Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
well then you import them into Zbrush, UV map them with UV master, subdivide and create texture from vertex colours
reduce the subdivision back to the original after creating the map
Thanks for the tips, Wendy.
Just wonder if any other program would help, like
Carrara, 3D Coat or some of Realusions programs.
After last big crash of my Windows computer and necessity to reformat it,
I am not installing any programs, than really needed for my hobby.
Guess, need to take a chance.
DOUBLE EDIT - No, Not Solved
Error message says I have a likely extension conflict
Trying to begin. Cannot.
Gang, I cannot get started. I have installed Stable Diffusion on local. I have installed controlnet. Followed online tutorials. But Stable diffusion does not seem to take into account my picture prompt even though I have enabled check. Neither the pose nor the clothes nor the... seems to be being taken into account. I've tried other photos and other figures and other poses, but only get even worse results. I followed some online videos.
I uploaded a picture of Leonardo in a suit, no tie, in basic sort of sitting pose. The result is nothing like it except for gray suit. I had 'canny' checked, and 'balanced' checked for this test.
DOUBLE EDIT - No, Not SolvedEdit #3 - Now it is solved.
Error message says I have a likely extension conflict
Found an error message. Not able to understand it, but something must be installed incorrectly, I assume. Hmmm.
Yes, it is just like another program knowledge, so learning curve from the beginning was high for me.
I have stopped using Automatic 1111, because my graphics card has only 8 GB of VRAM
and it was not sufficient to run Stable Diffusion XL models.
In ComfyUI, I have not reached controlnet level in knowledge, yet.
I have discovered, that FBX files from Shape-E does not preserve vertex colors.
Below is a sofa geometry created by Shap-E
Thought Solved, but NOTNow it is solved.
I went to extensions and did an apply and restart. I then ran from image2image. Better.
This might be rough, but at least it did what it was supposed to do with no errors.
ERROR Identification Progress
The new error log says I have a likely conflict among the extensions.
Has this happened to anyone elese? Does anyone know which ones are likely?
ARGH - I am off to the tech forums for Github, etc. Frustrating.
Many more errors. This one is typical.
Error running setup: E:\AI\stable-diffusion-webui\modules/processing_scripts\refiner.py
Traceback (most recent call last):
File "E:\AI\stable-diffusion-webui\modules\scripts.py", line 741, in setup_scrips
script.setup(p, *script_args)
TypeError: ScriptRefiner.setup() takes 5 positional arguments but 99 were given
Yes, I got some warnings with ComfyUI on Linux,
but at the end it works.
No idea about your errors, sorry.
In the meantime, I have UV mapped in Blender sofa from Shap-E
in 3 different ways, but Cube Projection gives the best results, so far.
The Piggie like it, as well.
Daz Studio render.
Some shapes generated today...
Need to find a way to convert vertex colors to another format,
that I can save in the file.
My 1st Render-to-AI-Output
Installation Solved (as far as I know).
Rendered G3M Santa reading a prop magazine sitting on a cube and plane against Orestes HDRI sky preset. Cube and plane with brick and roof shaders respectively. Loaded the result in Stable Diffusion with Controlnet. Prompt = Santa sitting on a chimney reading a magazine at night.
And here is the result. OK, the result sucks. But I am in the game!
Installation Solution
Some of the extensions for Controlnet had conflicts. I don't know for sure how many, but it appears to have started working correctly after I disabled the SD-CN-Animation extension (or something spelled similarly).
I got errors with some ControlNet extensions, it's very fussy which version is used with which checkpoint
you need the ones that match
I had some I really like that now won't work with anything but did before, all I can suggest is getting the models and extensions from the same source,
Automatic 1111 has stopped working on my PC altogether so I am little help, I am using Fooocus, NMKD and Visions of Chaos
yesterday and today using textural inversion i trained 3 awful embeddings that give caricatures of what they are meant to, I see the resemblence but they are horrible
my face, my late cat, my art (old scribbles I found in a book, truly awful but worth a laugh)
Nice experiments, Wendy.
like all experiments I learnt 3 ways NOT to do it
it would have takem 80 hours with 100 000 steps so I only did 10 000 steps (8 hours each)
more VRAM probably exponentially faster
better using the online services for such things like Leonardo.ai
Yes, I also use Shap-E online.
I like experimenting with the current technology.
Thanks - yes, I had it working again. Added some more extensions. And now it is not working again. Argh.
But here are some recent results with Santa Claus. Apparently, AI really likes hot curvacious female Santa Claus, and that is OK with me. (See first attachment). But in this case, I want a more traditional Santa Claus. So I put (Wilford Brimley) in with the prompts. Much closer to what I was looking for. (See second attachment). See my posts above for the original Daz render - really just a blockout.
