fifth one, send help.</d>

**overall_soundscape:**

Quiet living room tone with a close phone-microphone voice. A match strike and the wick's soft crackle are audible. No non-diegetic music.

Water flosser — South Asian woman, 30s, bathroom

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright bathroom lighting. A South Asian woman in her thirties with shoulder-length dark hair (S1) films a user-generated ad about a cordless water flosser in her bathroom sink. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the reservoir being filled, the tip clicking on, and a demonstration spray into the sink. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] My dentist called me out at every visit until I got this thing. Thirty seconds after brushing and my gums stopped bleeding within two weeks. It is cordless, waterproof, and the tank lasts a full clean. Last checkup: zero lectures. Worth every penny.</d>

**overall_soundscape:**

Bathroom room tone with a close phone-microphone voice. Water filling, a click, and a fine spray hitting the sink are audible. No non-diegetic music.

Felt desk pad — white woman, late 20s, home office

OpenVDN base — stage-b-step-2000 (bf16, 345f)
ours — audio_scratch step100 (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, clean daylight home-office lighting. A white woman in her late twenties with blonde hair in a claw clip and blue-light glasses (S1) films a user-generated ad about a felt desk pad organizer in her tidy home-office desk. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the pad unrolling across the desk, a laptop and notebook being arranged on it, and cables tucking into its edge loops. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] My desk went from chaos to this with one purchase. The felt pad defines the whole space, my wrists stop hurting on the hard desk, and there are little loops that hold my charging cables so they stop falling behind the desk. Ten out of ten, zero regrets.</d>

**overall_soundscape:**

Quiet home office tone with a close phone-microphone voice. Felt sliding on wood and a cable clicking into a loop are audible. No non-diegetic music.

Round 1 — audio075 (pilot lineage) vs fal (10 characters)

Vitamin C serum — Black woman, 20s, bedroom vanity

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, warm bedroom lighting. A Black woman in her early twenties with braided hair and gold hoop earrings (S1) films a user-generated ad about a vitamin C face serum in her bedroom, sitting at a small vanity with a ring light off to the side. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the dropper pulling up orange-tinted serum and two drops being pressed into her cheek. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I have been using this vitamin C serum for a month and my dark spots are actually fading. Two drops in the morning, that is it. It is not sticky, it smells like nothing, and my foundation glides over it. My before and afters genuinely shocked me.</d>

**overall_soundscape:**

Quiet bedroom room tone with a close phone-microphone voice. A soft glass dropper clink and gentle skin tapping are audible. No non-diegetic music.

NC earbuds — South Asian man, 30s, home office

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, ordinary daylight from a window. A South Asian man in his thirties with a short beard and glasses (S1) films a user-generated ad about noise-cancelling wireless earbuds in his home office with a cluttered desk and a monitor behind him. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the charging case snapping open, one earbud being twisted into his ear, and a phone screen showing the noise-cancelling toggle. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] My neighbor has been renovating for three months and these earbuds saved my sanity. I put them in, turn on noise cancelling, and the drilling just disappears. The battery lasts my whole workday and calls sound so clear people think I am in a studio.</d>

**overall_soundscape:**

Home office room tone with a close phone-microphone voice. A case snapping open and a soft earbud click are audible. No non-diegetic music.

Air fryer — white woman, 40s, family kitchen

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright ordinary kitchen lighting. A white woman in her forties with shoulder-length blonde hair wearing a casual sweater (S1) films a user-generated ad about a compact air fryer in her family kitchen with kids' drawings on the fridge. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing frozen chicken tenders going into the basket, the timer dial being turned, and crispy tenders being shaken out onto a plate. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I have three kids and this air fryer is the only reason we eat dinner on time. Twelve minutes, no oil splatter, no preheating an oven. The basket goes straight in the dishwasher. Honestly it replaced my stove for weeknights.</d>

**overall_soundscape:**

Kitchen room tone with a close phone-microphone voice. The air fryer basket sliding, a dial ticking, and a faint fan whir are audible. No non-diegetic music.

Protein powder — Latino man, 20s, apartment kitchen

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, ordinary apartment lighting. A muscular Latino man in his mid-twenties wearing a tank top (S1) films a user-generated ad about a chocolate protein powder in his small apartment kitchen, gym bag visible on the counter. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing a scoop of powder dropping into a shaker bottle, water pouring in, and the bottle being shaken. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I have tried like ten protein powders and they all taste like chalk except this one. One scoop, water, shake it, done. It actually mixes with no clumps and it sits fine in my stomach. Twenty five grams of protein and it tastes like chocolate milk.</d>

**overall_soundscape:**

Apartment room tone with a close phone-microphone voice. Powder scooping, water pouring, and a shaker ball rattling are audible. No non-diegetic music.

