Voices - TTS - with Vidnoz

For the OneSixthual dolls I need to assign each doll a particular voice, which fits to its character. I have experimented with several online tools, and here are my results.

Vidnoz for TTS Voices

I have used Vidnoz since 2 years, it was my first venture into AI-generated animations. At first I used their avatar mode: uploaded image of a doll, then made the doll speak by selecting a voice and added the text prompt. The result was then that the avatar spoke. 

This approach had the disadvantage that the avatars were quite unflexible, a bit stiff, and only the head moved together with the speaking. Mouth movement was OK.

So then I devised another more complicated method: Vidnoz also has a mode where audio-only can be generated, without the video. I used this to select from the voices the ones I liked, then entered the text prompt. The resulting wave file I then merged with DreamFace to an existing video snippet. 

Recording the voice

Vidnoz offers more than 1200 voices. It is quite difficult to find a suitable voice, one needs to try one-b-one. In the voice library one can filter by age:

Also one can filter by style: there are 34 different styles (e.g. newscast, whispering, etc.) and 17 scenarios (e.g. Training, Chat, Poetry, etc.), but it is still quite difficult to find a suitable voice, as these categories sometimes are not clearly describing the voice. This filtering is also available in editing mode.

And one can filter by language. It is actually great fun to select a language other than English, e.g. French, and then enter an English text - the voice then speaks English with an accent! 

And one can customize each voice: 

A great feature: one can preview the voice recording without needing to pay - this allows experimentation before the paid final version is recorded.

Using DreamFace for merging video with audio

The app DreamFace costs $30 per year, and credits can be purchased: 30 cr for $5, 1000 cr for $100. 5cr are added automatically weekly, but they expire after 1 week. I forgot how many credits a merging of the video with the audio costs... So I upload the video snippet, then the audio wave file, and I let the app do its magic. 

There are a few issues:

• DreamFace does not support NSFW videos, and it refuses the upload of a video which does contain such content. Therefore, one must only upload just a video close-up of the face that i supposed to talk.
• Multiple faces in a video are not supported. There must only be one clearly visible face.
• When the audio is longer than the video, the video at its and is added playing backwards. This ensures a smooth overall video, but it makes the video unusable in a context where its ending is pre-determined. Therefore, the video snippet should always be longer than the audio snippet.

Dreamface also does have its own voices, and they appear to have a great selection. But I have not used them yet, as I like the choices offered by Vidnoz.

Creating the final video

I then paste / merge the resulting video+audio into my overall video timeline, overwriting the previously non-speaking part with this new speaking part. For this I am currently using Microsoft ClipChamp, in the free version. Works fine for me.

PixelDojo seems to work again - albeit veeeery slowly

Since today was a news announcement on the PixelDojo site, without acknowledging or mentioning the big problems of the past few days, I followed my usual routine and tried the WAN 2.2. As the new instructions stated, one has to select "Image to Video LoRa" option in the drop-down menu on the left side. Then all the other options also appear: quality (480p), number of frames (100), fps (8),  LoRa (default), interpolate to 30 fps. 

Here are again the start and the end image:

 

And here are two results with this prompt:

"nsfwsks woman stands at pole, looking into the camera, camera zooms out until woman is completely visible, she is dancing at the pole with sexy movements"

  

Both are slightly different, which is the usual as I have set the seed to be random. There are a few AI artifacts, but I do not mind these very much. At least these videos both look mainly like what I wanted. So PixelDojo WAN 2.2 is just the very best for my needs.

(no) Alternative to PixelDojo: Kenerate AI

Since the recent demise of PixelDojo I am looking for alternatives. Kenerate AI was indicated on reddit as a possible alternative, so I gave it a try. 

Here are my results:

Price: $15 one time fee gives you 1300 credits. Which mean 1 cr costs about 1.15 Cent.

Trying out WAN 2.2 which I always used with great success on PixelDojo. 8 seconds cost here in 480p resolution 45 credits = 51 cents, which is 6.38 Cent per second of video.

Compare this with PixelDojo: 12.5 seconds for 18 Cent, which is 1.5 Cent per second of video. I did use there 5 seconds duration, but with a framerate of 8fps, which created then 12 second long video clips.  These were automatically upscaled to 30 fps.

Options at Kenerate: start and end frame.

The end frame was not used, and the requested zooming of the camera was not done, even though I had indicated this in my prompt.

So Kenerate AI is in no way a match for the lost PixelDojo capabilities.

Here is the input that I used:

Prompt: woman looks into camera, camera zooms out until she is in full view of camera, she stands at pole, moves slowly in sexy way

Start and end images:

 

Summary:

• Kenerate WAN 2.2 is more then 4 times more expensive than PixelDojo WAN 2.2.
• Kenerate does not obey camera prompt instructions.
• Kenerate does not obey last image as last frame.

Therefore, this site gets a NO as a PixelDojo replacement.

PixelDojo is now unusable

Since November 2025 I have used PixelDojo, and mostly the WAN 2.2 "spicy" model.

On 21 August 2026 I got the last message from the PixelDojo Discord server about news on PixelDojo. When I now tried to access them, the server no longer exists.

And when trying PixelDojo, WAN 2.2 functionality appears to be crippled. No more start and end image. 

And it is more expensive: instead of 3 credits, it is now 5 credits for 5 seconds.

I tried WAN 3.0 - just no comparison. Uploaded 4 pics, and the video just jumped to the pics instead of a smooth transition. This is useless for me, and expensive - 75 credits for 30 seconds 

Here are discussions and speculations on what has happened with PixelDojo:

https://www.reddit.com/r/PixelDojoAI/ 

Update:

It turns out that the WAN 2.2 which I had used up to a few weeks ago has now been renamed to "Video LoRa".

I will give it a try - OK, all the old functionality of WAN 2.2 seems to be there in the "Image to Video LoRa" option, to be selected from the drop-down menu.

---

My opinion is that this demise of PixelDojo is a temporary technical problem. Everything runs very slow at the site, but it appears that the functionality is back, and that new features are being added while the previous features are preserved.