|
|
--- |
|
|
license: apache-2.0 |
|
|
base_model: |
|
|
- depth-anything/Depth-Anything-V2-Giant |
|
|
library_name: diffusers |
|
|
--- |
|
|
Depth Anything V2 Giant - 1.3B params - FP32 - Converted from .pth to .safetensors |
|
|
|
|
|
The model was previously published under apache-2.0 license and later removed. See the commit in the official GitHub repo: https://github.com/DepthAnything/Depth-Anything-V2/commit/0a7e2b58a7e378c7863bd7486afc659c41f9ef99 |
|
|
|
|
|
A copy of the original .pth model is available in this Hugging Face repo: https://huggingface.co/likeabruh/depth_anything_v2_vitg/tree/main |
|
|
|
|
|
This is simply the same available model in .safetensors format. |
|
|
|
|
|
If you want to use it in ComfyUI, you can use Kijai's custom nodes (https://github.com/kijai/ComfyUI-DepthAnythingV2), select the model and it will be downloaded automatically. |
|
|
|
|
|
You may get OOM using the gigant model depending on your VRAM and the size of the image you're processing. In these cases, try to reduce the input image size. I can get 1024x1024 depth maps just fine with 24GB VRAM (uses about 56% of available VRAM). |
|
|
|
|
|
|
|
|
~~If you want to use it in ComfyUI, you have two options:~~ |
|
|
|
|
|
~~1. (Recommended) Use the .safetensors file with the modified version of Kijai's custom nodes (https://github.com/kijai/ComfyUI-DepthAnythingV2). Just replace the ComfyUI/custom_nodes/comfyui-depthanythingv2/nodes.py file with the nodes.py file in this repo and ensure depth_anything_v2_vitg_fp32.safetensors is in the ComfyUI/models/depthanything/ folder, as it will not be downloaded automatically.~~ |
|
|
|
|
|
~~2. Use depth_anything_v2_vitg.pth directly with Fannovel16's custom nodes (https://github.com/Fannovel16/comfyui_controlnet_aux). Use a node called Depth Anything V2 - Relative and select depth_anything_v2_vitg.pth. Ensure the file is in the folder ComfyUI/custom_nodes/comfyui_controlnet_aux/ckpts/depth-anything/Depth-Anything-V2-Giant/ folder, as it will not be downloaded automatically.~~ |
|
|
|
|
|
|
|
|
~~Kijai's nodes produce more detailed depth maps. However, you will likely get OOM using the gigant model depending on your VRAM and the size of the image you're processing. I can get 1024x1024 depth maps just fine with 24GB VRAM.~~ |
|
|
|
|
|
|
|
|
--- |
|
|
license: apache-2.0 |
|
|
--- |
