ฝัง
- เผยแพร่เมื่อ 16 พ.ค. 2024
- After testing a dozen new AI tools that popped out recently, I've chosen for you the most efficient one, and once again I'm sharing my tutorial about it!
Let me know your opinion on this new 3D modeling tool we are allowed to play with for Free :)
Important links:
CSM AI: 3d.csm.ai
Download Blender: www.blender.org/download/
Chapters:
00:00 - 00:30 - Video introduction
00:30 - 00:58 - Setting up the AI
00:59 - 02:55 - Generating 3D model from image with AI
02:56 - 03:27 - Enhance your 3D model
03:28 - 03:56 - Refined and preview model comparision
03:57 - 07:00 - Important tips
07:01 - 07:16 - Exporting models
07:17 - 11:23 - Prepare models for 3D printing
11:24 - 12:24 - Overview
See you all in the next vid!
![🔴 [TRỰC TIẾP] Tứ Xuyên (Trung Quốc) VS U20 Thái Lan | Cúp VTV9 - Bình Điền 2024 | JET STUDIO](http://i.ytimg.com/vi/bguJ41PcW0c/mqdefault.jpg)
Hey folks, I've found a way to make the A.I. models more detailed (like WAY more detailed). You can find a full tutorial on my channel, I hope it helps you achieve better results. Have a good one!
Thanks for the video! Can you tell me what the back ground music was please?
@@danlane9917 Sure thing, however I’m currently on a holiday trip. Give me a week and I’ll check the soundtrack name once I’m back in my studio :)
@@architech_the_engineer Thank you :)
ok go@@architech_the_engineer
I created some characters in Daz. I tried uploading 6 images from different angles but instead of creating one model it attempted to create separate models for each image then it used up all my coins and now it is forcing me to sign up. How do I create one model from multiple images with a free account?
This is exciting. Your clear step-by-step presentation merits 5 stars. I'll be following you closely.
That is so nice to hear, thank you very very much!
@@7HUNDERGOD-fx5ct nothing motivates me more 💪 thank you!
I hadn't even thought of having AI generate a starting model and then go in and finiah it up ourselves. Exciting stuff for sure!
Totally agree! Thanks for the comment!
very informative and straight to the damn point! awesome! would lov to see some of your blender tutorials!!! AI is scary AF!!! SKYNET HERE WE COME BABY!!!!
Hahah thanks a lot! I’m glad you enjoyed the video, stay tuned coz more tutorials are coming! 😎
Awesome video! This will help me and many others who have been looking to take advantage of our 3D printers.
Thank you so much! I’m working on more content that might be helpful on that topic, I hope you’ll enjoy it too!
If you haven't already taken advantage of your 3d printer without this, you probably don't deserve it.
In 5 years this is gonna be pretty dope
Amazing video ! Great explanation! Thank you so much
Thank you for such a kind comment!
thank you sir it helped me for my freelancing stuff good job and keep it up bro
I’m very happy to hear that! Hope my videos could help you in the future as well, cheers!
That was awesome. Thank you sir!
You’re most welcome!
so excited to try this out 🙏
That’s awesome! Let me know how it goes!
Great vid very useful thank you.More vids and tips on advanced sculpting please.
Thanks a lot! I will gather some useful tips for you with upcoming videos, stay tuned! :)
This is very nice! Though after tinkering a bit, I think one simple thing they might be able to do that would be incredibly useful would be to allow me to edit those four generated views from different angles that it uses as an intermediate step. A couple of the things I've tried had views that would have been great if I could have just gone in with a simple image editor and stretched and squashed parts of the images. Ideally I could download them and do some full blown editing work in my local toolset.
Great idea!
@@architech_the_engineer Would you prefer editing the preview mesh or the input views (or both)?
@@csm_ai Hello there! So nice to hear from you, I’m glad to see you’ve found my video! I hope you approve this little tutorial of mine :)
As to your question - while editing both options would be great, I think if I were to chose, I’d go for editing preview model. Manual edits on different views would require a decent understanding of the model and spatial imagination. I suppose in many cases it would lead to more confusion. That being said, I can’t wait to see what’s coming next with the app!
