Robert Rock retweeted
Help here: give.rarevillage.org/campaig…
Robert Rock retweeted
i see marketers churn from stefan brain every day...
it's the SAME reason every time...
AND IT KILLS ME...
they give it a 50 page "context" doc they made...
they wonder why the ads their are slop...
they cancel 😑
BUT they are NOT using AI like it's 2027...
they are stuck in 2023...
"CONTEXT" IS KILLING YOUR COPY
@StefanGeorgi just spent 1h 45m breaking this down live… plus exactly how he writes ads now (one of them spent $738k in 3 days off 3 prompts)
I watched every minute so u don't have to 👇
here's the paradigm shift first:
in 2023 and 2024 we all "context maxxed"
brand bibles. avatar matrices. VOC dumps. 40 page briefs
it made sense. models were dumber...
BUT now they're trained on ALL of the internet…
they already know SO MUCH about ur customer...
so every extra page u add isn't helping anymore...
it's steering...
the new game is:
LESS context -> BETTER swipes
here's how he actually does it:
1. CONTEXT IS THE PROBLEM (2:00)
he ran an experiment live
same product (a ~$25 amazon cupping massager)
same ad
same prompt
the only thing he changed was the context:
- a 15 page AI brand brief
- a 7 page version
- a 1.5 page product sheet
- just 2 reference ads. nothing else
the 15 pager wrote "three synchronized clinical modalities… targeted 660 nanometer red light"
for a massager lol
his reaction (12:12): does a woman buying a suction massager give a sh*t about negative pressure vacuum?
NO
but the 2nd ad version sounded like an actual woman:
"if you wake up with heavy legs, see deep red sock lines dug into your ankles by noon… it is not because you are lazy or eating wrong"
his verdict (19:08): "the two reference ads wins"
why it happens (10:11 and 23:03):
AI treats EVERYTHING u give it as important
the brief had a "mattress effect" metaphor he didn't even like… so the scripts kept talking about mattresses
it said "clinical big sister. no hype, no shouting"… so the yapper came out weirdly polite
it said "we do not promise suction melts fat like butter"…
so good luck ever getting a butter line when u want one
YOU forget the line is there but the model never does
his words (24:22): this is "like actually 95% of getting better outputs"
and this is the churn thing from the top (24:58)
it's not a stefan brain problem.
claude... chatgpt... ANY LLM does it
a claude-written context doc is full of claude-isms…
so of course everything it writes sounds like claude
the sports car thing (20:06):
a model built for marketing steers like a lambo
barely touch the wheel and it turns
a 40 page brief is yanking the wheel on a car that already knows the road
how his projects actually look (1:19:28):
2 to 3 great ads. not 12
if it's missing something like the price or the offer
write 5 sentences urself
STOP letting AI write u another brief
he doesn't even use voice of customer docs anymore (1:33:08)
VOC docs make it repeat the same pain points over and over
2. SWIPE LIKE AN ARTIST (21:03)
"reference ads are king"
here's the exact process behind the $738k ad (21:16):
1. found an ad he liked from a DIFFERENT brand
2. "we're basically just going to swipe this ad but for our brand. outline it for me"
3. read the outline. "swipe this one even closer. this is what the hook should be"
4. maybe one more note
5. script done. dropped it into a batch of 30 creatives. it became the banger
under 10 min with no manual writing
3. SWAP THE FORMAT (26:11)
same 2 yapper ads in the project. new swipe: a song ad from another brand
it wrote it like a yapper first. one note: "do another version where it's a song ad"
next note: match the swipe closer, new situation, new location, less classic rhyme
side gem (30:51): a lot of the song ads that are winning right now DON'T even rhyme
then he took that same song ad and turned it into a yapper for a totally different brand (acne)
copy was fine. voice was off. one line fixed it (39:12):
"imagine it's the same woman as from those ads talking. same voice, same writing style"
4. ONE WINNER → 100 ADS (46:32)
"give me 20 more concepts that would follow a similar structure. what would change would be the spokesperson, the situation, and elements of the story"
first batch was meh. and here's where ur human brain still matters (48:55):
the winner ran on ONE emotion. jealousy. running into the ex and the woman he left her for
so he made it keep the emotion and only swap the scene:
farmers market. dog park. daughter's engagement party. airport sky club. pilates studio. whole foods hot bar. college tailgate
then: "do concept 20. now 19. now 18…"
(50:46): "I don't know how you would not write like 40, 50, like 100 ads basically in like an hour or two"
