[Feature Request]: some feedbacks about load 3d node from Reddit

[jtydhr88]

Is there an existing issue for this?

• I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

I got some great feedbacks about load 3d node from reddit, here is the original message:
Thanks for your work! I use this node everyday in a VFX-like scenario to load glbs, objs and fbxs. Here's my feedback:

*It would be great if we could set the camera focal length. Right now is either ortho or perspective, but perspective is way too wide angle. The ability to set it to eg. 85 mm would be appreciated.

*A nice addition would be the ability to feed an image as the background. I use this node to position the 3D object on top of a background so that I can use it as a controlnet or low denoising guide, so I have eyeball it. If I could see the background directly in the node, it would streamline the process a lot.

*Would it be possible to add a soft shadow under the object so that it looks like it's contacting with a surface? This is useful when you process this image through a ksampler: it gets the crude shadow and interprets the object is actually on top of a surface and improves the look of this shadow.

*The integrated mask does not ouput anything. I would expect a classic alpha channel with the empty space in black and the object in white.

*This node stops showing the 3D model randomly when working, especially with heavy VRAM operations like flux generations. It shows a black empty square although the model is loaded. If you select the node and render again, the object becomes visible again.

Proposed workflow

N/A

Additional information

No response

┆Issue is synchronized with this Notion page by Unito

[jtydhr88]

just assigned to myself

[jtydhr88]

One more feedback:
Currently testing your loader with some archviz models. From my point of view:

• I miss a FOV camera adjustment option.
• The 3D viewport gets black after image generation.

Nice 2 have:

• Clip planes for Depth. I lost a lot of precision in the depth output. Had to switch to normals, which worked fine too with the union control net.
• Viewport (camera) aspect ratio matching the width/height dimension

I could not load some larger untextured models. Is there a vertex buffer size/object count limit?

You created an awesome and really helpful node. It saves so much time.
I really appreciate the effort you've put in.

[Amorano]

Noticed you ignore my Material support question ;p

Add:

• Material Support (MaterialX/OpenPBR)
• • Just make a "material" input and have a node that can change said material entries so its not a million inputs
• Support for loading 3D meshes from raw data instead of needing to "upload" everything aka IF's Trellis support -- directly consume GLB/OBJ (FBX)

[jtydhr88]

fov and mask implemented in #2116 and comfyanonymous/ComfyUI#6308

[jtydhr88]

One more feedback:
It seems I found a bug.
Exposing the Image width and height does not work. The image resolution seems to stay at the prior resolution (before exposing - 'convert widget to input'). This results in an error at the sampler node, even if the new resolution matches the prior regulation.

About the large model: Tried to figure out again which model it was. It was not the size of the model, but the .obj file format. After exporting the model to fbx I could load it without a problem. I downloaded it from Sketchfab. Here is a link to the file including the fbx and max file: https://we.tl/t-CkxbgIDqll

[jtydhr88]

one more feedback:
It would be great to be able to position the model once, then get access to all of the different images / materials, eg get the image, the depth and the wireframe to use as part of a ControlNet setup.

Maybe have image as is, showing the current setup ( eg with grid option ), and have 4 more outputs representing the 4 current materials - they only get rendered if they are connected.

Great job with supporting 3D file formats natively in ComfyUI, looking forward to future progress on this!

Having a camera setup node that could be fed into the Load 3D node ( and also other Load 3D nodes to sync them up ) would be a cool feature. This input could be set up to create turn tables, with advanced setups creating fly throughs etc.

When you make changes in the viewport in a Load 3D ( eg rotate the model ), and then clone it, the old camera position is copied. If you run the workflow and then clone it, the new camera position is copied.

To test, create a new Load 3D and import a model. Rotate / move the model in the thumbnail. Now clone the model - the old camera view is copied.

Run Queue ( even if it gives an error ). Now, re clone the model and the updated camera view is copied correctly.

[zerotransform]

Greetings. I run a indie motion capture studio and I'm looking for the cleanest way to use pure animation for pose data in comfy UI. I can render the output, but I feel that there is data loss when going from 3d animation source to a render as input.

[moebiussurfing]

hey @jtydhr88 , what do you used to code this 3d viewport? do you know if we can do / link something similar from a cpp app/code bind to a similar comfy node?

[jtydhr88]

@zerotransform can you give me more details about your case?

[jtydhr88]

@moebiussurfing sorry but I know nothing about cpp.
The code of load3d node is under https://github.com/Comfy-Org/ComfyUI_frontend/tree/main/src/components/load3d and https://github.com/Comfy-Org/ComfyUI_frontend/tree/main/src/extensions/core/load3d, and (BE) https://github.com/comfyanonymous/ComfyUI/blob/master/comfy_extras/nodes_load_3d.py

However, right now I am refactoring the load3d to support new vueNodes system, PR is #5765, after it merged, a lot of thing will be changed