Or even better if it let us replace them with pictures from different angles
@@csm_ai honestly being able to just replace the preview pictures with actual pictures of the thing would be great
If CSM would allow you to submit multiple images , that alone would improve the AI understanding the puzzle 🧩 pieces and how to assemble
Might also be a huge legal liability.
To elaborate - If I take 20-30 photos (similar to photogrametry) of my iRL truck, and submit them to the AI, it can assemble those 20 photos together to make the model
Bingo. That really is the thing i think we're all waiting for. Because the thing I really want is the ability to make detailed interiors from AI and I just imagine a tool in the near future where I can take a dozen pictures of the same room from different angles and import them and it figures out the space in a more detailed accurate way. This feels very much like "check back in 6 months" kind of technology.
@@MatthewWells_officialIs there any way I can reach you to discuss more use cases for this?
Nice video, congrats and thanks!
Thank you!
Yes please. Could you go more into detail on how to refine/sculpt in blender
Thanks for the comment, I will definitely share some tips!
Pleas make video. love this video @@architech_the_engineer
@@Morten_M working on it! Stay tuned 😎
Outstanding tutorial!
Breathtaking comment! Thank you so much! ♥️
Great job. 🎉
this was great, i'd love a video focusing on sculpting models in blender!
Roger that, thanks for letting me know!
ive been looking for something like this for a long time, it would be really cool if u made a vid on how to imporve the models in blender
Thanks for letting me know! Have a good one!
This actually came a lot sooner than I expected, but more or less in the way I'd expect it to work when it finally released. I can't wait to see how it evolves in the next 3-6 months
That makes the two of us 😎 thanks for the comment!
Lol you are not talking in years now but months... the next 3-6 months !!!! this is crazy
We already have an improved model in the pipeline and will have an update to share soon! Happy Cubing :)
@@Digit0wL absolute bonkers. Can't believe how fast everything is growing at
@@csm_ai you're too effective at this
Oooo so many possibilities
Definitely!
wow man u deserve a lot of subscriber.
Thank you so much, I’m putting a lot of effort into making videos 🙌 It makes me very happy to see you enjoy them!
Wow! I've got to check this out.
Definitely!
Can't wait to see more! Like
The more is coming! Thank you! 😃
@@architech_the_engineer I have also suscribed. Thanks
@@oscargonzalezbarber1119 so happy to hear that! Welcome aboard 😎
Excellent video. Thank you!
Thanks!
Super video.Thank you !
I’m glad you like it, thanks for the comment!
Thanks for yours tips
Always happy to help!
Yess please another vid please!!!
Will do! ♥️
Thanks for this vidéo !!🤯🤯
Thanks for the comment!
Could I take (say) 3 or 4 pics of an object for the AI to use or doers it only take one pic & then extrapolate?
this is great! i'm wondering how this would react to simple fonts? say for example the "thrasher fire font"? will the app give the letters volume?
That’s a great question! I’m gonna find out!
I've been looking for something like this for a year now! I'd like to print a bust of my mom, can't really scan her or use photogrammetry because she passed away and don't have multiple perspective photos from the same moment so 2D images are all that I have left!
This is absolutely amazing, I love the idea and really hope you achieve what you planned! If during the project you face any struggles, I would be honored to help you, feel free to come back and contact me. Best of luck, you really made my day with that comment, what a beautiful and noble way to use this technology! Thank you Sir, have yourself a great day and stay awesome
@@architech_the_engineer thank you very much! hope to be able to get good results after first optimizing the photos and manually removing backgrounds. But really appreciate the availability ❤️
I was looking for something like this in around may. I wanted a 3D print a bust of my newborn daughter. I was surprised there is nothing available, since head shapes and faces are pretty similar, at least compared to the search space of general text to image. So from a mathematical point of view it should be an easier task, but I guess there is just not as much training data available.
I ended up paying a sculptor on fiver and I think the results were better than anything I could have achieved myself or any of the tools I found, including this one, impressive as it is. It just lacks details so without a texture the models are really simple
@@powertomato that's the way to go! especially with sentimental things it's worth it to pay for someone who knows anatomy and can make it look exactly like the photograph. AI can't get those details in at all
@@powertomato could you link the sculptor you used?