small gem (47:46): when AI dumps a giant outline on u, tell it "a few sentences per concept max"
less is more is EVERYWHERE
5. YAPPER SECRETS (40:59)
secret 1: write how ppl actually talk
he has yappers where 7 sentences in a row start with "and"
secret 2: the creator is half the ad (41:47)
the $738k script had already been tested… and got no traction
then they gave the exact same script to the brand's best yapper creator. she re-recorded it
it spent $270k that thursday. then $200k to $300k every day since
80% of that brand's top yappers are now that one woman
so before u say "yappers don't work in my niche"… test more than one creator. and in his experience women beat men almost every time
the 3 transitions (1:00:44). almost every yapper he writes goes one of 3 ways:
- she's at an appointment, apologizes for her body, the expert explains what's really going on
- she's on the phone with a friend who happens to be an expert
- she's at brunch with a friend who happens to be an expert
another gem here... pick ONE lane (1:02:11): cellulite OR lymphatic drainage. not both
the split test that'll make ppl mad (1:09:17):
landing page for a demo that's 52. they tested a woman in her late 20s vs 40s vs 50s vs 60s
late 20s smoked it. 50 to 60 cent RPU lift
ppl don't want to see who they are today. they want to see who they are in their head
same with copy. write it how a 25 year old talks, have a 51 year old say it. still hits
6. HOOKS (54:25)
the money prompt (55:15). steal this word for word:
"give me 30 new ad hooks and concepts for [brand]. problem aware. avoid generic ones that every LLM would give. as you evaluate initial ideas, tell yourself to make them weirder, scroll stopping, include unique situations when relevant, and to just be 10x more creative and 10x less generic"
the hook he rated 10/10 (57:30):
"my husband tried to put his hand on my thigh while we were driving home from dinner on Saturday, and I literally swatted his hand away like a stray wasp"
why it works: super relatable. a real situation. insecurity baked in. one weird vivid image (the wasp)
AI cliches he cuts on sight (56:51): "we need to have a serious talk." "we need to talk about." "you need to see this"
and shorter beats clever (1:04:54)
AI wrote: "if you've ever set an alarm for 5:45 in the morning on a romantic weekend getaway just to spackle concealer over your jawline…"
he wanted: "so I almost broke up with my boyfriend last month"
open loop. don't explain. make them need the next line
7. STEM HOOKS (1:11:26)
this is the one most ppl will miss
ask AI for "20 hooks" and u get mid stuff
why? a hook only works inside a FORMAT. a great yapper hook is a bad song ad hook
so he makes "stem hooks"… like a stem cell
one raw situation that can grow into any format
one stem: fake tan rubbing off on a guy's white towel the morning after
casual yapper: "I almost climbed out of a third-story bathroom window last weekend just so a guy wouldn't see what my fake tan did to his white Ralph Lauren towel"
authority yapper: "as a cosmetic chemist, please stop believing that a fake tan that stains your bath towels is just part of the process"
faceless AI voiceover: "the real reason your self-tanner wipes off on towels no matter how long it dried"
same seed. 3 different ads
the only difference between a casual and authority yapper? confidence. one is discovering it. one already knows
20 good stems x 6 formats… his words (1:40:08): "you literally write thousands of ads from like this process"
8. DON'T SNEAK THE CONTEXT PILE BACK IN (1:27:22)
once ur rolling u'll want to save ur rules so u don't repeat urself
fine. keep it tiny
his example: "never reveal the product until this part" sounds smart
3 months later ur ads won't mention the product and u forgot u ever said it
rules pile up, start fighting each other, outputs get worse
how they're thinking about memory at SB: say a rule once and it's soft. repeat it and it hardens. start fighting it and it softens again
9. GET THE PLANE OFF THE GROUND (53:43 and 1:35:20)
the first reference ad is where u spend the time. he spent about 30 min making his first yapper great
after that it's next one. next one. next one
his analogy: getting the plane off the ground takes the most energy. then it coasts
the whole thing in 5 lines:
- delete the context doc ur proud of
- put 2 to 3 great ads in a project. that's ur context
- swipe a winner from another brand. outline first
- remix the situation, but keep the emotion
- be the thinker. let AI be the hands
AI just turned the swipe file into a remix machine
if u only watch one part, watch the first 25 min. it'll change how u write with AI
or just send this to ur agent ;)
Want to know how I can write like 100+ good ads in a single day using AI? For pretty much any format imaginable?
