XAVI retweeted
opus 5.5 edited this entire STREET INTERVIEW ad on its own
this is what AGENTIC EDITING looks like
a "change my mind" vox pop for a protein bar brand that doesn't exist.
all i did was approve the script, the frames and the final cut
what it handled on its own:
- 22 AI takes, each trimmed 0.18s after the last word so it moves at real street pace
- one take had a hidden hard cut at exactly 0.50s. it found it frame by frame and started the shot after it
- the 3-way split of everyone tasting it? stacked in code from 3 separate takes
- captions lifted above the sign so you never read two things at once
the part that still needed me: taste.
which shot felt fake, which joke actually landed
RT + reply 'STREET' and i'll send you the full workflow
(must follow so i can DM)
XAVI retweeted
everyone should try PIXAR UGC for yapper ads
a real-looking girl in a real kitchen, slicing apples, ranting about recipe bloggers
except her face is a pixar character
everything else is a normal phone video, so your brain files it as a real creator... until the eyes hit you
quick workflow:
- took one frame from a real kitchen yapper and turned her into a new girl with a half-pixar face (nanobanana pro, free on flow)
- split the 59-word rant into 3 clips of 10s in omni, each one starting on the last frame of the one before, so the cuts land on the same pose
- locked the voice in every prompt: slow atlanta drawl, every word fully pronounced
- claude code cut every pause, sped it up 1.15x and added the captions + hook sticker
the final video cost 45 flow credits
under $1 for an ad that looks like a creator filmed it on her lunch break
XAVI retweeted
I have the CHEAPEST ai video provider out there
everyone's gatekeeping this MCP
highlights:
seedance at $0.086/s
google omni 1.1 flash at $0.02/s
& more..
idk why no ones talking about it
cheaper than
higgsfield (by 40%)
Kie
& Sume
RT & comment "MCP" and i'll dm u the link
XAVI retweeted
this AI Ad is 100% CODE
no image gen. no video gen. $0 spent on generation
opus 5.5 built the whole thing: the phone, hinge, the rizz screens, the keyboard, every notification, even the music
if you have an app, this is the cheapest way to make a promo or a demo video. no screen recording, no editor, no actors
quick workflow:
- research real conversations first (i pulled hinge openers, "two truths and a lie" tested best so it became the hook)
- opus writes the phone as one html page where every element is set by the second
- headless chrome steps through it and saves 1,170 frames
- message pops, keystrokes, the screenshot flash all get sounds coded onto the same cues
- only outside media: one stock photo and the app's real icon
RT + reply 'CODE' and i'll send you the full workflow
(must follow so i can DM)
OPUS 5.5 just made an entire animated ad by itself for $0
no image generation. no video generation. no editing
completely one-shotted by OPUS 5.5 inside claude code
i tested it on RYZE mushroom coffee with a single prompt, and then i just waited
what it did on its own:
- researched the product, the audience and the angle
- wrote the script and built the storyboard
- pulled the real product images straight from the web
- produced the voiceover and the captions itself
- rendered the finished ad, ready to post
you don't need a stack of paid AI tools anymore. opus 5.5 handles the whole production end to end
RT + reply 'OPUS' and i'll send you the complete workflow
(must follow so i can DM)
XAVI retweeted
these STICKMAN ads have been performing better than any other format on organic
it only makes sense to run them as ads
dirt cheap to make
the simple look is weirdly pleasing to the eye, especially when your feed is packed with over the top high production ads
quick workflow:
- write the script first, one simple visual metaphor carrying the whole story
- doodle frames on Nano Banana Pro, black marker on white with orange accents only
- animate the key beats in Kling 3 at 480p and keep the rest as stills with cutaways on top
- a designed narrator voice on Gemini TTS, dry and a little self-deprecating
- Lyria for a light score that lifts the moment the product shows up
- karaoke captions, one yellow word at a time
if you're still paying for polished ugc and skipping this, you're leaving the cheapest win in paid social on the table
RT + reply 'STICK' and i'll send you my full workflow
(must follow so i can DM)
documentary ads are the most UNDERRATED format in ecom right now
this one opens on a real fact: the "coffee break" was a 1952 ad campaign
then it moves through the decades.