I will try it
Seems like it might be a good tool for creating reliefs to be cnc carved. Don't have to worry about the backs of objects with that. I'm interested what it does with multiple objects in a scene. Will it just chose a single object to focus on? Or will it generate stuff for the entire scene?
Basing on an experience with chair image (1:07) it focuses on one object and cuts it out of the background. You can actually find the cut-out image, to do this go to export in upper corner (like when downloading a model) and find INPUT IMAGE. It will show what part of the photo was used to create a model :)
@@architech_the_engineer interesting. I would probably have to break up an image into individual objects, then assemble the elements back together... Thanks!
@@DeusExNihilo Always happy to share a tip, personally I would assume the same approach :) good luck with your project!
I remember I saw some time ago some tools to generate reliefs from the map. There is no need to use image-to-3D tools for this task I believe:)
@@3Dklase yes, it is possible to generate a relief directly from a bitmap, but it relies on the pixels' color value (lightness/darkness) so for a regular image where brightness doesn't necessarily correspond to depth, it will be distorted. Under certain lighting conditions it's not too bad, and with an actual depth map (generated from a 3d model) it will work. The AI is fundamentally different in that it's actually trying to imagine the shape of the objects depicted in an image.
thx just what i needed
Happy to hear that!
great video, thank you
I’m honored to hear that, thank you for the comment!
Looks cool. there is a room for improvement, but its def better to have a model to start with than make it from 0. Can't wait to see more!
The more will come sooner than we all expect! Thanks for the comment!
Super Top 🔝 Thank you very much 😊
Thanks for the comment! 😇
Soooooo nice video, i was looking for something like this to make an .stl out of my drawings, so I can work later on merge cube, I'll give it a try for sure
Great idea! Thanks for sharing =)
Good class, thank you`
Pleasure to hear that, thanks for the comment!
I would love to see how you sculpt the models and how you might make them into action figures
Noted! Thanks for the comment!
wow thanks for this video! I wondered when AI generated STLs will flood the market... seems like just now!
Seems like the future is now 😏 I’m happy you enjoyed the vid, thanks for the comment!
Maybe it’s best to choose a well known object or one with plenty of historical references to get best results. We draw best based on what we already know so that might mean the same for AI.
Finally I can pass the univerity's finał projects! Thanks a lot. You are an exellent teaches with lovely voice❤
This is the nicest thing I’ve heard so far on youtube, thank you so much! ♥️
Multimodal models will create almost perfect models within the next 4-6 months, and they will probably also be game-ready with correct topology, and most likely weights and skeletons for animations.
Finally, I'll be able to save multiple months of time for my solo project thanks to these.
sell-out...lol
This gives me an idea I'll try. Use iPhone LIDAR to scan an object, render it in Polycam or similar and then use the rotary on a CNC to make replicate the object. Thanks
Can you use this for products?
please, an enhanced model tutorial :)
good to know AI helping to 3d now can advice how to use in procreate
I haven't had good luck with asymmetrical car parts like a headlight assembly but I'm wondering if the upgraded plan which allows Video to model would help... we will see for $20 a month ill give it a try
Let us know what you think!
is there any way to produce Lalique spirit of the wind ? front view only would not give enough information
Its free for NOW. Going to be subscription when they release out of beta testing.
I hope they add the option to add in more photos so that it has the information's to create a a better model.
Hope to see that in the future!
so basically a scan?
I doubt any of these comments are real. There is not a single comment mentioning how the HD render takes DAYS to be ready. I uploaded an image few days ago of a game character. Well, the HD render is not done yet and it has been 3 days already.
I think it's likely they are having humans make the models behind the scenes
Amazing
Thanks!
Very interesting. Please show some more prints, and try to push the limits. How well can it do busts in general?
Let’s find out! I will push the limits and let you know with next videos what else can we achieve 😎
@@architech_the_engineerThank you, sounds great. Busts of scholars and scientists are good sellers to educational establishments.
I like the skew monsters
What about if you have multiple perspective images?
