For Support Issues

[viral cloak]

(both nvme and m.2, which are technically different things)

And using a 7700X which while not the best, is no slouch

[trim dock]

ah it asked if i wanna keep models but well imma stick with fresh installs

i see why it might take a long time

[viral cloak]

Haha

To be clear its not like hours just like, maybe go make a coffee type of time frame.

[trim dock]

always the tinies

1.3mb read and 3mb write

Thanks again bub

[viral cloak]

My pleasure

[trim dock]

i just bought the card a few days back had one bad crash had to fully power cycle as my cpu was throwing error on post 🫠 (was streaming nier:automata on discord) and rolled back to 8331 havent had crashes yet🤞

[viral cloak]

The folks in here have helped me out tons, I only started playing around with this local AI generation like 8 months ago or something

[trim dock]

god the flashbang

Nice it installed comfyui v0.4 without any issue

Once again thanks a ton

[viral cloak]

No problem, my pleasure!

[exotic mortar]

Is there a config text file that contains the path to the checkpoints? The path is used to put civit sd models into doesn't seem to recognize my checkpoints in 3.0.1

[merry trout]

@glad portal Latest AIPG should work with alchemist now

3 more people

[glad portal]

It is now working with my A770! Noting that I have to disable my IGPU. Also it seems that while a few basic things are working smoothly like image generation/editing, however having tried Wan2.2 in ComfyUI, it disconnected (this failed) during generation every time, so I wonder whether it is not utilising the full 16 vram of the A770 or something

[merry trout]

How much RAM do you have?

[glad portal]

I have 64gb ram

[merry trout]

damn, that should be plenty for everything to work fine

Ultimately the crashing would depend on how AIPG does wan2.2. in its raw state, wan 2.2 is two 28gb models plus a 6GB text encoder plus a bunch more. those two giant models can be lossily compressed down to 7gb per and swapped around, and things should work then if you don't go overboard with the resolution/framerate, e.g. sticking to 0.4-0.3mp and 81 frames i think was the default

[merry trout]

@exotic mortar The latest AI Playground should work with alchemist GPUs now

2 more people

[merry trout]

@barren pollen new AIPG should work with alchemist now

[merry trout]

@surreal merlin You probably saw by now since I saw you reacting in the comfy thread but, aipg should work now

the newest aipg

[surreal merlin]

Hi Vik yes thanks for letting me know. Aipg v3.0.1 was working with driver 8250 but then I updated to the latest 8425 today and it seems to break AIPG and comfyui at the last stage of generation. I'll try downgrading the driver to see if it fixes it. The AIPG Agentic chat has been quite interesting to work with

[surreal merlin]

@merry trout Downgrading the driver to 8331 WHQL appears to have fixed it

[glad portal]

I will try this, as Wan2.2 image-to-video has failed every time (also at Vae Decode on Hunyuen 1.5). Also in order to get ComfyUI to connect, I have had to deactivate my IGPU, but I wonder whether this compromises the amount I can offload the workload from my A770...

[merry trout]

I haven't tested much but, the newest driver seems to cause segfaults (3221225477) when training sdxl loras under some circumstances (just plain opening blender, among other things I haven't nailed down). So it could be causing segfaults in other cases too like wan
I wouldn't strictly pin it on the driver, just that it happens with the new one

[exotic mortar]

Is there a config file we can look at to see where AIP is looking for checkpoints?

[torn solstice]

What hardware was the latest 8425 not working? Wanting to reproduce

[torn solstice]

The settings.json in /resources provides ComfyUI and other parameters. json files in /resources/presets provide the model and node dependencies needed and provides the workflow parameters for each node (essentially a ComfyUI API export of a workflow)
Back it up before editing. You can break the app easily here

[surreal merlin]

Hi Bob it's an A750

[zinc parrot]

it's also crashing comfyui when processing the last step on my A770 16Gb driver 8425 and playground 3.0.1

[glad portal]

So..... Has anyone with a 1st gen Alchemist and the latest updates/drivers, managed to successfully generate a video through ComfyUI?

[torn solstice]

For dGPUs especially with lower vRAM A750, B580 you will need to add back the reserve vram parameters into the settings.json. They were removed in 3.0.1 alpha and impact performance and stability. We are adding this back and to the app settings. Not sure this is your problem but it does impact your hardware. (Be sure the last item before the end bracket does not have a comma

"comfyUiParameters": [
"--lowvram",
"--reserve-vram",
"6.0"
],

[torn solstice]

The last step on all image gen uses a node for safety checking. I wonder if something is breaking with this node.

