DAINAPP
patreon
Dain-App 0.45
🕑 Added 2020-10-15 00:03:51 +0000 UTCMirror: Link
Note: The folder name was wrongly named DAIN_APP Alpha 0.35 instead of 0.45; I noticed it after the upload.
This is the first release I managed to run on my computer, since I managed to get myself a GTX 1650 \o/ Thanks all for the support, no more need to keep testing the app on colab. It's been really useful for development.
Change Log
I been mostly focused on fixing the broken stuff that the batch rendering caused.
- Fixed CLI
- Disabled half-precision by default. Right now there is pretty much no point in turning it on (it hardly get free memory and it is slower).
- There is two models in this version, the best and renamed best_so_far to experimental. Deleted the other ones. Experimental sometimes is better to animation. It also usually remove the blue artifacts.
- Detect scene changes should work, not sure if it was broken before.
- Did try to improve memory usage for each frames (without splits)
- Did try to fix the random spikes in memory.
Working On
- Fixing Resume Render
- Fixing Mode 3
- Adding tooltip to the GUI
- I managed to interpolate between models, so in the future you can use 70% of model 1 and 30% of model 2
Still not working On
- Fixing experimental interpolation (3X, 5X, 7X)
Comments
GRisk
Is so you can use the aplication on CMD instead of the interface.
Been a big fan but this one crashes on launch compared with the other versions so we have had to downgrade. Geforce 2070 Super 12gig ddr5. Have had to go back to 0.40 for stability issues. Keep up the great work!
Congrats on GTX1650 btw whats CLI?
GRisk
I did try to update a few libraries, but the new Cuda SDK still have some problems, so i'm not sure it will be able to run on this version. Any crash_log.txt from running on a RTX 30X0 is appreciated.
GRisk
0.45 already have some tensor core support, but is hardly noticed. But in 0.46 we should start to see some real improvement using Tensor Cores.
And there's still a compatibility issue with RTX3090
So when will it support tensor core (even only for Ampere)
