OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Llff

Llff is a TensorFlow research implementation of Local Light Field Fusion that synthesizes novel viewpoints from sparse images of a static scene. It converts images into multi-plane representations and blends rendered local light fields; its repository documents COLMAP-based camera-pose recovery, MPI generation, and novel-view rendering, as well as a zip-to-video quickstart.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Llff — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Llff when you have a sparse set of photos of a static scene and want to generate intermediate viewpoints or a rendered video using a local research-code workflow.

Who it is for

  • Researchers studying image-based view synthesis.
  • Developers prototyping light-field rendering or novel-view generation from scene photos.

Documented uses

  • Generate novel viewpoints and an MP4 from a zip of scene photographs.
  • Recover camera poses and depth bounds for a scene using COLMAP.
  • Explore a pretrained TensorFlow workflow for generating multi-plane images and rendering views.
Before you choose

Requirements & limits

The repository is public and lists the GPL-3.0 license. The documented workflows run from a local environment or Docker container.

  • This is a code implementation, not a hosted service; image processing is performed through the documented local workflows.
  • Camera poses can be recovered with COLMAP, while the README also describes using your own poses without running COLMAP.
  • Manual setup requires CUDA, TensorFlow, COLMAP, ffmpeg and the repository's Python requirements; the Docker image route is another documented setup option.
  • The example workflow downloads a pretrained checkpoint and sample data; custom-image use depends on suitable static-scene captures.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Render novel views from a zip of scene photos

Use the documented quickstart to generate rendered frames and an MP4 from a captured image archive.

Before you start

  • Llff repository and its documented dependencies available locally.
  • A zip file of images of a static scene; the README recommends limiting maximum view disparity to about 64 pixels.

Steps

  1. Prepare a zip file of the scene images.

  2. Run the command below.

    Documented command
    bash zip2mpis.sh [--height HEIGHT]

  3. Optionally set HEIGHT; the README recommends 360 pixels for faster results.

Expected outcome

Rendered novel-view frames and an MP4 video in the specified output directory.

Guide limitations

  • The workflow is for images of a static scene; capture quality and viewpoint spacing affect suitability.
  • The README's 64-pixel disparity figure is a rule of thumb, not a guarantee.
Documentation 1 ↗
Source-based practical guide

Run the example scene with the documented Docker setup

Prepare the example data and use the Docker demo workflow to generate its rendered video.

Before you start

  • The Llff repository downloaded locally.
  • Docker and NVIDIA Docker installed.
  • An NVIDIA GPU environment suitable for the documented NVIDIA Docker workflow.

Steps

  1. Run the command below from the repository base directory.

    Documented command
    bash download_data.sh

  2. Pull bmild/tf_colmap with Docker and tag it as tf_colmap.

  3. Run demo.sh in the repository through the documented nvidia-docker run command.

Expected outcome

The README says the demo generates `data/testscene/outputs/test_vid.mp4`.

Guide limitations

  • The Docker image download is listed as approximately 6 GB.
  • This recipe follows the README's Docker instructions; it does not establish compatibility with other environments.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
Fyusion/LLFF · Official homepage title · retained in its source language
How we review sourcesSuggest a correction