Home How It Works About ← View Baseline Try Free
✦ Improved

Now powered by CLIP + Cross-Image Attention + Transformer Decoder — generates 3 unique story versions per upload

Improved AI Model

Three stories.
One perfect moment.

Upload up to 5 photos and get three uniquely voiced narratives — choose the one that captures your memory best.

3
Story Versions
CLIP
Vision Encoder
40k+
Training Stories
Free
To Try

Generate your stories

Upload 1–5 photos. Get three different story versions — pick your favorite.

Drop photos here or click to browse
JPG, PNG — up to 5 photos, 10MB each
Photo 1
Photo 2
Photo 3
Photo 4
Photo 5
0 of 5 photos uploaded
Encoding your photos with CLIP...

Your 3 Story Versions

CLIP + Transformer · Choose your favorite
Version 1 · Warm
Personal & heartfelt
Version 2 · Vivid
Descriptive & visual
Version 3 · Reflective
Poetic & nostalgic
✓ Your Selected Story
CLIP ViT-B/32 · Cross-Attn · Transformer · CS 747 GMU

How it works

A state-of-the-art pipeline using CLIP and cross-image attention to understand your photos.

Step 01
Upload your photos
Select 3–5 photos from any event — travel, family, everyday moments.
Step 02
CLIP vision encoding
OpenAI's CLIP model encodes each photo into rich semantic embeddings — understanding objects, scenes, and emotions.
Step 03
Cross-image attention
A custom attention layer connects your photos together, so the story understands how scenes relate across time.
Step 04
Transformer story generation
A Transformer decoder generates three uniquely toned narratives — warm, vivid, and reflective — so you can pick the best.

Want to compare with the baseline ResNet-50 + LSTM model? See how much CLIP + Transformer improves story quality.

View Baseline →

About this model

The improved Storifai model — built for CS 747 Deep Learning at George Mason University.

This is the improved version of Storifai, replacing ResNet-50 with CLIP ViT-B/32 for richer image understanding and swapping the LSTM decoder for a Transformer decoder with cross-image attention. Trained on 40,155 VIST stories. The improved model generates three differently-toned stories per upload so you can choose the one that fits your memory best.