Working well. Wonder 2 is gentle, Lighting and Dust/scratch are working good. Dust/Scratch tends to remove some detail, like flowers on a skirt. But did get most of the annoying white spots from a scan. Lighting is brilliant.
Can I have Wonder2 in T-Video! 
I made a GIF while switching from local result (way less detailed) to cloud one:
We are working on storage improvements for the application. NeuroServer has opened doors for new AI models, so we are increasingly aware of how much space the models can take, and optimizing storage usage where we can.
Is NeuroServer taking up space before or after the processing completes? It should only use memory while processing is requested.
While NeuroServer enables large models to run locally. it does not reduce the computational requirements for processing so older GPUs will experience lengthly process times.
What happened to Wonder 2? Since the update to TP v1.3.0-v1.3.1, Wonder 2 can no longer be used for RAW files. See screenshot. Before the update, i.e. in v1.2.1, this was possible in the cloud without any problems.
Hmm so there is a major difference in quality then.
Can someone from Topaz comment on this? Should the expected outputs from Wonder v2 cloud vs local be the same quality? Is this a bug or a feature?
@Lingyu ?
From our testing local has similar quality with minor changes in detail. Examples shared in this thread differ more than the testing set. Iâll run batches this weekend with varying hardware to identify if itâs hardware related.
Hi there,
In Activity Monitor, Neuro Server takes up almost 4GB of memory.
4GB is nothing these days.
I run on RTX 3090 with currently newest 591.74 Studio driver (newer one was, apparently, recalled), on Win10 ESU.
And Wonder v1 doesnât even start locally now (just prepares the image forever) in TP v1.3.1.
It is if youâre on a 16GB machine.
You canât realistically expect to run cutting edge AI models on a 6 year old machine that was never that good to begin with IMO.
yeah, well, thatâs just your opinion. Iâm not on here to discuss the merits (or lack of) regarding the machine or how good it was to begin with. There was no need to add that little quip in at the end. Letâs just leave it at that.
Win 11 Pro desktop PC. TP 1.3.1 - Standalone. Processor = NVIDIA 5090 (32GB VRAM) GPU. 64GB RAM.
Download/Install Metrics:
15:10 mins. to download TP 1.3.1 @ 40-44 Mbps consistent download speed.
15:40 time at which Neuro Server began download & extract process.
29:45 mins. = Total download & install processing time.
Image: ~ 6 MB jpg (image captured with 1.5-2 mp mini keychain camera - HOTU) of 3D pop up strawberry and floral arrangement greeting card.
Processing Used: TP 1.3.1 Standalone. Wonder2 Local to PC. 2x upscale. No other TP filters. Took 3 minutes.
Output (I may be missing something but I donât see much difference b/wn the Before/After viewsâŠ). Didnât try same image with remote server (aka, âCloudâ) :
And thatâs just the beginning!
Face recovery with the âcreativeâ model is failing to run with either v.1.3.0 or v.1.3.1 - the application just hangs on processing, seemingly indefinitely as far as I can tell.
Very likely.
Iâm already looking forward to using huge models once the technology is fully developed.
Youâre a 55%.
Always use 4x with Wonder 2 (if that is valid still).
Topaz Photo 1.3.0/1.3.1 is broken for the Super Focus model. It displays an error message indicating that the model is missing. It worked perfectly before the major update.
Here are the logs
Logs.zip (32,8 Ko)
Error with V1.3.0/1.3.1: Dust & scratch crashes: missing model? Everything works with V1.2.x. Here is the log.
2026-03-01-12-12-5.zip (25,3 Ko)
I just checked my install. Recover Faces, Super Focus, and Dust and Scratches all are working normally. This update was a particularly large download, so I suspect the install partially failed on some systems.