Daily planner — East Asian student, dorm desk

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, soft desk-lamp lighting in the evening. An East Asian woman in her late teens with straight dark hair and a college hoodie (S1) films a user-generated ad about a daily productivity planner in her dorm room desk with textbooks stacked beside her. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the planner flipping open, a pen checking off boxes, and a filled-in weekly spread. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] My brain does not work without this planner anymore. Every morning I write my top three tasks and time block my classes. It has a habit tracker so I can see my streaks, and crossing things off is so satisfying. My grades literally went up this semester.</d>

**overall_soundscape:**

Quiet dorm room tone with a close phone-microphone voice. Paper pages flipping and a pen clicking are audible. No non-diegetic music.

Workout leggings — red-haired woman, 30s, living room

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright natural living-room light. A white woman in her early thirties with curly red hair in a ponytail wearing athletic wear (S1) films a user-generated ad about high-waisted workout leggings in her living room with a yoga mat rolled out on the floor. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the waistband being stretched, a deep squat to show they are not see-through, and a side pocket holding a phone. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] Okay these are the leggings I keep getting asked about. They are squat proof, I checked in every mirror. The waistband does not roll when I sit, and there is a pocket that actually fits my phone. I own four pairs now and I am not sorry.</d>

**overall_soundscape:**

Living room tone with a close phone-microphone voice. Fabric stretching and a light slap of the waistband are audible. No non-diegetic music.

Cold brew maker — Black man, 50s, kitchen

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, morning kitchen light. A Black man in his fifties with gray in his beard wearing a plaid shirt (S1) films a user-generated ad about a cold brew coffee maker in his kitchen counter next to a window. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing coffee grounds being spooned into the filter, water being poured, and cold brew streaming over ice in a glass. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] I was spending six dollars a day on cold brew until my daughter got me this. You add the grounds at night, put it in the fridge, and by morning it is smooth, never bitter. One batch lasts me four days. It paid for itself in two weeks.</d>

**overall_soundscape:**

Kitchen room tone with a close phone-microphone voice. A spoon scraping grounds, water pouring, and ice clinking in a glass are audible. No non-diegetic music.

Dog chew toy — red-haired man, 20s, couch + retriever

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, ordinary living-room lighting. A red-haired man in his late twenties with freckles sitting on a couch with a golden retriever (S1) films a user-generated ad about an indestructible rubber dog chew toy in his living room, the dog resting its head on his lap. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the dog chewing the toy, the toy being flexed between two hands, and peanut butter being smeared into its grooves. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] This dog has destroyed every single toy I have ever bought except this one. Eight months and not a single tooth mark. I stuff peanut butter in it and I get a whole hour of silence. If your dog is a shredder, this is the one.</d>

**overall_soundscape:**

Living room tone with a close phone-microphone voice. Soft rubber chewing sounds and a dog's contented huff are audible. No non-diegetic music.

SPF stick — Middle Eastern woman, 30s, car seat

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, bright daylight through a car windshield. A Middle Eastern woman in her thirties with wavy dark hair and sunglasses pushed up on her head (S1) films a user-generated ad about a clear SPF 50 sunscreen stick in the driver's seat of her parked car. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the stick being swiped across her cheekbone and blended with fingertips, leaving no white cast. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] PSA because I found the sunscreen that does not ruin makeup. It is a clear stick, SPF fifty, and I keep it in my car so I reapply at red lights. No white cast, no greasy shine, it feels like nothing. Dermatologist approved and it is under twenty dollars.</d>

**overall_soundscape:**

Parked car interior tone with a close phone-microphone voice, faint distant traffic outside. A soft cap click and skin swiping are audible. No non-diegetic music.

Desk converter — East Asian man, 30s, home office

ours — audio075 LoRA (bf16, 345f)
fal — minimax/h3 t2v (768P, 10s)
prompt
**integrated_multimodal_description:**

[Shot 1] Vertical smartphone UGC video, handheld phone-camera framing with slight natural shake, ordinary home-office lighting. An East Asian man in his mid-thirties wearing a zip-up hoodie (S1) films a user-generated ad about a clamp-on standing desk converter in his home office with a dual-monitor setup. The creator speaks directly to the camera with natural expressions and small hand gestures, intercut with close-up product shots showing the desk riser being raised with one hand, monitors lifting smoothly, and him typing while standing. Casual, authentic, non-professional look typical of social media content. The creator (S1) says, <d>[English] My back was wrecked from sitting ten hours a day, so I tried this desk converter. One pull and my whole setup lifts with me. It holds both monitors, does not wobble when I type, and took five minutes to clamp on. My lower back pain is basically gone.</d>

**overall_soundscape:**

Home office room tone with a close phone-microphone voice. A gas spring hiss, a soft clunk as the riser locks, and keyboard typing are audible. No non-diegetic music.