Traveling to a different country and forgot the remote at home. How do I use the Shield? I have an android phone.
The shield turns on but since it's a new location it does not automatically connect to wifi.
Hey all, hoping some networking folks can help me on this.
I work remotely for a company based in Montreal and connect to my office PC via a remote desktop tool (HP Anyware/PCoIP). Normally when I'm physically in Canada it's smooth, but connecting from India (Punjab, Airtel Xstream Fiber) the session is choppy. Dropping to ~6fps with the client itself reporting <1Mbps bandwidth and 260-270ms latency.
So bandwidth is clearly not the bottleneck. It's latency and especially the jitter spikes that seem to be killing real-time protocols like PCoIP even though raw MTR looks OK on a short run.
Questions for the sub:
Not trying to game-optimize here, just need a stable low-jitter connection for work. Any insight appreciated.
Landing at 11:55 PM so need a hotel for that night only before I continue on my journey via a Taxi the next morning.
I am guessing I won't get out before 12:30-12:45 am so hopefully a hotel that allows late check-in.
Budget is $90USD to $120USD. Can go slightly up if there is a better option
Tried looking up some Aerocity hotels on booking.com but there is such a huge difference between good and bad reviews that I do not know which ones to trust.
Trying out a few LoRA's on top of the character LoRA I trained but any realism LoRA I stack on top completely changes the body proportions. The face remains the same but it generalizes the body.
Without any extra LoRA's Krea2 retains different body types and heights that come from the character LoRA but they all change to the standard 1girl as soon as I stack anything on top.
Any LoRA + weight you can recommend so I get Realism concept without completely destroying the character?
Using Krea2 RAW model + Turbo Lora at 0.6 With Dual Sampler. Sampler 1 is set to 8 steps, 1 CFG, euler/simpler and Sampler 2 is set to 10 steps, 1 CFG, dpmpp_2m_sde/sgm_uniform starting at step 5.
Setup:
krea2_raw_bf16 (UNet, ~24.4GB)
qwen3vl_4b_bf16 as the CLIP/text encoder (~8.5GB)
Wan2.1_VAE
krea2_turbo_lora_rank64_bf16
one character LoRA
64 GB RAM
Running this in ComfyUI on an RTX 5090 (32GB). Model + text encoder combined is ~32.9GB, just over what the card can hold so both can never stay resident together. ComfyUI reloads the qwen3vl_4b text encoder from disk on every single generation, adding more seconds per prompt beyond actual sampling time.
Questions:
Please recommend a quantized version of the text encoder (or any other encoder that works with Krea2) that's more efficient/smaller but doesn't compromise on quality.
Anyone running this exact combo (krea2_raw + qwen3vl_4b TE) on a 32GB card without the reload-every-prompt behavior? What's different about your setup?
Any other Krea2-specific VRAM tricks that keep both TE and Model resident at once?
Can share full console logs if useful. Thanks!
Traveling to Chandigarh with some of my computer components and will need to rebuild the PC. Please recommend some good computer assembly shops where I can buy a PC Case, UPS, and some peripherals.
I know there is a computer market in Sector 20 but all those shops seem branded and selling pre-built PC's and laptops.
LAPTOP QUESTIONNAIRE
Country
Budget
Are you open to refurbs/used options?
Screen size
Weight limit
Purpose
Form factor
Intended usage
Desired battery life
Please list, in order of most important to least important, the priority between Size, Weight, Performance, Battery life
Info/Requirements
Trying to caption my dataset for a Krea2 LoRA but the Auto Caption is not working. Here is my settings :
The log just says "Starting job..."
Is there another way to caption the images? I would prefer to use a custom prompt.
I've got an approx 10 hour layover in Munich this Sunday (land 6:35 AM and depart 4:45 PM) and I'm planning to leave the airport and go see Marienplatz for a few hours instead of just sitting around MUC.
Travelling on a Canadian passport with one ticket through to my final destination with bags checked through (so no bag reclaim in Munich i am assuming). Just carrying a backpack with me.
My plan is roughly:
Has anyone actually done this exact kind of thing? long layover in Munich and have gone into the city and come back?
A few specific things I'm trying to figure out:
Trying to figure out if a 4-5 hour window in the city is realistic and safely buffered or if I'm underestimating how long airport processes take on the way back. Appreciate any real experiences especially from anyone who's done MUC specifically with a non-EU passport.
Thanks!
Posting this in case anyone else is hitting the same issue with ai-toolkit-perceptual on Windows.
When enabling the identity anchor (identity loss weight > 0), training fails with:
PermissionError: [Errno 13] Permission denied: 'C:\Users\xxx\AppData\Local\Temp\tmpXXXXX\tmpXXXXX'
The full traceback points to onnx2torch's safe_shape_inference.py which creates a NamedTemporaryFile, writes to it, then tries to read it back. On Windows the file handle appears to stay locked between the write and read, causing the permission error.
