Home › Video Tools › AI-WebTV

AI-WebTV - AI Video Tools Tool

Overview

AI-WebTV is a public, live demonstration hosted on Hugging Face Spaces that streams automatically generated video content using a fine‑tuned Modelscope-based model. The Space runs an automated prompt database with themed prompt sets (for example: nature, sci‑fi, abstract) and continuously renders short clips for web viewing. The project is presented as a research demonstration and explicitly enforces content guidelines to avoid violent or excessively gory material. The demo is intended to showcase video synthesis research, user-selectable prompt themes, and an automated pipeline for continuous streaming rather than a polished commercial product. According to the Space listing and discussion thread, the model produces outputs similar in style to the Zeroscope family of video synthesis models, while specific training data, model parameter counts, and an evaluation benchmark suite are not published on the Space. The project page and discussion on Hugging Face provide the canonical entry points for trying the demo and reading developer notes (see the Space and discussion pages for current status and usage details).

Model Statistics

Likes: 502

Model Details

Architecture and base model: AI-WebTV uses a fine‑tuned Modelscope‑based video generation model (Modelscope is the base framework referenced by the author). The Space description states the outputs are similar to those of the Zeroscope model family; however, the Space does not publish the full model architecture diagram, parameter counts, or training datasets. Capabilities: The demo streams short generated videos from an automated prompt pool with selectable themes. It supports continuous generation for live viewing and cycles through an internal prompt database to demonstrate variation. The project enforces research‑only usage and forbids prompts that request violent or excessively gory content. Unknowns and boundaries: The Hugging Face Space does not disclose pipeline identifiers, exact parameter counts, latency metrics, or an explicit API for programmatic inference. There is no published benchmark table or training corpus on the Space; model weights and a formal license are not documented on the public page. For implementation details and recent changes, refer to the project's Hugging Face Space and discussion thread.

Key Features

Live streaming of automatically generated short video clips.
Fine‑tuned Modelscope‑based model producing Zeroscope‑like outputs.
Automated prompt database with selectable themed prompt sets.
Research‑only usage guidance and content filters to prevent graphic violence.
Public Hugging Face Spaces demo for interactive viewing and discussion.

Example Usage

Example (python):

import requests
import re

# Example: fetch the Hugging Face Space page and try to find embedded video URLs (mp4).
# Note: many Spaces stream via blobs or websockets; this scraper works only if direct video URLs are present.

SPACE_URL = "https://huggingface.co/spaces/jbilcke-hf/AI-WebTV"

resp = requests.get(SPACE_URL, timeout=15)
resp.raise_for_status()
html = resp.text

# Very simple regex to find .mp4 links in the page HTML
mp4_urls = re.findall(r'https?://[^"\'>]+\.mp4', html)
mp4_urls = list(dict.fromkeys(mp4_urls))  # deduplicate while preserving order

if not mp4_urls:
    print("No direct .mp4 URLs found on the Space page. The demo may stream via blobs/websockets or require interactive controls.")
else:
    # Download the first found MP4
    mp4_url = mp4_urls[0]
    print(f"Found mp4: {mp4_url}\nDownloading...")
    r = requests.get(mp4_url, stream=True)
    r.raise_for_status()
    with open("ai_webtv_clip.mp4", "wb") as f:
        for chunk in r.iter_content(chunk_size=8192):
            if chunk:
                f.write(chunk)
    print("Saved to ai_webtv_clip.mp4")

# For programmatic interaction with Spaces that expose an API, consult the Space owner or the Space's /api endpoints.
# The project page and discussion thread are the best place to ask for an official programmatic interface.