AxialView LoRA: Aerial Top-Down Video Generation
AxialView is a LoRA for Wan2.2 T2V 14B that produces axial aerial shots with circular and symmetrical compositions from coral atolls to stadium light shows.
On this page
AxialView LoRA: Top-Down Aerial Video Generation with Wan 2.2
TL;DR:
- AxialView is a LoRA for Wan-AI's Wan2.2-T2V-A14B that biases generation toward axial (top-down / aerial) viewpoints.
- Start prompts with "AxialView." and describe circular, symmetrical scenes for best results.
- Use
AxialView-WAN22.safetensorsin ComfyUI's Load LoRA node with strength_model around 0.8-1.0.
What "Axial View" means here
Axial view, as used here, is a strict top-down perspective — a scene observed from directly above, as if from a drone or satellite looking down on a circular subject. The model emphasizes compositions built around a center, where geometry repeats or radiates outward: atolls surrounding a lagoon, stadium fields laid out in mandala patterns, rotating machinery seen on its axis, and storm systems spiraling into an eye.
What the model does
The AxialView LoRA steers the base Wan2.2-T2V-A14B model into this specific visual grammar. Its bundled widget prompts show six representative generations: a coral atoll with manta rays in synchronized circles, a stadium light show forming pulsing LED mandalas, an ocean maelstrom with a spotlighted rescue helicopter, a particle collider ring described as a brutalist monument, a montage of industrial turbines and propellers, and a supercell storm viewed from the eye outward.
Across these examples the recurring visual language is wide, symmetrical framing with the camera looking straight down, strong radial or circular structure, and a cinematic color grade — gradient blues over water, neon against night-black for stadiums, storm grays and purples for weather.
How to use it
Place AxialView-WAN22.safetensors in ComfyUI/models/loras/. In your ComfyUI workflow add a Load LoRA node after your Wan base model loader, select the file, and connect the MODEL output to your KSampler. The card suggests a strength_model starting point of 0.8-1.0.
In prompt form, begin with the trigger phrase AxialView. and then describe the scene from a top-down framing, calling out the circular or symmetrical structure. The model card recommends explicitly describing the viewing angle ("extreme wide aerial shot, viewed from directly above") and the style ("tropical aerial cinematography", "high-contrast night photography", etc.).
Training and base model
The LoRA is adapted for the Wan-AI Wan2.2 line and is trained against Wan-AI/Wan2.2-T2V-A14B as its base model. Training hardware was sponsored by reDMOND Ai. Because it is a LoRA adapter, it is used on top of the full base checkpoint — you still need a Wan2.2 checkpoint loaded in your runtime; AxialView only modifies steering.
License
The model card declares license cc0-1.0 (Creative Commons Zero 1.0 Universal), placing the weights in the public domain.
Common errors and fixes
Error | Cause | Fix
|---|---|---
LoRA loads but output shows no aerial perspective | strength too low or prompt does not lead with the trigger | Set strength_model to 0.8-1.0 and start prompt with "AxialView." "File not found" in ComfyUI Load LoRA node | safetensors not placed in the loras folder | Put AxialView-WAN22.safetensors in ComfyUI/models/loras/ Asymmetry / off-axis framing | Prompt does not describe the top-down angle | Lead with an explicit "extreme wide aerial shot, viewed directly above" clause
FAQ
What base model do I need? Wan-AI/Wan2.2-T2V-A14B or a compatible Wan2.2 checkpoint. AxialView is a LoRA, not a standalone checkpoint.
Can I use it for still images? Yes — the card lists text-to-image among its tags, so the LoRA transfers to Wan2.2 image pipelines as well.
What trigger word is required? Start prompts with AxialView.
Sources
- Hugging Face model card: https://huggingface.co/artificialguybr/AxialView-Redmond-WAN2-T2V-14B — description, trigger words, base model, settings, license.