What I've tried:
Added toolkit folder to antivirus exclusions
Disabled antivirus entirely
Redirected TEMP environment variable to a different drive
Deleted all cache files and restarted
Latest git pull
Environment:
Windows 11
RTX 5090
Python 3.11
onnx2torch 1.5.15
Is anyone else hitting this on Windows? And does anyone know if there's a fix for NamedTemporaryFile locking behavior on Windows in onnx2torch?
Posting this in case anyone else is hitting the same issue with ai-toolkit-perceptual on Windows.
When enabling the identity anchor (identity loss weight > 0), training fails with:
PermissionError: [Errno 13] Permission denied: 'C:\Users\xxx\AppData\Local\Temp\tmpXXXXX\tmpXXXXX'
The full traceback points to onnx2torch's safe_shape_inference.py which creates a NamedTemporaryFile, writes to it, then tries to read it back. On Windows the file handle appears to stay locked between the write and read, causing the permission error.
What I've tried:
Added toolkit folder to antivirus exclusions
Disabled antivirus entirely
Redirected TEMP environment variable to a different drive
Deleted all cache files and restarted
Latest git pull
Environment:
Windows 11
RTX 5090
Python 3.11
onnx2torch 1.5.15
Is anyone else hitting this on Windows? And does anyone know if there's a fix for NamedTemporaryFile locking behavior on Windows in onnx2torch?
Just picked up MSFS 2024 and looking to 'acquire' some content beyond what's in the standard edition. I used to play MSFS 2020 a lot and simplaza was my goto website but they don't host MSFS 2024 content so I am here looking for help.
Is skybound considered a safe website? I remember sometime back there was this thing about different discords and reddit communities being compromised and potentially infected.
Also, I found AeroHub's Telegram but not their website or discord. Are they considered a safe group as well?
Are there any other websites I should keep an eye on for airplanes, sceneries and plugins?
Hey everyone, I've been dealing with this issue for a while now and have spent hours tonight trying to fix it with no luck. Hoping someone here has seen this before. Setup:
The Problem: Xbox app takes forever to load, games don't show in the sidebar, and when they do, clicking Play results in "Launching" then immediately back to "Play" with no game window opening. Microsoft Store is also slow and unreliable.
What I've tried:
Current State: Every time Gaming Services is freshly installed, sc query GamingServices immediately shows:
STATE: 2 START_PENDING
(NOT_STOPPABLE, NOT_PAUSABLE, IGNORES_SHUTDOWN)
It never progresses to RUNNING. Force killing the process and rebooting doesn't help — it goes right back to START_PENDING after every fresh reinstall. Error codes encountered along the way: 0x8007041d and 0x80073cfb.
Any game install is stuck paused in queue and Resume is greyed out because Gaming Services can't start.
Has anyone seen Gaming Services permanently stuck in START_PENDING like this? Is a Windows repair install the only option at this point?
Hey everyone, I've been dealing with this issue for a while now and have spent hours tonight trying to fix it with no luck. Hoping someone here has seen this before. Setup:
The Problem: Xbox app takes forever to load, games don't show in the sidebar, and when they do, clicking Play results in "Launching" then immediately back to "Play" with no game window opening. Microsoft Store is also slow and unreliable.
What I've tried:
Current State: Every time Gaming Services is freshly installed, sc query GamingServices immediately shows:
STATE: 2 START_PENDING
(NOT_STOPPABLE, NOT_PAUSABLE, IGNORES_SHUTDOWN)
It never progresses to RUNNING. Force killing the process and rebooting doesn't help — it goes right back to START_PENDING after every fresh reinstall. Error codes encountered along the way: 0x8007041d and 0x80073cfb.
Any game install is stuck paused in queue and Resume is greyed out because Gaming Services can't start.
Has anyone seen Gaming Services permanently stuck in START_PENDING like this? Is a Windows repair install the only option at this point?
I am very new to metamask so trying to figure this out. I sent SOL to my metamask wallet. But the payment processor where I am trying to buy stuff only accepts 'Base' or 'Polygon' USDC.
I am trying to swap SOL for USDC but it keeps saying 'Unable to get a quote right now. Try again.'
How do I do it?
Woke up to see FenLight AM not working. So I switched to Umbrella and connected my RD account and seems like I did get the infringing media notice on an episode but then it kept searching and played it.
So I'm wondering did I just get lucky? Or does Umbrella do any local/internal torrenting in the background.
I worry because I am not behind a VPN.
I'm using the Switch (Any) node from ComfyUI Impact Pack, which takes 3 image inputs and selects one based on an integer index (0, 1, 2). The problem is that bare integer selection isn't very user-friendly. I'd love a dropdown that shows something like:
(0) Photoreal
(1) Sketch
(2) Material
...so the user knows what each index actually represents, rather than having to remember what 0, 1, and 2 mean. Ideally I want a node with a labeled dropdown whose output is an INT, which I can then pipe directly into the Switch node's index input. Has anyone solved this with an existing node from a popular pack (Comfyroll, Easy Use, rgthree, etc.)? Or is a small custom node really the only clean way to do this?