Try this.

1. Launch AI Playground
2. Click ComfyUI button in image settings
3. Drag this image and run

See if it finishes. Bypass safety check node see if it works

[zinc parrot]

I was trying video gen using wan fast. I'm not sure what you want me to do with comfyUI I don't see the workflows and when I try to load them manually from the preset folder it complains about a version parameter missing.

Things I've tried so far:

• Image gen instead of video gen but also gives an error on the last step.
• Downloading templates from inside comfyUI but still gives an error on the last step.
• Lowering the total frames & resolution but same result

[torn solstice]

Dragging the landscape image into the ComfyUI workspace doesn't give you a workflow?

[zinc parrot]

Do you mean like this?

[torn solstice]

try this

This is workflow from AI Playground.

I want to see if it works from ComfyUI, and see what happens if you get rid of these two nodes

Have you edited or changed anything about the comfyui environment other than making this change to the settings.json
"comfyUiParameters": [
"--lowvram",
"--reserve-vram",
"6.0"
],

FYI this is to first test image gen, to see if you can at the very least ComfyUI to work

[zinc parrot]

the landscape2.json works

p.s. I still had the two nodes I will now try without those

Still working

So the draft image works fine, also in aipg

[torn solstice]

Good to know. Is it only wan video that doesn't work

[mint geyser]

Does AI Playground need an internet connection after everything is installed?

AI PG seems to take 1 to 2 seconds to generate an image with an internet connection and 70 seconds without an internet connection.

Is this because with an internet connection it is using remote generation and then reverting to local generation without an internet connection ?

[mint geyser]

Further investigation reveals AI PG is trying to get a few json config files for the NSFW checker

I'll post a log file

It attempts many times and eventually times out

[mint geyser]

It is trying to get to some json config files here ----> https://huggingface.co/AdamCodd/vit-base-nsfw-detector/tree/main

[torn solstice]

Yes the safetychecker installs after first run. And the Control Face is similar, needing to do something when first run but should not require anything afterwards.

In other words if you run Control Face once while online, it should have both tje safetycheck and Adam Codd files and should not be trying to access to files while offline. If so then we may have a bug in alpha

[mint geyser]

Hello Bob, The issue is always the same

Online it is fine but as soon as I unplug the internet it happens

It is trying to get to 3 files

[torn solstice]

Adam Codd is the inswapper Reactor Node part of the fave swapping features. Is the issue just limited to that?

[mint geyser]

Max retries exceeded with url: /AdamCodd/vit-base-nsfw-detector/resolve/main/preprocessor_config.json

Max retries exceeded with url: /AdamCodd/vit-base-nsfw-detector/resolve/main/processor_config.json

Max retries exceeded with url: /AdamCodd/vit-base-nsfw-detector/resolve/main/config.json

[torn solstice]

Is it doing this on Draft?

[mint geyser]

Online is fine offline it =tries to get these files for about 60 seconds

[torn solstice]

Which preset?

[mint geyser]

Yes apologies AI Playground Version v3.0.1 alpha

[torn solstice]

Which feature

[mint geyser]

Did you want me to make a real support ticket? this was just meant to be a quick question.

[torn solstice]

First I need to understand which feature. Which preset

Adam Codd is specific to two presets

[mint geyser]

This is Image gen with safety checker

[torn solstice]

Please try Draft and tell me if this is an issue

[mint geyser]

I'll get to it ASAP The room with the AI PCs is currently being used to deliver a Masters lab session

[torn solstice]

Please run Draft

[glad portal]

Wan video and also LTX-2 are the video models I've tried that do not work on my A770

[torn solstice]

I need you to confirm the Adam Codd files are tied to something other than Control Face. There are multiple safety check solutions. I need confirmation these exact files are related to presets not tied to them

Not Another Safety Check does download files at first run, not after
Adam Codd is a Safety check part of Inswapper. Part of ContolFace. So I need your help to isolate

[torn solstice]

Ok thanks. I'll look at these

[mint geyser]

Default settings

Draft

512x512

Same 3 JSON file errors

'(MaxRetryError('HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /AdamCodd/vit-base-nsfw-detector/resolve/main/config.json (Caused by NameResolutionError("HTTPSConnection(host='huggingface.co', port=443): Failed to resolve 'huggingface.co' ([Errno 11001] getaddrinfo failed)"))'), '(Request ID: 176d15c4-1892-4485-a985-2a7b619f84fb)')' thrown while requesting HEAD https://huggingface.co/AdamCodd/vit-base-nsfw-detector/resolve/main/config.json

With the internet unplugged 🙂

[mint geyser]

It's probably only specific to my use case.

