Building Flair - Day 4 🇨🇴
Today’s update comes with a lot of emotional damage 🚬
my goals were to submit Flair to the app store and watch Colombia make it to the quarterfinals.
HOWEVER… Colombia had other plans
before the disaster, i recorded a demo of Flair’s avatar generation flow to get screenshots ready for the app store submission, so i figured i could talk about that flow today:
it doesn’t only create digital avatars. It also helps Flair estimate body proportions and measurements, so outfit recommendations can feel more personalized to different body types.
NERD ALERT: under the hood, it uses mediaPipe Pose Landmarker to detect body landmarks + create a segmentation mask. Then it uses the user’s height to calibrate pixels to centimeters and estimate body measurements from the distance between key body points
it’s not perfect, but these measurements give Flair a much better starting point than generic sizing. Specially because each store handles different sizes 💀
aaaand back to today’s goal: i finally generated the app store screenshots (using https://t.co/0oFB2eE93c) and sent the submission 🕺
@shahriyer wow i didn't know you could do things like that in figma haha, i loved your poc! and yes, i do think it's possible to get cheaper try-ons, that's something i'll definitely look into in the future
Building Flair - Day 3 💸
The original plan was to keep Flair free and monetize through affiliate links from partner stores. basically: if someone found a piece they loved through the app, we’d earn a small commission from that sale.
cute in theory!! 🤡 but it’s been over two months with no response from the affiliate programs, my guess is they want to see real users before approving us, which is fair, but also… expensive for an AI app lol.
so to avoid burning too much money, especially with image generation features, today i worked on adding a paywall. (with a generous free trial before the paywall 🤝)
integrated RevenueCat, tested the purchase flow, and it’s workingg👇
hosting a consumer founder's dinner in mexico city, end of july.
for builders, designers, tinkerers, growth marketers.
proof of work required and no anonymous profiles.
@shahriyer is it live? i'd like to seee. And yes, image generations is too expensive, i'd like to try the vton part with diffusion models in the future
@yul_magdaleno generé una imagen de la modelo usando ese outfit con nano banana y luego le pedí a veo 3.1 que hiciera un vid giratorio (en https://t.co/oyGeMeFxAm). Finalmente me aseguré de que el fondo fuera totalmente blanco (sin sombras) con https://t.co/LxcnhZhDJW 👩💻
Building Flair - Day 2 ⚽️
TIL that soccer and productivity don't mix lol
Between going to the office and watching the argentina and colombia matches, i only managed to build one feature, but here it iss:
Flair can now recommend outfits using the clothes you already own. Before this, the AI stylist could only build looks using store products. Now you can simply ask it to style your own closet instead 🧢
The feature itself was pretty straightforward to implement because i’d already built the enrichment pipeline for closet items. All that was left was classifying the user's intent (closet vs store catalog) and updating my outfit planner agent to use closet data instead of the store catalog.
i’ll probably nerd out about the technical side of this in the next weeks
Meanwhile... woohoo and viva colombiaaaa 🇨🇴🫓☕️