Creating Human 3D Models Has Never Been Easier Because of AI! (Outdated)

In this video I will show you how you can generate a 3D Character model from a single image using PIFuHD research paper code that I Fixed, Updated and simplified on Google Collaboratory.
This method is really simple, any beginner with no prior experience in 3D art or Coding can do it.
UPDATE: The Google Colab script still generates the 3D model and normal mapping, but unfortunately, the model preview is not working for some reason. If you encounter an error, it is most likely due to this issue. I have attempted to fix it multiple times since the video was released. However, every time even one library updates, some functions stop working, and I have to fix them again. As a result, I have decided to stop maintaining the PIFuHD Colab file. If the code stops working due to a new update, I will not deploy a new fix. This is not because I do not want to, but because the effort required is no longer worth the gain. The code breaks almost every week, and it is not sustainable for me to keep fixing it.
Google Collaboratory used in the video: colab.research...
Don't forget to like and subscribe, it really helps me out!
Software Used: Blender, Adobe Premiere Pro, Photoshop
Follow me on TikTok: www.tiktok.com...
Follow me on Instagram: / 3dbulletgio
#blender #vfx #cgi #ai #artificialintelligence

[@3DBullet]

A few other KZbin tutorials show this method to generate the 3D Models, but their code is old, messy and mostly doesn't work because of the new dependency updates, however I fixed the code and compressed/simplified it to only be 2 Cells in Google Collaboratory and you require no prior experience in coding. Even the original PIFuHD Google Collab doesn't work anymore. UPDATE: The Google Colab script still generates the 3D model and normal mapping, but unfortunately, the model preview is not working for some reason. If you encounter an error, it is most likely due to this issue. I have attempted to fix it multiple times since the video was released. However, every time even one library updates, some functions stop working, and I have to fix them again. As a result, I have decided to stop maintaining the PIFuHD Colab file. If the code stops working due to a new update, I will not deploy a new fix. This is not because I do not want to, but because the effort required is no longer worth the gain. The code breaks almost every week, and it is not sustainable for me to keep fixing it.

[@spatil689]

Can u provide link to ur collab?.Please 🙏🙏🙏

[@3DBullet]

@@spatil689 It's in the discription 🙌

[@Marc-hw6sb]

@@3DBullet it just became outdated as well.

[@nahlene1973]

Thank you for the tutorial. I think if you used Mid-Journey or Stable diffusion to generate the initial human image it would be even better because you can specify it to be in the T-Pose or A-pose. Being in standard pose would make it much easier to rig later. However this would a great method to fast generate some extra actors. Currently my workflow is to generate a human portrait with mid journey and use Character Creator's headshot plugin to transfer the face to a standard body, then export to Blender. All rigging (including the face) would be precise and automatic.

[@jaimepena2052]

My character looks like the fossilized remains of a crushed samurai warrior

[@aquamirrorX]

wow, randomly recommended by youtube this is legit awesome combined with Stable Diffusion character generation😱 definitely gonna try it out for my future projects, thank you for your work!

[@mreflow]

Is this still working? I keep getting a "FileNotFound" error while trying to run the second step after uploading my image.

[@3DBullet]

That error usually means that the 3D Mesh couldn't be reconstructed, try using a different image where the human is clearly visible ( Preferably with transparent background )

[@mreflow]

@@3DBullet Thank you!

[@blenderarchviz]

You can adjust project mapping to mesh with same texture and fill unmapped faces with similar textures

[@3DBullet]

I'm working on an effective way to texture the models to at least use them as background characters, I will post a tutorial if I find a way to get good results.

[@blenderarchviz]

@@3DBullet i trust in you 👌🙂

[@WisdomShrimp1]

How do I add the colour/texture of the PNG to the model? (I just started using blender to make 1 model)

[@user-gq2bq3zf1f]

I'm always thankful. There are clothing photos from various angles. Is there a way to naturally change the clothes of the character in the video?

[@YukiTSnowyOwl]

It won't work for flat pencil and digital drawing images

[@tellthex]

I could imagine using this to create a character, skipping the sculpting process and just start retopologizing

[@jasmine8182]

Here from Reddit. Thanks for this. Any alternatives which work for any image, not only human bipeds?

[@3DBullet]

There are no alternatives as far as I know.

[@LinusKVisuals]

You could try to generate a depth map of the image and simply extrude it to get a 3d effect.

[@3DBullet]

@@LinusKVisuals The process actually gives you a normal map that you can convert into a depth map.

