Deforum

Verified

Open-source Stable Diffusion toolkit for keyframed AI animation

Visit Deforum →No rating yetFree

Product preview

Deforum

What it is

Deforum is a community-built, open-source toolkit for making animations with Stable Diffusion. It began as a Google Colab notebook and grew into an extension for the AUTOMATIC1111 Stable Diffusion WebUI, which is now its most complete version. Artists write prompts that change over time and control 2D and 3D camera motion with math-based keyframes, with depth models for 3D movement, video input for style transfer and frame upscaling. Companion tools include Parseq for audio-synced keyframing. The extension is free under the AGPL license and runs locally on your own GPU or in Colab. Its hosted Deforum Studio web app is not accepting new users.

Key features

Prompt scheduling across frames
2D and 3D camera motion with keyframes
Video input for style transfer
Parseq companion for audio sync

What we liked

+ Free and open source
+ Very fine control over motion and timing
+ Large community and shared settings

Worth noting

· Steep learning curve for new users
· Needs a capable GPU or Colab setup

Editor's notes

More context from AI BirdView

Deforum earned its reputation as one of the early tools behind the dreamlike, morphing AI animations seen in music videos and visual art. Its strength is control: instead of a single prompt and a button, you script how prompts, zoom, rotation and camera paths change frame by frame. That depth has kept it popular with artists who want a specific look and are happy to experiment until a sequence feels right.

Math-based keyframing for zoom, angle, translation and strength
3D motion using depth estimation for a sense of camera travel
Video input mode for restyling existing footage
Parseq support for syncing motion to music and beats
Open-source code under the AGPL license with an active community

It is best for digital artists, VJs and musicians who are comfortable with Stable Diffusion and want frame-level creative control. A practical tip: start from a shared community settings file and change one parameter at a time, rendering short low-resolution tests before committing to a full-length animation, since long renders can take a while.