black and white archive photos, then pencil sketches, then modern colour, until the product shows up as the fix
quick workflow:
- tore down a winning supplement ad frame by frame to steal its structure
- 31 nanobanana pro stills with a separate look per decade
- only 4 kling clips, the rest are stills with slow push-ins cut to the voiceover
- gemini TTS narrator, lyria score, karaoke captions synced word by word
RT + reply 'DOCU' and i'll send you the full workflow, every prompt, every fix
(must follow so i can DM)
XAVI retweeted
OPUS 5.5 just made an entire animated ad by itself for $0
no image generation. no video generation. no editing
completely one-shotted by OPUS 5.5 inside claude code
i tested it on RYZE mushroom coffee with a single prompt, and then i just waited
what it did on its own:
- researched the product, the audience and the angle
- wrote the script and built the storyboard
- pulled the real product images straight from the web
- produced the voiceover and the captions itself
- rendered the finished ad, ready to post
you don't need a stack of paid AI tools anymore. opus 5.5 handles the whole production end to end
RT + reply 'OPUS' and i'll send you the complete workflow
(must follow so i can DM)
XAVI retweeted
opus 5.5 is CRACKED for ai drama ads
pocket fm runs 40 minute story ads that stay live for 4+ months
i pulled their 199 longest runners and turned the whole format into a claude skill for ecom products
drop a product link and it hands you a 3-6 min drama script,
plus every image and seedance prompt to make it
what it actually does:
- digs through reviews, comments and deal-site threads, then builds a bank of real customer lines so the dialogue sounds like your actual buyer
- remixes premises that already ran 100+ days (the secret heir, the family mistake, rejected then chosen) into your buyer's world
- every concept runs on life force 8 stakes, so the story holds up even with the product taken out
- your product slides in softly once they're already hooked
- storyboard first, then paste-ready nanobanana pro stills and seedance 2.5 clips with character lock
tested it on RYZE mushroom coffee
and wrote a thanksgiving taste-test drama with zero health claims
RT + reply 'DRAMA' and i'll send you the skill + the full playbook
(must follow so i can DM)
HOLY SH*T...
i just cracked organic traffic at scale
built a reaction hook generator from my personal SWIPE file, then mined 1,000+ REAL viral reaction clips (with their actual stats) to make it hit even harder
upload your product link, it researches it, locks a sharp audience identity, and one-shots the entire reaction clip using seedance 2.5
quick workflow:
- give it a product link or a plain description
- it locks a sharp identity, "dog dads," not "pet owners"
- matches 1 of 6 proven hook formats, for physical products it defaults to the classic "walked into the store and found this" angle
- writes 5-8 on-screen hook lines to test against each other
- one-shots the whole reaction clip in a single seedance 2.5 generation
best used to hit organic traffic: dropshipping, affiliate offers, promoting your app
RT + reply 'HOOK' and i'll send you the skill
(must follow so i can DM)
XAVI retweeted
HOLY SH*T...
i just cracked organic traffic at scale
built a reaction hook generator from my personal SWIPE file, then mined 1,000+ REAL viral reaction clips (with their actual stats) to make it hit even harder
upload your product link, it researches it, locks a sharp audience identity, and one-shots the entire reaction clip using seedance 2.5
quick workflow:
- give it a product link or a plain description
- it locks a sharp identity, "dog dads," not "pet owners"
- matches 1 of 6 proven hook formats, for physical products it defaults to the classic "walked into the store and found this" angle
- writes 5-8 on-screen hook lines to test against each other
- one-shots the whole reaction clip in a single seedance 2.5 generation
best used to hit organic traffic: dropshipping, affiliate offers, promoting your app
RT + reply 'HOOK' and i'll send you the skill
(must follow so i can DM)
XAVI retweeted
dropshipping demons are about to start PRINTING this Q4
while everyone is focussed on halloween, this guy is targeting veteran's day
found a page running AI UGC ads made entirely in Seedance 2.5
within 5 days, 9 videos posted, already at 55k followers. one clip alone pulled millions of views organic
quick breakdown:
- same broll on every single video, pretty sure it's lifted straight off a competitor's ad
- new hook every time, that's the only thing that actually changes
- trending audio slapped on top so the algo does the pushing
- comment a keyword for the product link instead of a direct link, boosts engagement and cvr
- link in bio catches everyone who doesn't bother commenting
running this at scale costs way less than people assume
RT + reply 'Q4' and i'll send you the full teardown on how to replicate this
(must follow so i can DM)
XAVI retweeted
made this motion design for @whop
it has ZERO ai video or image
the whole thing was built inside claude code
using a single prompt and a link to the whop homepage
quick workflow:
- gave it one prompt and the whop(.)com link, nothing else
- broke down a reference ad frame by frame first for the shot rhythm and audio timing,
- then pulled the real screenshots, copy and the $4.6B stat straight off the site
- storyboarded 24 keyframes, then rendered every frame as a pure function of time so any single beat could get redone without touching the rest
built and synced its own soundtrack in python, timed to the same numbers as the visuals
RT + reply 'WHOP' and i'll send you the exact prompt + full workflow
(must follow so i can DM)
XAVI retweeted
omni is still UNDEFEATED for yapper content
nothing else touches how natural the talking-head performance comes out
only reason i still open kling at all cos of google being a b*tch with the policy filters. half my scripts get flagged before they even render
here is something i have noticed
black and other ethnic creators look MORE natural on omni than white ones do
my read on why: white AI voices got used so much they're recognizable now. you hear it and something in your brain goes "that's the voice"
ethnic voices and faces haven't hit that oversaturation wall yet, so it just reads as a real person
if i come across such clips on my home feed, there's no way im clocking it as AI unless im really looking hard
that's the actual bar right now, not "does it look good," but "would i personally get fooled by my own output"
RT + reply 'OMNI' and i'll send you the full 16-block prompt structure i use to get this locked in every time
(must follow so i can DM)
XAVI retweeted
I onboarded a new media buyer
Taught her how I run $400K+ per month in casino deposits off FB Ads
7 Concepts:
> 1. How to set up Keitaro so every campaign has clean tracking from day one
> 2. How to install pixel + CAPI so postbacks fire correctly in 25 minutes
> 3. How to structure campaigns so reg-to-dep stays consistent across GEOs
> 4. 15+ case studies of my tracked campaigns each doing over $30K+ per month (accounts I manage)
> 5. How to close any lander with 85%+ success rate ( cloaking + funnel skill )
> 6. How to get $5000 payouts approved fast from casino networks
> 7. How to do this completely faceless and with less than 1 hour per day on average of work ( BM farm + antidetect systems )
She already got her first Keitaro campaign live and tracking clean.