Well, here you go.
I did an epic live call today where I broke it down. Had like 400 people on at one point.
If you watch even the first 20 mins or so of this, I’m willing to bet you’ll get significantly better outputs with AI than you are right now. Vid is long-form, but it’s also saucy AF. Hope you enjoy :)
Robert Rock retweeted
Want to know how I can write like 100+ good ads in a single day using AI? For pretty much any format imaginable?
Well, here you go.
I did an epic live call today where I broke it down. Had like 400 people on at one point.
If you watch even the first 20 mins or so of this, I’m willing to bet you’ll get significantly better outputs with AI than you are right now. Vid is long-form, but it’s also saucy AF. Hope you enjoy :)
Robert Rock retweeted
Sadly surgery didn’t go well and tumour too big now. You are going to try and get me into the hospice tomorrow. From there I think it’s probably a few days left. I’ve loved it all and have no regrets. Achieved everything I ever wanted with a perfext wife and 2 kids,
Robert Rock retweeted
this Seedance 2.5 character method is STUPIDLY good.
(the full method is explained in the quoted post)
here’s the exact JSON prompt behind this one:
{
"output": {
"aspect_ratio": "9:16",
"style": "Photorealistic vertical indoor front-facing selfie.",
"image_quality_tier": "Clear face, lightly softened skin texture, mild video compression and softer background.",
"master_description": "A young adult woman of mixed West African and Latin American appearance has warm light-medium caramel skin in a brightly lit selfie. Her tapered oval face has full cheeks, a narrow rounded chin, dark almond eyes, long curled lashes, shaped brows, a small straight nose with rounded tip, full glossy rose-pink lips and a tiny mark below the lip. Dense near-black rope curls reach her shoulders beneath fine swooped baby hairs. One large thin gold hoop hangs at the viewer's right ear. She wears a pale mint-ivory halter with white scalloped lace edging its deep V and a front tie. Her arm at the viewer's right extends toward the camera. A cream dog and dark chairs sit behind her in a dim gray room."
},
"subject": {
"overview": "Young adult woman, visually early twenties, feminine presentation. She fills the vertical frame, face near center and looking into the lens. Warm light-medium caramel complexion, peachier on face than neck and arms. Medium frame, rounded shoulders, average upper-arm fullness and fuller bust relative to visible torso. Upper body only.",
"apparent_ethnicity": "Mixed West African and Latin American appearance as a visual impression: caramel-brown skin, dark curls, almond eyes, full lips and rounded nose tip.",
"skin_color": "Light-medium caramel-brown with warm golden-peach undertones. Approximate display colors: forehead #E4A790, cheek #D18E7B, nose #EBAC96, arm #D0957C. Face is lighter and pinker than arm; neck is shaded. Values include frontal light and processing.",
"distinctive_traits": [
"Warm light-medium caramel skin, peach-golden in frontal light.",
"Soft oval face, full cheeks and narrow rounded chin.",
"Dark almond eyes looking into the lens.",
"Long curled upper lashes.",
"Dark groomed brows with gentle arches.",
"Small straight-bridged nose with rounded tip.",
"Full rose-pink lips, fuller lower lip and gloss.",
"Dense near-black rope curls with shoulder-level ends.",
"Fine curved baby-hair swoops at the forehead.",
"One large thin gold hoop at viewer's right ear.",
"Pale mint-ivory lace-trim halter top with front tie."
],
"body": "Medium frame, rounded shoulders and moderate upper-arm fullness, without sharp muscle definition. Shoulders span about twice head width. Neck is long and narrow with visible collarbones. The arm on the viewer's right is enlarged by lens proximity. Bust is fuller than upper ribcage and supported by the top; only upper contours show. Lower torso is cropped.",
"hands": "Hands are outside the frame; no fingers or nails are visible.",
"body_skin": "Neck, upper chest, shoulders and arms are warm caramel-brown with satin highlights. Tiny dark speck on outer viewer's right upper arm. No visible tattoos, scars or prominent veins."
