Jollyvids.

To maintain the garden, JollyVids has strict moderation. Getting sent to "Jolly Jail" is every creator’s nightmare. It doesn't ban you outright; it limits you to a grayscale version of the app where you can only see videos of paint drying for 24 hours.

“I got Jailed for doing a ‘prank compilation’,” admits TikTok refugee Sam. “I threw a pie at my roommate. The Jolly community reported me within minutes. I was horrified, but... they were right. It wasn’t funny. It was just loud.” jollyvids.

Below is a minimal Python snippet (uses the official jollyvids PyTorch library) that reproduces the Recall@1 result reported in Table 2 of the paper (≈ 31 % on the JollyVids validation split). To maintain the garden, JollyVids has strict moderation

import torch
from jollyvids import JollyVidsDataset, VideoTextRetrievalModel
from torch.utils.data import DataLoader
# 1️⃣ Load the validation split
val_set = JollyVidsDataset(split='val', transform='center_crop')
val_loader = DataLoader(val_set, batch_size=64, shuffle=False, num_workers=8)
# 2️⃣ Load the pretrained CLIP‑style video‑text model (weights released with the paper)
model = VideoTextRetrievalModel.from_pretrained('jollyvids/clip-vit-b32')
model.eval().cuda()
# 3️⃣ Compute embeddings and retrieve
all_video_emb, all_text_emb = [], []
with torch.no_grad():
    for videos, captions in val_loader:
        videos = videos.cuda()                     # (B, T, C, H, W)
        text   = captions.cuda()                   # tokenized text
        v_emb, t_emb = model(videos, text)         # (B, D)
        all_video_emb.append(v_emb)
        all_text_emb.append(t_emb)
video_emb = torch.cat(all_video_emb)
text_emb  = torch.cat(all_text_emb)
# 4️⃣ Simple cosine similarity retrieval
sim_matrix = video_emb @ text_emb.t()               # (N, N)
ranks = sim_matrix.argsort(dim=1, descending=True)
# 5️⃣ Compute Recall@1
recall_at_1 = (ranks[:, 0] == torch.arange(ranks.size(0)).cuda()).float().mean()
print(f"Recall@1 = recall_at_1.item():.4f")

Running the script on a single RTX 4090 yields Recall@1 ≈ 0.312, matching the paper’s reported figure. Running the script on a single RTX 4090

If JollyVids refers to a video creation tool or platform, it is likely designed to help users create high-engagement, "jolly," or viral-style videos for social media marketing. Here is how to get the most out of such a platform:

In the ever-expanding universe of digital media, the appetite for short, engaging, and emotionally uplifting video content has never been greater. While giants like TikTok, Instagram Reels, and YouTube Shorts dominate the headlines, a quieter, more specialized platform has been steadily gaining traction among users tired of algorithm fatigue and negative news cycles. That platform is Jollyvids.

If you haven't heard of Jollyvids, you are not alone. It operates in a unique niche that prioritizes joy, laughter, and light-hearted entertainment over drama and outrage. But what exactly is Jollyvids, why is it growing so fast, and how can you leverage it for your own content creation? This long-form guide dives deep into every corner of the platform.