was probably a week before I joined the forum as came here after I messed up my library trying to "organise" it 

I was already using Carrara 5 by Evocia for a few months and Sketchup with iClone3 and 3DX2, started with those about January after getting a Dell laptop with a lite version of Crazytalk installed so going to Reallusion to get more content, of course tnen I needed to model props as theirs dear and Carrara 5 was on a magazine disc, my modelling was atrocious but it all seemed pretty cool back then, I bought Poser7 in October because DAZ studio crashed all the time on my pathetic laptop but Poser and Carrara worked

I finally did manage to install Crazytalk 8

it was hidden and a PITA to navigate to but I found a way to reset my count on their site in an obscure tab called "more information"

last time I had to do that it didn't work as the computer had to verify it

I will need to do the same for iClone6 if I install it

assumptions are being made based on personal usage

there definitely are people who prefer click and load and render readymade scenes 

I have encountered them many times wanting to know exactly what was used in promotion images down to the expressions and hand poses

there are also AI art users who tweak and redo generations many times inpainting, resubmitting the results to img2img, edit it in an image editor, combine elements, painstakingly remove individual pixels (me in Gimp) before loading it into Crazytalk replacing inner mouth and eyes with edited CT textures to match

loading a depth map of the image into Zbrush as an alpha and converting alpha to mesh

Zremeshing and reuvmapping that to fit the image

rigging bits of the mesh and weightpainting in Carrara and using soft selection to create morph targets

yes I actually do all that

 

so no one size does not fit all

I show the wireframe in Carrara at the end of the video 

I have the free opensource model D-iD and other programs like it use

it's called Thin Plate Spline but the one I have in Visions of Chaos is limited to 256x256 pixel videos 

which is like a postage stamp 

I personally prefer Reallusion Crazytalk 8 but that's a commercial product, it does however work great with DAZ/Carrara whatever renders and anything else with a face including animals

unlike D-ID

however it would be nice if someone could make a Thin Plate Spline plugin for DAZ renders!!!

or Carrara

an online version here
https://huggingface.co/spaces/AlekseyKorshuk/thin-plate-spline-motion-model

his x said:

WendyLuvsCatz said:

I myself use Crazytalk

for 3D Poserecorder isn't too bad but it works better on other people's faces in videos than my own crooked asymmetrical one

I also have a machine learning AI called Thin Plate but it only does 256x256 sized videos

I have CT8. It has a handy ability to make a 3D head and send it to iClone. iC can do a decent iRay render too. It's too bad RL is such a cloistered environment. DAZ is so conspicuously absent from their world view, I've wondered if there is an old feud in effect.

I have a copy of Poserecorder but have never road tested it. I wonder if the DAZ timeline would allow me to swap out the Poserecorder lipsync with, say, a Mimic lipsync.

I found ThinPlate on YT. A very nice implementation of existing tech, although the UI hasn't arrived yet. Do you remember when Adobe cancelled the Voco project in 2016? They were scared of the potential of the tech. Thin Plate is a good example of why. I'd love to see Thin Plate taken forward to become a deliverable tool.

Speaking of facial symmetry, it doesn't exist.

Visions Of Chaos which I use for ai on my computer has a very limited UI for it (no settings you can change)

but only good for postage stamp sized videos to stick in the corner as a narrator or Vtuber or something

Poserecorder saves a duf animation, you can edit/delete keyframes to add a mimic lipsync

also use Handbrake to encode the mp4 used as it doesn't like many codecs https://handbrake.fr/

WendyLuvsCatz said:

I myself use Crazytalk

for 3D Poserecorder isn't too bad but it works better on other people's faces in videos than my own crooked asymmetrical one

I also have a machine learning AI called Thin Plate but it only does 256x256 sized videos

I have CT8. It has a handy ability to make a 3D head and send it to iClone. iC can do a decent iRay render too. It's too bad RL is such a cloistered environment. DAZ is so conspicuously absent from their world view, I've wondered if there is an old feud in effect.

I have a copy of Poserecorder but have never road tested it. I wonder if the DAZ timeline would allow me to swap out the Poserecorder lipsync with, say, a Mimic lipsync.

I found ThinPlate on YT. A very nice implementation of existing tech, although the UI hasn't arrived yet. Do you remember when Adobe cancelled the Voco project in 2016? They were scared of the potential of the tech. Thin Plate is a good example of why. I'd love to see Thin Plate taken forward to become a deliverable tool.