},
"face": {
"structure": "Oval face tapering downward. Widest across upper cheeks; moderately broad curved forehead; softly full cheeks, not hollow. Jaw narrows gradually with a soft angle. Chin is narrow, rounded and not pointed. Cheekbones form a gentle contour.",
"proportions": "Face length is about 1.6 times cheek width. Forehead and upper cheeks are similar in width; jaw is narrower. Lower face third is slightly shorter. Eyes sit just above midpoint; nose centered between eyes and mouth. Mouth is about 1.4 times nose-base width. Nearly frontal face with mild selfie perspective.",
"asymmetry": "Viewer's right eye is slightly more open; curl partly interrupts the viewer's left outer eye and cheek edge. Viewer's left brow sits slightly higher. Viewer's right cheek is more exposed. Mouth corners are subtly uneven, the viewer's left corner marginally higher. Nose has no strong deviation.",
"eyes": "Dark brown almond eyes, horizontally elongated and medium-sized. Outer corners rise slightly, about 2 to 4 degrees; inner corners softly pointed. Natural spacing, about one eye-width between inner corners. Shallow upper-lid crease. Eye on the viewer's right is slightly rounder and more open. Deep brown irises, round pupils, warm off-white sclera and small frontal catchlights. Upper lashes are long, dark, curled and dense with separated tips; lower lashes short and sparse. Mild under-eye shadow, no pronounced bags. Subtle warm lid shading and fine upper-lash definition.",
"eyebrows": "Dark brown to near-black, medium-thick and groomed. Fuller softly squared inner ends, low smooth arch at outer third and tapered tails. Hairs are blended, not a solid block. Brows sit close to eyes; viewer's left arch is slightly higher, brow on the viewer's right flatter and longer. Color matches hair.",
"nose": "Small-to-medium nose with straight narrow bridge and softly rounded tip. Narrow highlight along bridge. Moderate rounded alar base, near the distance between inner eye corners. Small dark nostrils, viewer's left more shaded. Mild tip highlight and warm shadow below. No piercing.",
"lips_and_mouth": "Closed relaxed mouth with mild composed pout, no smile. Full lips; lower lip is rounder and about 1.3 times upper-lip volume. Soft cupid's bow, muted rose-brown edge and pink-rose gloss. Bright glints at lower-lip center and smaller upper lip. Short-to-medium philtrum; corners nearly level with slight asymmetry. No teeth.",
"ears": "Viewer's right ear partly visible between cheek and curls; lobe carries the hoop, upper ear is covered. Viewer's left ear is concealed. Do not add or mirror a second earring.",
"expression_and_gaze": "Calm, mildly posed expression. Brows neutral, lids naturally open, cheeks relaxed and lips together with slight forward tension. No teeth or open mouth. Gaze straight into the lens."
},
"skin": {
"texture_by_zone": "Forehead smooth with soft central sheen; temples mostly hidden. Under-eyes softly shadowed. Nose has the brightest bridge and tip highlight. Cheeks are smooth with faint texture and peach-pink blush. Around mouth even; chin smooth with tiny mark; jaw softly shaded. Neck has faint folds at base. Chest, shoulders and arms have satin highlights. Light camera smoothing remains visible.",
"tone_shifts_by_zone": "Forehead and nose are lightest and peachiest (#E4A790 to #EBAC96); cheek is warm peach-brown (#D18E7B). Opposite cheek darkens at hair. Neck and chest are warmer and darker; upper arm is muted caramel (#D0957C). Frontal light lifts face and shoulders; cool room light neutralizes edges. Keep base complexion unchanged.",
"marks": "Tiny dark beauty mark below lower lip, slightly viewer's right. Separate tiny brown speck on outer viewer's right upper arm. No clear freckles, acne clusters, scars, tattoos or large patches.",
"shine": "Medium shine on forehead and nose; low-to-medium on cheeks, neck and arms. Nose has a small bright reflection; cheek highlights are soft, not oily. Lips have high gloss.",
"makeup": "Light base, warm peach-rose blush high on cheeks, subtle neutral lid shading, fine upper-lash definition, groomed brows, long curled upper lashes, small nose highlight and rose-pink gloss. No heavy contour."
