Recently tried SV3D, found there is no cleaer and simple instructions online, so…
最近尝试了SV3D,发现网上没有更清晰和简单的说明,所以…
SV3D 使⽤⽅法:
(English version is in next page)
仅包含最简单的运⾏⽅法(使⽤⾃⼰的图⽚三维重建)
⾸先下载第三⽅的 ComfyUI(⼀个第三⽅的 stable diffusion 的使⽤⼯具,不包含模型本体)
以下两个都安装
安装 ComfyUI – Comflowy
安装 ComfyUI Manager – Comflowy
在 hugging face 下载官⽅模型
stabilityai/sv3d at main (huggingface.co)
sv3d_p.safetensors,放⼊ models/checkpoints
运⾏ ComfyUI,下载⼯作流并导⼊
SV3D workflow.json
Stable Video 3D ⼯作流 ComfyUI 本体已经⽀持-ainewsto(这个⽹站有些可能会⽤到的依赖
项)
进去后会有提示缺失节点
点击右下⻆ manager
点击 Install Missing Custom Nodes
安装缺失的节点(有可能会失败,可直接点击链接下载,⼿动放在相应⽂件夹中:⽂件夹⼀
般为:cuntom_nodes, models/相应⽂件夹)
如果是使⽤的这个⼯作流:SV3D workflow.json
会需要下载 CLIP_VISION, VAE 和缺失的节点,放⼊对应⽂件夹
补充完这三个⽂件以后,可以刷新或重启 comfyUI,导⼊⼯作流,左下⻆选择⾃⼰的图
⽚,可能需要⼿动选择⼀下 CLIP_VISION 和 VAE,然后点击右侧 QUEUE PROMP 即可运⾏
SV3D Usage:
Only the simplest running method is included (using your own image 3D reconstruction)
First of all, download the third party ComfyUI (a third party tool for using stable diffusion,
model body is not included).
Install both of the following
Install ComfyUI – Comflowy
Install ComfyUI Manager – Comflowy
Download the official model at hugging face
stabilityai/sv3d at main (huggingface.co)
sv3d_p.safetensors, put in models/checkpoints
Run ComfyUI, download the workflow and import it into
SV3D workflow.json
Stable Video 3D workflow ComfyUI ontology already supports -ainewsto (there are some
dependencies on this site that may be used).
Once inside, you will be prompted for missing nodes
Click on manager in the bottom right corner
Click Install Missing Custom Nodes
Install the missing nodes (it may fail, you can download them by clicking on the link and put
them manually in the appropriate folder: usually: cuntom_nodes, models/appropriate folder)
If you are using this workflow: SV3D workflow.json
you will need to download CLIP_VISION, VAE and the missing nodes and put them in the
corresponding folders.
After adding these three files, you can refresh or restart comfyUI, import the workflow,
select your own image in the bottom left corner, you may need to manually select
CLIP_VISION and VAE, and then click on the right side of the QUEUE PROMP to run.