AnimateDiffWhile AnimateDiff's core motion modules excel at creating natural, general-purpose animation, artists often need more explicit control—especially over the virtual “camera.” This is where the AnimateDiff Motion LoRA comes in. A Motion LoRA is a specialized, lightweight LoRA (Low-Rank Adaptation) model that, when applied, injects a specific, directed camera movement into your animation. This powerful tool separates camera motion from the subject's animation, giving you a new layer of directorial power.
Instead of hoping the model correctly interprets “zoom in” from a text prompt, you can now apply a dedicated Motion LoRA for that exact effect. This guide will cover the official set of camera motion LoRAs, explain how to install and use them, and show you how to combine them for more complex shots. Mastering the AnimateDiff Motion LoRA is a key step towards creating more dynamic and professional-looking AI animations.
Czym jest Motion LoRA?
A Motion LoRA is a small, supplementary model that modifies the behavior of a primary AnimateDiff motion module. Unlike a style LoRA that changes a character's appearance, or a concept LoRA that adds an object, a Motion LoRA is trained specifically to inject a consistent vector of movement across all frames. In this case, the official models are trained to produce common camera moves.
Crucially, the official camera Motion LoRA set is designed to work with the v2 motion module (mm_sd_v15_v2.ckpt). While they might have some effect on other modules, they are tuned for and perform best with this specific version. Think of it as an add-on: the base motion module provides the animation of the subject, and the Motion LoRA provides the movement of the camera filming that subject.
Instalacja i konfiguracja
Using a Motion LoRA is a straightforward process involving downloading the files and placing them in the correct directory for your AnimateDiff setup.
Pobierz LoRA
Download the official Motion LoRA files (v2_lora_*.ckpt) from the AnimateDiff Hugging Face repository or Civitai. There are 8 in total.
Umieść w folderze
Place the downloaded .ckpt files inside the dedicated Motion LoRA folder for your UI.
A1111: stable-diffusion-webui/extensions/sd-webui-animatediff/data/motion_lora/
ComfyUI: ComfyUI/custom_nodes/ComfyUI-AnimateDiff-Evolved/motion_lora/
Włącz i ustaw wagę
In the AnimateDiff settings in AUTOMATIC1111 or the node in ComfyUI, select the desired Motion LoRA and set its weight. A weight of 0.7 to 1.0 is a good starting point.
8 oficjalnych Motion LoRA kamery
The official collection provides a comprehensive toolkit for basic cinematography. Each Motion LoRA is designed to be clear and predictable in its effect. Here's a list of what each one does:
| Nazwa pliku Motion LoRA | Efekt ruchu kamery | Typowe zastosowanie |
|---|---|---|
| v2_lora_ZoomIn.ckpt | Gradually zooms into the center of the frame. | Creating focus, emphasizing a detail, adding dramatic tension. |
| v2_lora_ZoomOut.ckpt | Gradually zooms out from the center of the frame. | Revealing a wider scene, creating a sense of scale or distance. |
| v2_lora_PanLeft.ckpt | Pans the camera horizontally to the left. | Following a subject, revealing a new area, creating a sense of travel. |
| v2_lora_PanRight.ckpt | Pans the camera horizontally to the right. | Similar to Pan Left, but in the opposite direction. |
| v2_lora_TiltUp.ckpt | Tilts the camera vertically upwards. | Revealing something tall (e.g., a skyscraper), creating a sense of awe. |
| v2_lora_TiltDown.ckpt | Tilts the camera vertically downwards. | Looking down on a subject, revealing something on the ground. |
| v2_lora_RollingAnticlockwise.ckpt | Rolls the camera counter-clockwise around its axis. | Creating a sense of unease, dizziness, or a stylized action shot. |
| v2_lora_RollingClockwise.ckpt | Rolls the camera clockwise around its axis. | Similar to Rolling Anticlockwise, but in the opposite direction. |
Wskazówki użytkowania i łączenie LoRA
Using an AnimateDiff Motion LoRA is as simple as selecting it from a dropdown in your UI. However, to get the best results, consider these tips:
- Adjusting Weight: The LoRA weight controls the intensity of the camera motion. A lower weight (e.g., 0.5) will result in a slower, more subtle camera motion. A higher weight (e.g., 1.0 or slightly above) will produce a faster, more pronounced movement. Pushing the weight too high can sometimes introduce artifacts, so it's best to start around 0.8 and adjust from there.
- Stacking Multiple Motion LoRAs: The true power of the AnimateDiff Motion LoRA system is its stackability. Both AUTOMATIC1111 and ComfyUI allow you to apply multiple Motion LoRAs at once. For example, you can combine “Zoom In” and “Pan Left” to create a diagonal camera move that pushes into the scene while moving sideways. When stacking, you may need to reduce the individual weights of each LoRA (e.g., set both to 0.6) to prevent the movements from becoming too extreme and overpowering the base animation.
- Negative Weights: You can use a negative weight (e.g., -0.8) on a Motion LoRA to reverse its effect. For example, applying a negative weight to the “Zoom In” LoRA will make it zoom out. This effectively doubles the number of camera moves at your disposal.
By experimenting with different combinations and weights, you can create a nearly infinite variety of complex camera motions, adding a layer of cinematic storytelling to your AnimateDiff projects that was previously difficult to achieve with prompts alone. This is a crucial technique for any serious AnimateDiff artist.