},
"hair": {
"style_and_length": "Very full dark twisted curls made of narrow rope-like sections and corkscrew ends. Dense rounded volume at temples and sides. Length reaches shoulders and upper chest with ends at varied heights. Crown is slightly raised into a small loose tuft at top-back.",
"parting_and_hairline": "No broad exposed part. Hair is directed back with a small concealed central division. Curved, softly irregular hairline; fine baby hairs form swoops across central forehead and shorter curves at temples. Little scalp shows.",
"color": "Near-black dark brown (#261F21), with espresso ridges (#3C3234); frontal light makes some ridges look brown-gray.",
"texture_and_finish": "Compact springy twists with irregular spiral sections, mostly pencil-width or narrower. Raised ridges reflect light; gaps stay dark. Dry-to-satin, dense, little visible frizz.",
"loose_strands": "Fine laid curves at forehead, short curled temple wisps and separate twist tips around shoulders. One narrow curl falls beside viewer's left cheek. Keep irregular ends.",
"placement": "Viewer's left curtain falls past jaw toward shoulder, hiding ear and cheek edge. Viewer's right bulk spreads beside cheek and over shoulder, partly hiding chair. Visible ear and hoop sit in a gap; hair overlaps top straps.",
"facial_hair": "No facial hair."
},
"pose": {
"head": "Head tilted about 3 to 5 degrees toward viewer's right. Face nearly frontal, turned 3 to 6 degrees toward viewer's left. Chin near level with slight downward angle.",
"torso_and_shoulders": "Torso mostly faces camera, subtly turned viewer's left. Shoulder line slopes down toward viewer's right by 3 to 5 degrees. Shoulder on the viewer's right is closer and larger. Shoulders relaxed and rounded; collarbones visible. Upper torso leans slightly forward.",
"arms": "Viewer's right upper arm descends diagonally from the shoulder on the viewer's right toward lower edge on the viewer's right, enlarged by proximity. It continues forward beyond crop; forearm and hand are outside frame holding camera. Viewer's left arm is mostly hidden, with a small section at the lower viewer's left.",
"hands_and_fingers": "No hands or fingers visible. Camera-holding hand stays outside frame; do not show a device.",
"body_language": "Relaxed deliberate selfie pose, angled shoulders and face toward lens; composed rather than animated."
},
"clothing": [
{
"garment": "Pale mint-ivory sleeveless halter top, fitted over bust and upper torso. Very pale green with cream cast, not pure white; deep plunging V exposes upper chest.",
"fabric": "Light stretch knit with fine ribbing, mostly matte with soft sheen on folds; opaque and without visible print.",
"construction": "Wide halter straps pass behind neck, leaving shoulders bare. White scalloped lace edges the deep V. A small center-front fabric knot has two long narrow tie ends. Curved seams outline bust cups; a horizontal under-bust seam crosses torso.",
"fit_on_body": "Close fit follows upper torso and bust. Deep V tapers to center tie; lace has fuzzy scallops. Small creases gather at under-bust and lower torso. Viewer's strap on the viewer's right is broad and clear, one on the viewer's left rises diagonally."
}
],
"accessories_and_objects": [
{
"item": "One large thin gold-tone hoop earring at the viewer's right ear.",
"details": "Thin polished gold-tone hoop, about 5 to 6 cm across, with bright pale-gold and warmer yellow reflections. No stones.",
"placement": "At viewer's right earlobe in a gap between jaw and hair; lower curve reaches upper-neck level. Ear on the viewer's left is hidden."
},
{
"item": "Fine gold-tone necklace with a small pendant.",
"details": "Very fine gold chain with a tiny bead-like glint and small curled openwork pendant; no lettering.",
"placement": "Rests across collarbone and upper chest; small pendant hangs near upper sternum, slightly viewer's left."
},
{
"item": "Small dog in the background.",
"details": "Small fluffy cream-white dog, beige around head and ears, with dark eyes and nose. Standing on all four paws, facing camera, softly blurred.",
"placement": "Rear, viewer's right, around 68 to 80 percent frame height, partly behind foreground arm and below chair legs."
}
],
"environment": {
"location": "Dim indoor residential dining area with gray-green walls, gray ceiling and pale-gray floor.",
"objects_by_position": "Gray ceiling with diagonal junction at upper viewer's left. Viewer's left: dark vertical cabinet-like panel, dim ledge with a small reddish-brown object and lower panel edge. Viewer's right: pale-top table and dark rounded-back chairs with slim wooden legs, partly hidden by hair. Plain gray wall has a brighter low patch. Small cream dog stands on pale-gray floor; darker baseboard along wall.",
"depth_and_blur": "Subject is close and clear; furniture and dog are several feet behind and soft, especially the dog.",
"edges_and_corners": "Top: gray ceiling; upper viewer's left diagonal ceiling plane, upper viewer's right plain wall. Viewer's left: dark cabinet and reddish object; viewer's right: dining furniture. Bottom crops torso; arm on the viewer's right exits lower viewer's right. Dog and floor remain at the lower viewer's right."