Is there a tool to generate 1 3D model from multiple 2d shots of the same object?
Yes there is! It’s called photogrammetry, but you require a lot of pictures to get nice results
I'm surprised how good it turned out
Same! Can’t imagine what will be possible in the next 6 months 👀
@@architech_the_engineer :D
Very interesting, is there a version where we can upload multiple images?
Not yet I’m afraid, but I hope to see it in the future!
Very interesting and great video. Unfortunately been trying now to have an image of my own generated and all I get is a backlogged error and a prompt to upgrade to the paid version.
You are correct, I’ve just tried and also experienced error. Probably too many people using the site, hopefully they will fix it soon
A good first step but nowhere near what I would call a "detailed" model and frankly even an entry level 3d modeler would do a better job. Will be interesting to see how this matures though.
You are correct, there is always room for improvement, but keep in mind that half a year ago the most advanced free AI tools would generate only the simplest shapes like a bucket. There is definetely a field for progress, and that is why I am here to deliver you the news when another level will be achieved. Thank you for your opinion and have a Great Christmas! 🎄
Actually it's a good tool for a 3d modeler... like 80% of the work is done just needs some details defined... a bit of sculpting and it's good to go
@@hd-be7dithe clean up process really wouldn’t be worth it, it’s getting better though
@@evil1knight for a 3D printing purposes the heavy and messy topology is not an issue at all, for other purposes it is good to run some retopology on the model, but it is still better to have some starting point to save some time
@@hd-be7di Only if you don't care about your topology so, not for any real professional use. If you're just futzing around with your own stuff sure.
Yeah it absolutely needs to let you upload multiple angles and stitch them together
Can they not allow multiple images to capture other angles of the object?
what if I already have the side, back and top view? can I input multiple images or only one? It would be better to have the option to provide more than one image of the same object.
Agree, however for now there is only 1 input picture available :(
Vapourwave Goldmine.
This year i am about to learn 3d, but after watching video like this, I am scared😢
Should I learn this or should do something else 😔
The thing is, I were very excited to learn 3d because it's really fascinate me but now i am in dilemma 😢
Please advise me, what should I do,
Is learning 3d worth now?
We doing these on agisoft metashape with better precision. what is difference?
This isn’t photogrammetry. It uses a single image.
any way to make it look more detailed? for now it's missing all the good stuff you want for your 3d printed model, like eyes, mouth etc
I’m happy you asked, because I’ve found a way to do that and I’m just now finishing a video about it! Hopefully it will be ready in few days, stay tuned!
I have 3d models I created in daz but the topology is not good for 3d printing. If I upload multiple angled images will it create a 3d print ready model or will I be back to square one and have to clean this topology in blender? If so this program is useless for my purposes.
You can’t upload multiple images, however I have another video (the last one on my channel) on how to get incredible topology details based on textures. If your models have textures prepared in DAZ, you can use that tutorial to achieve much more complex topology for printing
👏👏👏
🫵♥️
Looking good! I'm trying to turn a few ideas of miniatures I had into ones I can print, it's looking good so far but the site has taken over an hour to show the preview of the model.
That’s awesome idea! The wait time can be extanded if there are many people using this feature, and right now, during the holidays, we could assume there are a lot of folks playing with this tool. Most of my models were generated in 5-8 minutes, but it was about a week before Christmas. I guess you kinda have to wait it out, but I hope it will be worth the wait :)
i wonder if this would work for making figers like for dnd
can i choose this idea as my project for my bachelor's.need some assistance on this
..
Go ahead, if there are any questions feel free to ask in the comments! :)
Great video!
Thank you kindly!
I go to that link and I never see that for my picture. it says intermidate mesh final mesh.
What if we have multiple views of the same object? Say 4 sides of a car and an overhead shot?
This sounds very similar with the 3D scan (photography)
The biggest difference is that it only takes the info for a one image
My question is it's possible to make multiple images to convert in 3D with this AI tool
For example to get one image for the fond back left right Ect for imagine you want to comvert in 3D model
Yup photogrammetry requires many views but we can go from text, image or a sketch.
How about small object's like jewelry ring or pendant
Can this run locally
can you bring the models into a rendering engine?