Now I have to go back and try to figure out what is not working with what. But I am getting an error with controlnet again.
These are great! You have inspired me. My goal for my hobby this week is to have me as Santa Claus sitting on a rooftop reading a magzine, using an AI program. I hope to share the results so folks can get a chuckle.
Error Identified
The problem was the extension for 'eyemask.' I removed that, and now controlnet is working again. So here is a first pass simple grid of four results using my Daz render as a blockout in controlnet and then prompting for santa claus sitting on a chimney reading a magazine at night on a rooftop. No need to include Wilford Brimley this time because no hot young pinup result.
--------------------------------------------------------------------
And that is derived from the same Daz render included below, with the first Stable Diffusion attempt. Some progress, to be sure.
you can use Roop or one of the othe faceswap extensions to put your face on him
I am in effect still using Automatic1111 but in the Visions of Chaos python environment
I get the advantage of Xformers and reduced VRAM that way too albeit slower
I found I can continue training the embedding too so getting something closer to my face now
I'm not faulting your workflow or shaming your progress, but I am curious to know what you're getting out of Stable Diffusion with this project that you aren't able with Daz? The only stark difference I see between these two images is the lighting, and what SD is generating can be achieved in Daz with a photometric point light.
- Can only answer the first question, why not just use Daz, with a question. Why does anyone use Daz Studio if the resulting images can be created with paints, pencils, cameras, digital image editors, and other 3D programs?
- On the nod to my feelings. First, thanks for the effort to be considerate. We can all use more of that. Second, I don't feel shame for my Santa posts. I myself posted that they suck. See above. Must crawl before walk and there is no shame in that.
- Main question asked, what am I getting out of experimenting with AI?
...... To see for myself what all the fuss is about. For systems like Stable Diffusion, I definitely see one of the ethical issues, referring to its caveat that it is only for research purposes yet is being used for profit, breaking the terms of use that they promised to contributors while gathering the data. But to the extent that contractually paid for systems are coming from people who honestly paid Getty Images etc for the data, I also see great potential. I will be glad to have some background knowledge to use those programs when they get full releases..
...... To see what the aforementioned potential is. And it is amazing. I am only barely dipping my toe in, yet the speed at which edits can be made to existing images, or new starter images generated, is amazing compared to using a fake 3d photography studio like Daz Studio, or Blender, or... combined with Photoshop, Gimp, etc.
...... To add another tool to the toolbox. For example, my Santa images were created combining a Daz Studio render of Santa sitting on a cube with an AI processor, not using AI instead of Daz Studio.
...... To know things for the sake of knowing them, even if I never use Stable Diffusion again.
awesome stuff ...
had me thinking if I could do something similar ( Santa on a roof ) in MS Paint... ended up with this after 5 minutes..
must incorporate this stuff with Carrara ,, hmmmm .
Fair point.
I can't argue with any of those points. I briefly dabbled with Stable Diffusion for the same reasons.
Carry on, and best wishes. :)
Beautiful Santa.
RE: Carrara. Oh, the missed opportunity for Daz3d (company) becomes all the more glaring. Carrara is the perfect tool for basic staging, editing, rigging, and posing as part of an AI workflow. Carrara would be perfect to block out concepts, arrange, rig, and pose on the fly, add an arch or a custom tree in the background, then recursively submit to the AI processor. Apparently, Blender is already being combined in this way. Daz Studio does not have a true vertex modeler or a tree editor or a landscape editor, etc. A combo of Studio/Bryce/Hexagon would be great, but they allowed Bryce and Hexagon to whither as well.
This guy has a tutorial on a plugin for Blender. Studio could do this particular example (pumpkins) because he only uses primitives. But Blender or Carrara could add an arch or edit a custon tree or a custom landscape. Studio can only load premade such things.
I would love to transfer some of my custom toon Carrara characters (the Marx brothers) to AI but Carrara's FBX exporter is old.
Yes, all good. Cheers.
Perhaps some young comic will be inspired to reboot the Carry On series and make Carry On 3D.
Stable Diffusion with ControlNet has a menu for loading rigged FBX, JSON, and similar models. But my custom toon Marx brothers do not appear to transfer correctly from Carrara FBX to Stable Diffusion.
A video using my trained Textual Inversion Embedding
a recent pic of me for comparisson
I added my embedding to try, the words to trigger it are myself, AND wendyvain,
beginning and end of prompt
it will sometimes work with just wendyvain in the prompt but for specific use something like "myself, a woman doing such and such or wearing such and such etc, wendyvain" should trigger it
it was trained on Dreamshaper7 specifically but I think any SD v1.5 model should work