},
"lighting": {
"sources": "Strong frontal light near camera, slightly viewer's left and above eye level (11 to 12 o'clock), like a selfie light. Glossy highlights and soft shadows; dim cool ambient room light.",
"on_face": "Forehead, nose, upper cheeks and lip center are brightest. Soft shadows under nose, lip, chin and jaw. Hair shades viewer's left temple and cheek; small frontal catchlights.",
"on_hair_clothing_background": "Curl ridges reflect light; gaps stay near-black. Top approaches ivory in bright folds; arm on the viewer's right is warm-lit. Walls are cool and dim, brighter on the viewer's right; chair and cabinet lose detail in shadow.",
"color_temperature": "Foreground is neutral-warm, about 4300 to 5000 K; room ambient is cooler gray-green. Skin shifts peach, walls blue-gray, jewelry gold.",
"exposure": "Face is bright but detailed; nose and lip highlights do not clearly clip. Top approaches white; background shadows are dense."
},
"camera": {
"type_and_lens": "Handheld front-facing selfie camera, small sensor, moderately wide field about 24 to 28 mm equivalent. Camera not visible.",
"position_and_angle": "Arm's-length distance, roughly 45 to 65 cm. Lens slightly above eye level, angled down slightly; vertical framing, nearly level roll.",
"perspective": "Mild wide selfie perspective enlarges nose, lips and central face; near shoulder on the viewer's right and arm also enlarge. Background furniture appears smaller.",
"focus_and_depth": "Face and front hair in focus; near arm softer; dog and furniture blur with distance.",
"processing": "Compressed selfie-video still with mild facial smoothing, soft background and slight dark-area noise."
},
"composition": {
"framing": "Vertical 9:16 close-medium selfie from crown to upper torso; lower torso cropped. Hair nearly spans width; face upper-center, near arm diagonal at viewer's right.",
"positions_percent": "Crown 12 to 13 percent from top; eyes 29 to 31 percent down. Face center 44 percent left and 34 percent top; nose 44/36, mouth 44/41, chin 44/47. Shoulder on the viewer's right at 60/56; dog near 88/72.",
"crop_lines": "Small headroom. Both shoulders partly visible. Arm on the viewer's left is cut at viewer-left and bottom; arm on the viewer's right exits lower viewer's right. Hands and lower body out of frame; crown not clipped.",
"layers": "Foreground: face, hair, neck, halter and near arm. Behind: furniture, gray wall, cabinet, floor and dog.",
"hierarchy": "Face and gaze first, then dark curls, glossy lips and hoop. Pale top brightens lower frame; dog and furniture are secondary."
},
"color": {
"palette_with_hex": "Hair #261F21, curls #3C3234, forehead #E4A790, cheek #D18E7B, nose #EBAC96, arm #D0957C, lips #B67D69, top #E1E0D6/#C9CBC6, wall #4E5556/#7F8184, chair #322F37, wood #453B2C, gold jewelry.",
"white_balance": "Cool gray-green room contrasts with peach skin and ivory-mint top; gold accents remain warm.",
"contrast_and_saturation": "Medium contrast, stronger in hair and eyes, softer on skin. Warm moderately saturated skin; low-saturation gray room and pale top.",
"look": "Natural selfie color, bright frontal light, warm skin against cool-gray surroundings."
},
"realism_cues": [
"Tiny beauty mark below lower lip, viewer's right.",
"Small speck on outer viewer's right upper arm.",
"Unequal eye openness and uneven brow heights.",
"Viewer's left cheek edge partly hidden by a dangling curl.",
"Fine curved baby-hair swoops across forehead.",
"Irregular twist widths, spiral tips and dark gaps.",
"Single gold hoop at viewer's right ear.",
"Bright nose-tip reflection and lip-gloss glints.",
"Face peachier than the deeper caramel arm and neck.",
"Near arm enlarged by perspective; dog blurred behind it."
