Pinned Tweet
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
AI UGC is now IMPOSSIBLE to spot.
if yours still looks fake, it’s not the model. it’s your prompt
the right keywords and precise details are the whole gap
and this isn’t even the limit. it gets better every single day
everything I used to make this is inside my V3 system
just watch:
if you want the full system I used to make this, prompts and workflow included, it's here:
whop.com/creativelab-access/…
74
if your AI character looks like a GTA cutscene people will notice lol
the starting frame decides how real the whole video looks.
if your reference already looks smooth and rendered, the video model isn't gonna save it. it just carries that game look into every single second.
start with a reference that already looks like a real person filmed on a phone. then the video actually has something real to build on.
fix the frame first. that's what stops your videos from screaming AI.
2
6
814
the AI yapper format is still undefeated.
and it doesn’t have to cost you a fortune...
Omni Flash makes yappers dirt cheap, and the quality is really good. the lip sync, the expressions, all of it
I made this one with Omni 1.2 to test it out, but to be honest... I’m sticking with 1.1. it’s cheaper and the results are basically the same
just two things to keep in mind: clips cap at 10 seconds, so you’ll stitch them together, and there’s no voice reference
most people upscale inside Google Flow. don’t do that pls
for the best quality, download the clips, stitch them together, then upscale the whole video at once with Topaz upscale
watch this👇
I put Kling Image 4.0 against Nano Banana 2.1 and GPT Image 2.6. the results honestly surprised me.
same prompt. same character. 3 models.
which one looks most believable to you?
prompt in the comments 👇
🤖 Made with AI
Candid phone snapshot, shot with front phone camera, real phone photo of an attractive adult man in his early-to-mid 20s, an unedited camera photo, SOOC. He has thick, voluminous dark brown curly hair with naturally defined loose ringlets, styled in a slightly messy shape with several prominent curls falling across his forehead. His hair has natural volume, subtle shine, and individual flyaway strands. He has light olive skin, blue-grey eyes, thick dark eyebrows, a straight nose with a small silver nose stud, naturally full pink lips, and a defined jawline. He has a neatly trimmed dark mustache and a distinctive short, fuller goatee concentrated around his chin. Small sparkling silver diamond-style stud earrings are visible in both ears.
He is wearing a relaxed white cotton graphic T-shirt with oversized green and purple printed lettering across the front. The shirt has a slightly loose fit, natural fabric creases, and a simple rounded neckline.
He is reclining comfortably against a sofa or bed, holding his iPhone slightly above his face to take a close-up front-facing selfie. His head rests naturally against the surface behind him, shoulders relaxed. He looks directly into the camera with a subtle, confident half-smile and a calm, relaxed expression. His face is the main focal point.
The framing is vertical 9:16 portrait, close-up selfie from the upper chest upward, shot with front phone camera at a slightly elevated angle. His face is centered, with his voluminous curly hairstyle occupying much of the upper frame and his white graphic T-shirt visible across the lower foreground. Candid phone shot, spontaneous and slightly imperfect, like a casual selfie taken while relaxing at home.
The environment is a cozy, ordinary bedroom or living space. Behind him is a light cream-colored wall with a subtle textured painted finish. A large black flat-screen television is partially visible on the upper-left side of the background. On the right side, a dark black backpack or soft travel bag hangs casually near the wall, with realistic fabric folds and straps visible. A small section of light beige furniture or bedding is visible behind his shoulders.
Lighting is warm, soft indoor household lighting, creating natural illumination across his face with gentle shadows beneath his curls, nose, and chin. His skin has natural highlights, and the white shirt retains visible cotton texture without appearing overexposed.
Real photo look, IG phone snapshot: visible pores, natural skin-tone variation, under-eye texture, subtle freckles and facial imperfections, individual defined hair curls, natural catchlights in the eyes, accurate facial proportions, natural mustache and goatee texture, visible cotton weave. Natural colors, authentic handheld perspective.
Avoid the look of modern high-definition digital fashion photography. No plastic skin, no excessive smoothing, no beauty retouching, no sharp digital look, no cinematic lighting, no dramatic color grading, no filters, no distorted anatomy, no extra fingers, no watermarks or added text overlays. Strictly vertical 9:16 portrait orientation.
1
4
668
Seedance 2.5 made AI street interviews way TOO easy.
the realism? AI couldn’t come close to this a few months ago...
there are no limits anymore. you can literally create any street interview you want
picture it, give Seedance the right prompt and it’s done
and you’re not stuck with short clips. you can go way longer...
ads have never been easier to make.
watch this:
none of these people exist.
not the interviewer, not the strangers, not the street.
this AI street interview took me 10 minutes with Seedance 2.5
watch this👇
most people burn their credits on AI video because they skip this one step
and it takes like 30 seconds lol
your clip comes out wrong so you type "make the hands better, fix the lighting" and hit generate again. bro…the model can't read your mind. your vague notes are the reason it keeps missing.
download the video and drop it back into the chat where you built the prompt. claude, chatgpt, whatever you use. let it WATCH the clip and compare it to the prompt.
now it actually sees what went wrong instead of guessing from your description. it spots what got ignored and rewrites the prompt to fix exactly that.
sounds stupid simple, but it saves you a TON of credits.
6
22
1,770
more AI characters made with my Seedance 2.5 method.
none of these people exist:
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
get access to my V3 system here:
whop.com/creativelab-access/…
1
1,121
I genuinely can't tell this is AI anymore.
this storytime yapper took me under 10 minutes with Seedance 2.5
the face, the lip sync, the little pauses. nothing gives it away
just watch this👇
get access to my V3 system here:
whop.com/creativelab-access/…
1
4
731
Nano Banana 2.1 vs GPT Image 2.6 vs MY Seedance 2.5 method.
same prompt. three very different results.
prompt is in the comments. judge it yourself:
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
I genuinely had to double check if this stain remover ad was AI.
the coffee spill, the panic, the wipe, the stain disappearing...
this is exactly the stuff that usually gives AI UGC away. Seedance 2.5 nailed all of it
a real creator would need a shoot, a product and a ruined shirt. this took me minutes
most brands still think AI UGC is just talking avatars
watch this:
I just created a Shark Tank episode with AI in under 10 minutes
when I say you can create anything with AI, I mean LITERALLY anything, there are no limits
give Seedance 2.5 a direct prompt and it’s done
watch this:
someone posts one clip saying "this new model is way better than seedance" and half of my timeline wants to switch lol
bruh you don't know what prompt he used. you don't know how many tries it took him to get that one good clip either.
test it yourself first. put your own character and your own prompt into both models and compare them, then decide













