Laura Remade has recently released a collection of approximately ten camera effects (LoRAs). This article will examine several of these effects, focusing on their functionality, stability, and potential applications. The tests are conducted using a 9:16 aspect ratio, differing from the typical 16:9 ratio, to assess performance in a vertical format.
Workflow Setup on Running Hub
A pre-built workflow is available on runninghub.ai for testing these LoRAs.
Accessing the Workflow
- Visit runninghub.ai.
- Click on the "Workbench" on the left-hand side.
- Build and import the provided workflow.
Workflow Components
The workflow includes the following key elements:
-
Width and Height Configuration: Adjustable according to project requirements.
-
Model Loading: A 12.1 i2v 14-bit model (FP8).
-
Encoding Machines: Both text-based and visual encoding machines.
-
Model Settings: It is recommended to enable Sage Attention for improved results. Tcache is also loaded, specifically the LoRa model itself.
-
LoRa Model: The speaker has uploaded LoRa models to Running Hub. Search for "interesting 805 programmers" to find them.
-
LoRa Page Options:
-
Shift value set to 5.0.
-
CFG value set to 6.0.
-
Each LoRa requires different keywords.
-
-
Reference Images: Resized images are used with Florence for basic hint generation.
-
Hint Word Merging: Combining Florence hint with trigger words.
-
Visual Coding Machine: Used for encoding reference pictures, feeding into image-to-video conversion.
-
Output: Final positive and negative conditions, as well as latent representations.
The image-to-video settings are consistent with standard workflows.
Individual LoRa Effects
1. Crash Zoom In (Sudden Magnification)
This effect simulates a sudden magnification of the subject, typically the face, to enhance drama or detail. It is well-suited for stylized editing or unexpected reversals in the shot. The settings, including intensity, CFG, and flow shift, are explained. When applying this LoRa, adjust the hint word accordingly. The tested effect shows a distinct two-stage magnification.
2. CreateApp (Mechanical Arm Simulation)
This LoRa attempts to simulate the movement of a mechanical arm or camera. It requires changing the trigger word. The test results show only a subtle camera movement effect, which may not always be apparent, and the results aren't consistent.
3. Hacker Empire (Matrix Effect)
This LoRa mimics the "bullet time" effect seen in The Matrix, where the camera rotates around a frozen subject. The test demonstrated a stable performance with a girl appearing frozen mid-air while the camera circled her.
4. Shaking and Shaking (360 Rotation)
This LoRa is intended to create a camera rotation effect, not a subject rotation. Remember to update the trigger word. Tests indicated this LoRa's performance was inconsistent, yielding only a slight downward motion in some cases. The speaker stated that "the stability of this is actually not very good."
5. Curved Mirror (Arc Mirror)
This effect simulates a 360-degree mirror reflection. It offers more stable results compared to some others and can potentially complement the 360-degree rotation effect.
6. Shaking the Mirror (Exaggerated Reflection)
Similar to the curved mirror effect but more exaggerated. While the example strength is noticeable, the speaker's test results show less stability and an average outcome.
Overall LoRa Performance Summary
The following LoRAs demonstrated relatively strong stability during testing:
-
Hacker Empire
-
Crash Zoom In
-
Arc Mirror
The speaker suggested that the stability varies with each LoRa and the aspect ratio and the speaker mainly tests vertical video. It's important to consider performance in a comprehensive manner and not rely solely on examples provided, and to test a variety of aspect ratios for more robust results.