ProjectGraySweatpants

Come for the faces.
Stay for the football.

Start thirsting → Find your teamHot & Good

The Starting Five

The five hottest faces on active rosters right now. Yes, they all play. Yes, you should be watching.

What makes a 9?

We swapped each feature of his face (eyes, cheekbones, that smile…) for another man's and re-scored it. Wherever the score drops, his own is doing the heavy lifting. The hotter the outline, the more it's working for him, and his top three are numbered so you can see exactly why he made the cut.

costing himmehdoing the most

Points his score drops when that feature is swapped for another man's (16 men scored anywhere from 2 to 9, averaged).

How it works

A computer vision model trained on male beauty standards, a fine-tuned vision transformer.

  1. Face detection & alignmentMediaPipe FaceLandmarker fits a 478-point 3D face mesh. From the eye centers, each face is rotated level and scaled so the crop spans 4.6× the distance between the eyes, with the eyes at 46% height: 256 px, resized to 224 for the network. Every face lands in the same frame.
  2. Labels faces from public face datasets and athlete headshots, each scored 1–10. Within each score, faces were ranked by head-to-head picks and fit with a Plackett–Luce model, turning whole numbers into fine-grained targets (an 8.2 vs an 8.8).
  3. ArchitectureA CLIP ViT-B/16 vision transformer (86M parameters, pretrained contrastively on 400M image–text pairs) turns the face into a 768-dimensional embedding. A 512-unit MLP head, conditioned on a gender embedding, regresses the score.
  4. Face featuresNobody tells the model what a jawline is; it learns face geometry straight from the pixels. To see what it picked up, the 478-point mesh maps each face's regions (eye area, cheekbones, brows, nose, mouth, jaw and chin, hair) and swaps each one for someone else's. The eyes and cheeks move the score more than any other feature: for 8s and 9s, the eye area alone is worth about half a point. Eye color is measured, not guessed: iris pixels inside the eyelid outline, minus pupil and glare, read in CIELAB color space. Looks like curls, beards and that smile are tagged zero-shot by matching each face to text prompts in CLIP.
Technical details

The hottest players on every team

Gray sweatpants season, team by team. Pick a team to see its players ranked by face score.

Slide into my inbox. A player we slept on, a photo doing him dirty, a feature you're dying for, a league to add: tell me.

✉ UncleSam@phorager.com

Keep the thirst alive

Enjoying the view? Tips keep the site going.

☕ Buy me a coffee 👖 Buy me a pair of gray sweatpants