Definitely yes! The render may take some time since the geometry is dense, but it is totally possible
Exciting, but I still wonder where people will be in the end....it's all just a gimmick because it's all interesting, but we're breeding our competitor here who will replace us at some point😊
What formats or file types can these be exported as from the website?
Nevermind! Just went on their site
lol and finished the video haha
@@itsMBWAAA hahah, no worries, let me know if I can help 😁
**Laughs Nervously..**
I am just waiting for the moment when someone uses a Games Workshop box cover to print out some miniatures with the help of this AI. Ensuing legal battle will be so legendary, it will put anything within 40k lore to shame.😂
What 3d printer are you using there?
It is my good old Creality Ender 3v2, the print quality is a bit lousy because the printer would really use some repairs. It made hundreds of prints for me in the past, but I can’t afford to get a new printer as for now, so I guess it will stick around for a while 😄 Still, I am very happy with all the work it’s done over the years
As much as it's admirable that AI can do such things these models are just fairly disappointing even if you'd like to edit them further for better result... has this become a real thing with actual quality, we'd see it behind a solid paywall. I think 3D artists should feel rather safe.
Safety will last us 3D artists a year or two max. This tech is evolving at neckbreaking speeds
Models like GPT-5 and Gemini 2 will be able to generate 3d models within the next 4-6 months, and they will cost similar to current ChatGPT Plus price which is $20-25/month depending on the country for a lot of uses.
And they will most likely be able to create correct topology, weights, and skeletons for animations, making them completely game-ready for any major engine or 3d software.
It will not even take a year for them to be better than the best 3d artists at almost everything, and they will not be restricted with the exception of NSFW and maybe similar 3d models (open source AI models will also do those 4-8 months later).
Hmmm... we're not there yet. AI is basically generating a "shroud" meaning it will not pick up detail, just surface gradation of the most defined areas.
The video version would probably be better, or no? Would guess it would be able to generate much better objects as it has information for all sides?
I would also assume that, but I have not tried this technique yet
thanks
You’re welcome!
It would be great if you could upload multiple views of the same object.
its free to train the IA and when its done its going to be paid, user are free trainer to turn the model into a paid one xD
Very interesting, thank you for the video.
Too bad that the AI tool is still a bit underdeveloped, because just as we can guess from a front view of a car what the rest looks like, especially if we recognize the vehicle model, so should the Ai.
The AI should recognize that it is a hydrant and that it has perfect technical shapes and not artifacts floating in the air. But in a few months or a few years, it will be able to do that too. Just like when I see a picture of a hydrant, I can model a perfect model of a hydrant.
Good thinking, I expect to see much improved results in following months
I recently watched a documentary on Netflix about World War 3. A lot of the old footage had been cleaned up and colorized by AI, but it still looked very old. I think in a few years, when AI has a three-dimensional understanding of the world, it will be able to re-render old footage in the highest quality. If a Ford T appears in the old footage, then it can recreate it perfectly using the data on the internet. And if she doesn't recognize a vehicle exactly, she can come up with a technically and visually perfect model. @@architech_the_engineer
This is too low res to be useful. It looks good when it's textured because once you will look at the 3d model, you will be disappointed.
However, it could be useful as a starting point to either build on it by adding details, use AI to add details, or something like TopoTransfer in Houdini to deform a good known 3d model so that it looks more like the target mesh (like a pose)
Does generated models must be shared with the community, can they be private?
I’m pretty sure you can disable sharing with the community, but I’d have to double check
excuse me, how long until that. Thank you@@architech_the_engineer
@@architech_the_engineer excuse me again, have you checked?
I'm surprised it doesn't allow you to upload your own sides, bottom and top views for precision
There's 0 detail on those, the only "details" come from using textures on the models, there's no point on printing the 3d models
why i can`t sign up with my gmail?
unfortunately it is not very reliable and sometimes takes forever to create awful results. i am eager to see how it will improve over time and if we can get more control over the result.
However it’s possible that these AI generated 3D models may not be protected by copyright including if you heavily edit them yourself so you can use them as references but you need to fully model assets yourself or from human made assets you have permission to use.