Image to 3D in Seconds Using Trellis

[@васяпупкин-ъ9п]

What a wonderful time we live in!

[@MarkieWay]

This is awesome! unfortunate I use a Apple M3Max and I can’t seem to find a way to run this tool on the comfyUI that I run locally. Have any ideas?

[@alexvillabon]

@@MarkieWay Not sure. I gave up on trying to make it work inside of comfyui or standalone before this new repackaged release. You can still use huggingface though.

[@paulaaron4]

Hey Alex, waiting on your comfy ui course. Do you know when it would be out?

[@arash5550]

ModuleNotFoundError: No module named 'torch' Something went wrong, consider removing code/trellis_init_done.txt and venv folder to re-initialize from scratch.

I didn't manage to make it work on my machine, it throws some errors which are difficult for me to understand. It could be connected to vram, but I am surprised because I should have 11 vram, and it needs only 8, according to your post.

though. maybe it was something wrong with my instalation, now I tried again and now it is downloading a big dependency, and before I don't remember doing that. Maybe everything is fine. We'll see :) And, btw, I see and agree it is not production ready, but it can be a huge help in previs and blocking stage, especially when you use it with Unreal Engine.

what a s**tstorm, I am still struggling with it and by hand adding dependencies, cuda driver... maybe you were just lucky

[@KT-pv3kl]

dont give up hope. we are working on the bleeding edge of AI generation with those tools that are held together with prayers and duct tape. things are bound to break or randomly not work with some machines for no good reason. i had to set up my old comfy install for sdxl images at least 3 times before it worked as intended ( just to randomly break one day needing a reinstal). best of luck to you!. maybe try backing up your old confy version and do a complete reinstall from scratch with no other custom scripts and nodes. i noticed many scripts and custom nodes interfere with each other causing all kinds of errors.

@I'm not giving up hope, I like the tool, but the failed installation took up the whole day yesterday when I could have been doing more useful things. As long as it's not available in a handy package that I don't have to compile myself and deal with a bunch of incompatible Python modules, I'm good to play with. So far, the online demo works well for my needs.

[@nmatthes2927]

Is the installer using virtual env stuff with python and co. so that it does not mess with anything on your pc?

[@MMMMeeeeMMMM]

Back with AI video!!!!

[@guillermosepulvedaf]

ModuleNotFoundError: No module named 'gradio'

[@octopuss3893]

+1

[@imonedesign]

Same

[@kobayashi19891805]

need 16G vram @@

[@alexvillabon]

There is a way of launching low vram. I believe you can go as low as 8gb. I cover how to do it in the video. Cant remember off the top of my head what the name of the launcher is.

[@aegisgfx]

So on a 3060 12g Im waiting over 10 minutes for a single generation at 12 sampling steps.. (Edit1: ran the FP16 on that same setup and it was 10 seconds for a generation, I have no idea why the huge difference) (Edit2: second run of the run-gradio.bat was a lot faster, maybe only the first run has slow generations?)

[@alexvillabon]

The dev that put this package together explained how the f16 would be better for lower vram systems. Glad you got it running smoothly now :)