Comment "KEITARO" if you want ALL OF THIS that I gave her so that you can learn the same
If you comment, you get the following:
> 1. The full Keitaro Setup Cheat Sheet, my $15M+ tracking stack, FREE for the next 24 hours
> 2. 7-part reference covering installation, tracking, pixels, CAPI and PWA settings. usually $3K+ to access.
→ 20+ case studies of Keitaro-tracked campaigns doing $30K+ per month EACH
→ the exact postback setup generating $151K in 7 days on autopilot
→ how Parker did $570K in 3 months with one tracker setup
→ why campaigns with 6.5M impressions convert LESS than mine with 13K clicks
→ the CAPI configuration cutting wasted spend for free (send link to the cheat sheet)
→ the tracker crash course formula turning raw clicks into $10K–$100K in FTDs
→ how to hit $100K/month running campaigns that take minutes to set up
DELETING IN 24 HOURS
comment "KEITARO" and I'll DM it to you
**must be following + retweet
XAVI retweeted
i just built my own HORMOZI
the dave ramsey below ran off a single frame carrying the whole ad. this one goes further - 2 characters, a live crowd, and a full Q&A
watched his conference clips and realized the format is already an ad. someone in the crowd states a real revenue number then admits the bottleneck underneath it
the coach on stage answers in two sentences flat. the room full of people validates the whole thing.
quick workflow:
- generated both characters through GPT image 2 with separate anchor images, accents, and delivery styles
- every clip re-locks identity, outfit, and environment from those two source frames since Omni carries no memory between generations
- the crowd member's clips open on his exact reference pose so the cuts back and forth feel like one continuous shoot
- captions and end card in CapCut, everything else straight out of Omni
if you sell a biz op or coaching offer, this format has the objection-handling and social proof built into the structure itself
RT + reply 'MOZI' and i'll send you the full production doc
(must follow so i can DM)
the best AD FORMAT i've seen in months came from scrolling past dave ramsey on instagram
a call-in advice show. authoritative host behind a desk takes a live call from a woman who can't sleep,
diagnoses her problem on air, and recommends the fix with the kind of quiet confidence that makes you forget you're watching an ad
quick workflow:
- generated a single studio shot of the host at his desk and used it as the identity lock for all six clips
- the caller is audio-only through his headset, so there's no second face to keep consistent across cuts
- wrote full performance direction into every Omni prompt: the southern drawl, the listening beats where his mouth stays shut, the product pickup
- rendered all six clips and stitched in CapCut, nothing reshot
the format works because the host sounds like he's run this show for twenty years, and the caller's exhaustion is relatable
RT + reply 'DAVE' and i'll send you the full workflow
(must follow so i can dm)
XAVI retweeted
ad rejection spikes are still mad random.
it’s less about the creative, and still costs $100k–$200k/mo in FTDs when you can’t tell why you’re flagged.
i logged the full rejection pattern across my BM farms and antidetect setups.
like, RT + Comment “SPY” and i’ll DM you the GEO Spy Keyword Vault for free.
must be following or i can’t DM
XAVI retweeted
posted an AI UGC in spanish today, took 15 minutes, and I didn't write a single original word
pulled a viral organic rant that already worked, translated it to sound native instead of textbook, animated it, done
most people are burning creative budget testing new angles in the US and UK, the two most saturated and expensive ad markets on earth.
