Deforum
What is Deforum?
Deforum is an open-source tool that turns Stable Diffusion image generation into animation by letting users define how the image, camera, and prompts evolve over time across a sequence of frames.
At a glance
- Type of model
- Open-source animation extension for Stable Diffusion, not a standalone model
- Developed by
- Open-source community (Deforum contributors)
- Key capability
- Keyframe-based animation of Stable Diffusion outputs with camera movement controls, prompt scheduling, and frame-by-frame transformation parameters
- How it fits in AI workflow
- Used for producing AI-animated video sequences within the Stable Diffusion ecosystem, particularly for experimental, stylized, and parameter-driven animation that requires more control than dedicated video models provide
Ready to create?
Direct scenes, design characters, and ship full films
All-in-one AI creative platform with simple, transparent pricing, no speed throttles, and an infinite Canvas for max creativity.
How it compares
Dedicated video generation models such as those from Runway or Kling generate video as a unified sequence with learned temporal coherence, producing motion that looks natural and physically plausible. Deforum generates frame by frame with transformation applied between each, producing a characteristic flowing-morphing aesthetic that is visually distinctive but less temporally coherent than dedicated video models. Deforum offers greater parameter-level control; dedicated models offer more natural-looking motion and simpler operation.
Pro tip
When using Deforum for animation, setting small per-frame transformation values and using longer sequences produces smoother, more controlled motion than large per-frame jumps. A zoom increment of 0.02 per frame over 300 frames creates a steady, gradual zoom that feels cinematic; a zoom of 0.2 per frame over 30 frames produces the same total movement but looks rapid and jerky. Subtle settings with longer sequences are almost always preferable for polished output.
Types and variations
- 2D mode applies transformations directly to the generated frame as a flat image, including zoom, rotation, and translation, producing animation through frame-to-frame image manipulation.
- 3D mode uses depth estimation to apply perspective-correct camera movement simulation, creating a more convincing sense of moving through three-dimensional space.
- Video input mode uses an existing video as the initialization for each generated frame, applying Deforum's stylization on top of real footage.
- Prompt scheduling allows text prompts to change at defined keyframe points, enabling the animated content to evolve between different subjects or styles over the duration of the sequence.
Ready to make your first scene in Morphic?
Try MorphicCommon use cases
- Producing experimental and psychedelic AI animation sequences for art projects, music videos, and creative exploration.
- Generating dream-like morphing visuals that flow continuously between subjects and environments over time.
- Creating looping animated backgrounds and visual environments for live performance, installation art, and motion graphics contexts.
- Stylizing existing video footage using Stable Diffusion aesthetics applied frame-by-frame through Deforum's video input mode.
- Building long-form AI animation sequences with precisely controlled camera movement and prompt evolution that dedicated video models cannot replicate at the same parameter level.
Ready to create?
Direct scenes, design characters, and ship full films
All-in-one AI creative platform with simple, transparent pricing, no speed throttles, and an infinite Canvas for max creativity.
FAQs
Deforum is an open-source extension for Stable Diffusion that enables keyframe-based animation and camera movement control in AI-generated video. It generates each frame individually with incrementally adjusted parameters, producing animated sequences from the Stable Diffusion image generation pipeline.
Deforum generates each video frame as a separate Stable Diffusion image, applying incremental transformations including zoom, rotation, translation, and prompt changes between frames according to a user-defined keyframe schedule. The resulting frames are compiled into a video sequence.
Prompt scheduling allows the text prompts guiding generation to change at specified keyframe points throughout the animation, enabling the image content to evolve between different subjects or aesthetics over the duration of the sequence.
Yes, for specific use cases. Deforum offers granular parameter-level control over animation schedules that dedicated video models do not provide, and its characteristic flowing-morphing aesthetic is distinctive and valued for experimental and art contexts. It remains relevant for creators working within the Stable Diffusion ecosystem.
2D mode applies transformations as flat image manipulations, directly zooming, rotating, or translating the generated frame. 3D mode uses depth estimation to apply perspective-correct camera movement simulation, creating a more convincing sense of moving through three-dimensional space.
Deforum produces a distinctive flowing, morphing animation style where image content continuously evolves and shifts as the camera appears to move through or around generative visual forms. This aesthetic became recognizable as a genre of AI video art in the early diffusion model era.
Deforum can be accessed through Automatic1111 or other Stable Diffusion interfaces with a graphical user interface, reducing the need for direct code interaction. However, advanced use of prompt scheduling and custom parameters benefits from familiarity with the underlying parameter structures.
Yes. Deforum's video input mode uses existing video as the frame initialization, applying Stable Diffusion stylization on top of the source footage frame by frame. This produces a stylized version of the original video content rendered in the aesthetic of the chosen model and prompt.