Speaking of facial symmetry, it doesn't exist.

I myself use Crazytalk

for 3D Poserecorder isn't too bad but it works better on other people's faces in videos than my own crooked asymmetrical one

I also have a machine learning AI called Thin Plate but it only does 256x256 sized videos

I have access to all the Windows 10 voices and languages thanks to a registry hack I shared in another thread but will share the video again

(there is an executable to do it for you)

I can use them in Balabolka, iClone and Crazytalk

 

I can do text to speech with all the Windows 10 narrators

I had to do a registry hack but apps like iClone, Crazytalk, Balbolka have all the voices in them

https://www.daz3d.com/forums/discussion/comment/6511536/#Comment_6511536

WendyLuvsCatz said:

I had a go at Midjourney but really don't know what I am doing, that feed moves so fast

anyway I got some handsome elf lords in opulant robes and possibly wasted half my free goes upscaling them so better wait and think harder before I try again

meanwhile I will make Carrara cutouts of them and animate them with Reallusion Crazytalk

$10 a month for under 200 images is not bad though and will think about it ($15 AUD)

 

It's $10 for 200 minutes. If I don't upscale much or at all, I can get around 450-500 images. 

I had a go at Midjourney but really don't know what I am doing, that feed moves so fast

anyway I got some handsome elf lords in opulant robes and possibly wasted half my free goes upscaling them so better wait and think harder before I try again

meanwhile I will make Carrara cutouts of them and animate them with Reallusion Crazytalk

$10 a month for under 200 images is not bad though and will think about it ($15 AUD)

 

rnollman said:

What about CrazyTalk? You can export the character files to IClone. Not sure if you can export from IClone to DAZ.

I don't use CrazyTalk.

Forum search ;-)   It looks like you can use Crazy Talk with IClone.

What about CrazyTalk? You can export the character files to IClone. Not sure if you can export from IClone to DAZ.

Philemo_Carrara said:

I'm working on a project mixing face mocap and mimic like lip sync program. I don't use mimic because mimic understand mainly English phonemes and ranges from not very good up to very very bad in other languages. As I need my script to understand French, I'm in the process of implementing voice recognition VOSK python library with my OpenFace implementation to get a mix of both.

With this, you're out in front of the crowd. Good luck with it. This mocap field is moving very fast. Recently I have seen stabs taken at markerless full-body mocap. Mocap for the common man! Bravo!

I've looked more than once for TTS voices that are English spoken with a certain European accent -- French, German, Italian, etc. There aren't any. I await them.

My own first experience with lipsync was CrazyTalk in 2004. I was blown away by it, as I am by mocap today. Computer art in general has taken a turn toward photographic realism in the past 5-10 years. It puts me in mind of an old engineering joke. If I may:

An engineer and a scientist stand in a large room, side by side, with their backs to the wall. Facing them, back against the opposite wall, is a beautiful woman, scantily clad. The two men cross half the distance to the opposite wall, then half the remaining distance, then half the remaining distance, etc.

SCIENTIST: At this rate we'll never get there!

ENGINEER: Hey, this is close enough for me.

..............................................................

https://www.bitchute.com/video/MQGyir3J3ktN/

original image Wikipedia CC

using Philemo's CutOuts and some DAZ bits

lathed the vase and mapped the texture to it in Carrara, baked it with Inagoni baker and tidied it in Gimp

most of this was heaps of work in Gimp, Carrara the easy part

made hair density maps too there

am rendering an animation, will be a while

I rigged the cutout and animated the texture (with Crazytalk on the face)

Carrara render because it crashes Octane

4.6 screwed up on the AMD. That is why I upgraded to 4.12. Even Reality 2 with luxrender won't work on this current install on the AMD. I kind of gave up. Just doing single renders with 3Delight. Was planning on using this AMD for iClone, CrazyTalk and CrazyTalk Animator and not use Daz3D on it anymore after I get another pc that is better for Daz Studio. Reallusion licences are limited to a certain number of installs. Daz tends to destroy computers to the point of bsod doom and destruction. So then I have to contact Reallusion and request a seat removal so I could install on a new computer drive. I want to keep these two separate from each other to the point that if Daz dies, it doesn't take prisoners with it anymore.

my video is done

Wow! You have talent! This is fascinatring and the only one I've ever watched all the way through. Most bore me after the first short bit, including the original.