[@suryakanatanayak9104]

Thanx bhai love from Indian ♥️♥️♥️♥️

[@g_niac]

here from Reddit. sub’d and liked. thank you for sharing this

[@3DBullet]

Thank you!

[@harisaisenthilkumar2880]

same 👍. Pretty cool and scary at the same time ngl

[@brandireaves9538]

i keep getting errors in my 3d mesh area of the program you are talking about

[@dailyartro]

I have one question that there is error comes out and no obj file, only png that I got

[@DanFrederiksen]

cool. does it get textures as well? this could make renderpeople obsolete. generate 2000 people in a bundle

[@christelord9120]

Thank you so much for the tutorial and cleaned up code. This worked wonders the first time I ran the code, but strangely stopped working thereafter, no matter whether I’m using a copy of the code or how many times I disconnect and restart the runtimes. The error is the same each time: filenotfound. No obj was downloaded either. But you said it’s still possible to download the render? Please help. Thanks!

[@brandireaves9538]

on the 3D Mesh area it keeps saying my file is not found what am I doing wrong

[@M4Pxls]

Really nice, thanks for updating it! Is there a way to use generated normals and color image?

[@3DBullet]

Thank You! Currently I have not found a way to effectively map the image to the model.

[@MKDMusicPrestigeWorldwide]

Hey thanks for making this. Definitely will help with consistent 2D characters! I am getting the following error message though "Maximum call stack size exceeded". Any advice? I tried both png and jpeg along with different sizes. always the same error message.

[@kingpin4152]

I keep having errors with the second cell I uploaded a .png idk why even should get an error since it's just two cells and it didn't give me an error with the first cell

[@RogenLothena]

It's still easier to do it yourself. It's too early for AI and people are really overestimating it. Countries should have very restrictive laws on artificial intelligence. AIs need to be developed for another 5-10 years before they can actually make a proper professional 3D model.

[@SmashtoonGamer]

Any videos you recommend for doing it yourself?

[@kumar....4243]

Why single image !! As we can give multiple images, to get best result !!

[@juanc.saravia8668]

Thank you for the Collab notebook, do you know or can make a video on how to run it locally on your own computer with an nvidia card?

[@3DBullet]

Running on your local machine can be done, But I can not make any promises about a video on that, all I can say is that I will try.

[@paullacki8482]

@@3DBullet I think there would be a lot of interest in this. All the people running stable diffusion would be a target audience. (Me for one). Create a 2D character in SD, then making a 3D version that can be easily posed in Blender certainly helps with character consistency issues.

[@Gil84]

doesn't work.....lots of error messages....

[@streamology]

how limited is this? is it only for humans?

[@3DBullet]

It's currently limited to humanoid creatures. If you gave it an image of a human statue it would most likely work.

[@Harrisonjnr]

Crazy

[@narrativeninja]

not clean so it will still be difficult to have a good model but i know it will get better

[@Kemeslim_kel]

would love to see you texture it

[@3RDEYELOVE]

Now how do u add textures for skin and clothing?

[@squidboimusic]

Great to see another fellow Brave user! 🖖

[@bgmiishorts123]

not work

[@zaidxdev]

that will take our jobs D:

[@stefanogrillo6040]

Gud results

[@GreenMagic0]

Is it free?

[@3DBullet]

Yes it is, you can use it as much as you like.

[@MR.DEVIL0x]

Can u help me it gives me error

[@3DBullet]

Sorry about that, the code stopped working due to Google Collaboratory updating the default python version, I am working on a fix as we speak. Check back tomorrow and hopefully it will work.

[@3DBullet]

Update: The 3D Model generation works as expected, but I am still working on getting the model preview render to work. (if an error pops up, it is most likely just the 3D preview not working.)

[@MR.DEVIL0x]

Ok thank u

[@duffmcshark]

@@3DBullet what’s the max size for image?

[@3DBullet]

@@duffmcshark it doesn't really matter that much, the process automaticly crops any image to 512x512 pixels.

[@unknown6781]

he's Georgian I bet

[@Spookykooks-xv2bj]

tried it 5 times now and it just wont work. THANKS FOR NOTHING!!!

[@burhanvfx]

It give me error

[@BCABPP]

same this part kzbin.info/www/bejne/omXIlKOnerGll80 doesn't let me choose a file :-(

[@vaclavmuller]

kzbin.info/www/bejne/sKvKdKKvjt6ohJI

[@bikurifacebook4553]

buuuuuuuuuuuuuu No module named 'pytorch3d' lol