],
"negative_constraints": [
"No text, captions, logos, UI, icons, borders or watermarks.",
"No extra people; keep one small cream dog in the rear, viewer's right.",
"No duplicated accessories; one gold hoop at viewer's right and one fine necklace.",
"No phone or camera body; camera and holding hand stay outside crop.",
"No generic face, altered proportions, symmetry correction or beautification.",
"Do not lighten or even skin; keep caramel tone and face-to-arm difference. No plastic, waxy or airbrushed skin.",
"No straight hair, uniform curls, different hair color or missing baby hairs.",
"No open smile or teeth; lips stay closed and gaze on lens.",
"No different top or room; keep pale lace-trim halter and gray dining background.",
"No heavy HDR, fake bokeh, oversharpening, strong color grade or lens flare.",
"No distorted anatomy, extra fingers, visible hand or stretched shoulder."
]
}
I stopped using image models for AI characters. here's what I do instead:
image models make people look too perfect. smooth skin, perfect light, that fake AI look. you can spot it in a second
video models don't do that. Seedance 2.5 makes people look like they were filmed on a phone. so I make my characters there
here's how:
go on TikTok or Pinterest and search what you want. "girl in kitchen", "guy in car", "UGC in bedroom". find a video that looks real and capture the exact frame you want
put that frame into ChatGPT and ask it to describe everything in a JSON prompt. face, hair, skin, clothes, light, camera angle...
this part matters most. I have my own way of writing these prompts and it makes a huge difference. but even a simple version works
paste the JSON into Seedance 2.5. set it to 4 seconds, 9:16, 1080p. hit generate
watch the video, pause on the best moment and save that frame
that's your character
it costs more than an image model. but the skin, the light, the small mistakes that make people look real... you can't get that any other way right now
check the examples below
Robert Rock retweeted
20 static ad formats worth testing:
1. Comparison
2. Myth vs fact
3. Apple Notes
4. Problem solution
5. Venn diagram
6. Offer-centric
7. Bullet points
8. Before after
9. Text message
10. Meme
11. Testimonial
12. Zero stars
13. Stats surround
14. Product + FB comment
15. Us vs them
16. IG story
17. Fake press
18. Feature callout
19. Question headline
20. Bold statement
Which one haven't you tested yet?
Robert Rock retweeted
30 yaşından önce emekli olacak kadar para nasıl kazanılır?
28 yaşında 40 milyon dolar kazanan Charlie Morgan, nasıl yaptığını anlatıyor.
Tamamen ücretsiz ve Türkçe altyazılı.
Robert Rock retweeted
I've spent $7.8M on Facebook Ads last month.
Here's EVERYTHING you need to know about what's WORKING and what's NOT (a thread 🧵):
If your ad isn't working, 90% of the time it's the hook.
⠀
Took me 6 years and millions in ad spend to figure out what makes a hook convert.
⠀
You're getting it for free.
⠀
Here's how to write hooks that actually scale your brand to $100k/day:
Robert Rock retweeted
At what point do we stop focusing on who needs to pay more tax and start asking where all of our tax money is going.
Robert Rock retweeted
NASA's released view of Mars' night sky (4K resolution)
Readers added context they thought people might want to know
Profile is based in”NY” meanwhile their location is UK and they are connected via Pakistan App Store. This is an engagement based account where they get paid for engagement traffic and post nonsensical videos for views. Also isn’t mars twitter.com/NASAMars/statu…
Robert Rock retweeted
Introducing Parker: The world's first AI Creative Director.
100+ brands like Grüns, Lume and Legends use us already.
RT + Comment "Parker" and I'll send you 100+ AI prompt engineers for hire.
Robert Rock retweeted
I earn £100 grand a year.
I have a detached house. 2 cars and my wife is a full time home maker.
I can just about afford this. I probably couldn’t afford the house I own if I was buying for the first time today.
My salary puts me in the top 5% of earners in the UK.
Top 5% to have a similar standard of living to Homer Simpson.
That is a disgrace.
We have a flatter salary distribution than Soviet Russia and the highest taxes and borrowing costs we’ve seen in generations.
Success is almost impossible because our tax system means hard work makes you a stressed out tax cow, not wealthy and certainly not free.
We need something radical.
Less state.
More family.
Less handouts.
More reward for risk.
Instead. Tomorrow, Labour will offer another batch of policies which will send one message to aspirational Brits: why bother.
I read a piece in the FT this week about the price of being middle class in Britain now and it said a couple earning £436,000 combined take home £216,000. That’s a crazy amount of tax, and it doesn't include VAT on purchases, council tax, stamp duty or other taxes (I think does include pension contributions). I can’t stop thinking about it.
