meanwhile the exact same proven winners haven't even touched markets sitting on billions of people and real spending power
quick workflow:
- find a proven winning ad, organic or paid, doesn't matter
- translate it for spoken cadence, not literal grammar, ask explicitly for how a native speaker actually talks
- lock a character reference once, face and wardrobe frozen across every clip
- generate with a model that can actually hold the accent, most of them default back to english under pressure
LATAM alone is 650 million people. brazil is 213 million with a growing middle class. india just crossed 1.4 BILLION. germany has the highest GDP per capita in the EU. russia's another 140 million+. scandinavia is small in population but sits near the top of the world in spending power per person
none of these markets are short on money or people. they're short on competition
RT + reply 'GEO' and i'll send you the full localization guide, translation, model picks, prompts, everything
(must follow so i can DM)
XAVI retweeted
full AI vlogs are possible now
this is exactly the kind of content an AI influencer runs on, and most people aren't even trying it yet
a full 30-second GRWM, shower to hair to makeup to a dress to walking out the door, and it ends with her looking straight into your eyes and toasting like you're the date.
one consistent character across the entire thing, entirely ONE-SHOTTED with seedance 2.5
quick workflow:
- generate 8 reference images, each one a different beat of the story, all locked to the same identity
- use those images as style and identity references only, never as literal start or end frames, this is what stops the small face-drift most GRWM attempts get killed by
- split the story into 2 clips, 5 hard cuts each, narrator voiceover matched precisely to every on-screen action, never lip-synced
-load clip 1 into clip 2 as a voice reference only, for consistency, not visual framing
this is the actual unlock for anyone trying to build an AI influencer.
you don't need a camera, a studio, or a real person. you need one locked identity and a story people actually want to watch
RT + reply “VLOG” and i’ll send you the full production doc including all the prompts
(must follow so i can DM)
seedance 2.5 is sick
you can literally become an influencer without camera
and don’t even get me started about how much those gym leggings brands pay for collabs to micro influencers 😭
full prompt:
CLIP 1, Getting ready (0:00–0:15)
use @image 1 (damp hair, no makeup), @image 2 (blush/lipstick), @image 3 (curling hair), @image 4 (fit-check dress) as style and identity references only, not as start or end frames.
0-3s: hard cut open, she stands at the bathroom vanity in a white robe, hair damp and pushed back, gently patting it dry with a small towel, no makeup yet, gaze on her own reflection in the mirror.
3-6s: hard cut, she smooths on a light layer of lotion and dabs concealer under her eyes with her fingertip, focused on the mirror.
6-9s: hard cut, she sweeps blush across her cheeks and applies lipstick, quick confident motions, still focused on the mirror.
9-12s: hard cut, she curls a section of her hair with a curling iron, one hand holding it near her face.
12-15s: hard cut, now in a fitted cocktail dress with soft curled hair, she turns side to side in a full-length mirror checking her look, small satisfied smile.
voiceover (her voice, narrating over the visuals, not lip-synced, she is never shown speaking on camera): "Fresh out the shower, hair's still damp. Skin prep first, lotion, then a little concealer. A little blush, and this lipstick I've been obsessed with. Now let's get these curls going. Okay... this dress might be the one."
audio: the voiceover plays over quiet bathroom ambience for the first four beats, shifting to quieter bedroom ambience for the last beat, no music, no on-screen dialogue.
style: candid mobile camera quality, natural color, effortless aesthetic raw, hard cuts between every beat, no cross-dissolves, soft bathroom daylight through the first four beats, warm lamp light for the fit-check beat, deep focus, no artificial blur.
constraints: preserve identity across all five beats using the reference images for face, hair, and build only, do not use any reference image as a literal start or end frame, no readable text, no logos, no extra characters, no morphing hands, no revealing or suggestive framing.
XAVI retweeted
yapper ads are still the highest ROI format for our clients
so i gave opus 5 my exact yapper workflow and it packaged it into a skill
-avatar prompts matched to your target audience
-copy that doesn’t sound like AI wrote it
-gpt image 2 first-frame prompts
-omni video prompts, character locked
rt + reply "opus" and ill send it to you (must follow so i can DM)
XAVI retweeted
CLAUDE FABLE 5 IS BACK
their most capable public model, a full tier above Opus 4.8
Fable is built to run long, complex tasks for hours without you touching it, which means the entire AI content pipeline can now run itself start to finish, unattended
you can GO VIRAL every single time on autopilot with Fable, here's how it works:
- fable 5 builds the full persona: name, backstory, attitude, one niche, one mission
- generates a face that doesn't exist, a composite, never a real person
- locks that face with a character-lock line so the exact same person shows up in every single post
- scans what's going viral in your niche and rebuilds it in your character's voice
- generates every first frame so the character looks identical across the whole feed
- kling 3 animates them into short clips with real expressions and delivery
- schedule everything and let it run
RT + reply 'FABLE' and i'll send you the full build so you can run it right away (must follow so i can DM)