I didn't tag DAZ though because it was almost all done in Carrara 

they want DAZ studio

that very brief queens chamber shot was DAZ studio iray and took the longest

the baby is of course Reallusion Crazytalk as I said

thankyou 

another often overlooked bunch of Reallusion products are Crazytalk and Cartoon Animator (the latter you could also substitute OpenToons, Clip art studio, etc for)

these work well with DAZ studio iray renders as the base images.

rendering images as layers and a sprite like process it is surprising what you can create from a few images, a walk cycle can loop, a few gestures and the face image animated with Craztalk.

Hi there,

I've bought this plugin and I wanted to see how this works. 
But when I add a audio file (wav) into the Speech recognition tab, I'm always getting this pop up message which says 

'The requested lookup key was not found in any active activation context.'

What does that mean? 

Can you please check that speech recognition is working in your Windows installation. Please follow the tests here to make sure that your Windows 10 is configured to support speech recognition. 

I thought we said that Realusion used WSR, did we not? If that is true, then CrazyTalk and iClone would not work, not so? They are working fine. I have never seen an error from them since 2004. This is the kind of behavior I had hoped for in Anilip.

I began taking pains to set my record signal level at around .7, as .5 is "too low" and 1 is "too high." Anilip is very sensitive to signal levels. Then I switched mics, replacing my Blue Snowball with an Audio Activa AT2020. Last I ran Audacity Noise Reduction using the default settings. Anilip accepted this recording without errors. I am hoping to be able to repeat this. At the moment I'm surprised a microphone could make this kind of difference. I am learning about microphones.  My recording room is more reflective that it should be, and I am hoping that addressing this will also contribute to more reliable Anilip transcription.

The purpose of "training" your STT software, as we once had to do in Dragon Dictate, is to increase the typing accuracy. It doesn't help the software accept the signal. Anilip has a signal quality issue, not a transcription accuracy issue. True, the transcription accuracy is not very good, but it's as good as AWS. It's less important in lipsync than it is in typing. It is not our primary concern.

I came here to get help, and hopefully to learn. Instead I find myself trying to convince you that these problems are not of my own making. This isn't helping. Lipsync is crucial to the future viability of DAZ3D as an animation tool, and Mimic has "one foot in." It is important that Anilip's idiosynchrasies are identified, one way or the other.

 

I don't think it's fair to say we won't help you. You have reported a speech recognition problem and we have given you the possible causes. Of course we want to continue to develop and improve the handling of the software in the future. Your input is very helpful for this. However, we cannot shoot from the hip and have to look closely at possible improvements. So far, we haven't had many customers who have reported speech recognition problems. But maybe that's because the focus is mostly on TTS. We are already thinking about configurable error handling, or rather to handle such 'errors' as notes and rather to log instead of displaying them in the popup. But please give us some time to properly design these things. Your problems have arrived with us. No reason to convince us.

Hi there,

I've bought this plugin and I wanted to see how this works. 
But when I add a audio file (wav) into the Speech recognition tab, I'm always getting this pop up message which says 

'The requested lookup key was not found in any active activation context.'

What does that mean? 

Can you please check that speech recognition is working in your Windows installation. Please follow the tests here to make sure that your Windows 10 is configured to support speech recognition. 

I thought we said that Realusion used WSR, did we not? If that is true, then CrazyTalk and iClone would not work, not so? They are working fine. I have never seen an error from them since 2004. This is the kind of behavior I had hoped for in Anilip.

I began taking pains to set my record signal level at around .7, as .5 is "too low" and 1 is "too high." Anilip is very sensitive to signal levels. Then I switched mics, replacing my Blue Snowball with an Audio Activa AT2020. Last I ran Audacity Noise Reduction using the default settings. Anilip accepted this recording without errors. I am hoping to be able to repeat this. At the moment I'm surprised a microphone could make this kind of difference. I am learning about microphones.  My recording room is more reflective that it should be, and I am hoping that addressing this will also contribute to more reliable Anilip transcription.

The purpose of "training" your STT software, as we once had to do in Dragon Dictate, is to increase the typing accuracy. It doesn't help the software accept the signal. Anilip has a signal quality issue, not a transcription accuracy issue. True, the transcription accuracy is not very good, but it's as good as AWS. It's less important in lipsync than it is in typing. It is not our primary concern.

I came here to get help, and hopefully to learn. Instead I find myself trying to convince you that these problems are not of my own making. This isn't helping. Lipsync is crucial to the future viability of DAZ3D as an animation tool, and Mimic has "one foot in." It is important that Anilip's idiosynchrasies are identified, one way or the other.