I'm using 15 PCs to do local AI things with no or limited internet connection

And it has to be very SFW

The events will likely have school kids using the PCs.

The preview feature has been a problem too 😂

It shows the preview before NSFW checker

[torn solstice]

Thanks. I've alerted the dev. Seems we should be able to elimate or reduce the timeout when offline
There is a check above safetychecker to turn off preview. Does that work\

uncheck this

[mint geyser]

Hi Bob, The Preview check box does stop the preview from showing. I just nee to stop the kids from changing things

[torn solstice]

Lol... fun being a kid. There will be a demo mode coming that locks the settings when turned on, but it will be a limited experience. For now turn off preview, turn up Safety Checker, add bad stuff to the Negative Prompt (nsfw, nude, blood... ) and I guess keep an eye out on 15 PCs

[glad portal]

It really is most models on comfyui that it disconnects on Vae Decode. (Using A770 with the latest drivers and AI Playground v3.0.1 alpha

[surreal merlin]

Ive done a quick test and Wan video seems to be working now in ComfyUI with the new driver 32.0.101.8509 and the A750

[torn solstice]

FYI Reserve vram setting needed in ComfyUI for WAN Vace on Arc dGPU

[trim dock]

is it not possible to manually add models in current version of aipg (absolute newbie here)

i tried using the add model but it is asking me to point to repo for download but i already have the models downloaded

[torn solstice]

You can add models yes. When in chat use "Add Models" button.
The rest of the modes are managed via ComfyUI presets. Add models to the proper models/ComfyUI folder for use in manual or just launch comfyui and do it from there

[trim dock]

i did use the add models but its asking me to add from a repo on hugging face

[trim dock]

i'm using comfyui directly

[exotic mortar]

How or where do we add a model (checkpoint) for image gen? I found at least 3 "checkpoints" directories

[half dawn]

Why does this keep happening?

[mint geyser]

Do you have space for the model?

[half dawn]

i just restarted my pc, deleted the model file and it worked, but i think it was space since i deleted other stuff for more space

[torn solstice]

Adding models when directly in ComfyUI, you need to be sure to place them in the correct place. Most workflows from ComfyUI will tell you where these need to go. Those end directories are the correct place but be sure you place in /resources/models/ComfyUI/[directory]

The Add Models from Chat presets is specific to Chat models. You need the Huggingface id and need to tag properly. If you add models in manually outside of Huggingface drop them into the LLM directory then use the Advanced chat preset to use it

[exotic mortar]

Where do we add image gen models from Hugginface or Civit when using the AIP UI / app ? The PDF doesn't say.

[torn solstice]

For image gen: /resources/models/ComfyUI/
Place in the appropriate model type folder which for
Manual Mode I think is "checkpoint" but Id have to check

[torn solstice]

I do see an issue. Looks like the model list is not refreshing and stuck on the AIPG default models. Opening a ticket. Here's a work around

1. Download and place your checkpoint model in /resources/models/ComfyUI/checkpoint
2. Start AI Playground and generate a sample image using Draft or HD (choose the one using the base Stable Diffusion that matches your Checkpoint. Draft if SD1.5, HD if SDXL)
3. Click the ComfyUI Button in settings.
4. Drag in your image generated from step 1 to ComfyUI screen
5. In the checkpoint node, select your CivitAI model
6. Put your prompt in the postive CLIP node. Adjust parameters for resolution, steps, CFG, Scheduler as desired
7. Optional - Add a LoRA node, and connect between the Checkpoint and Clip nodes and put a LoRA model here. Download LoRAs to /resources/models/ComfyUI/loras

This should get you going until I fix the issue

[exotic mortar]

This explains what I was seeing. Being an AIP user for a long time now with additional image gen models, I thought I was doing something wrong in version 3 of AIP

[cinder geode]

anyone have the same problem? it's set on auto and my b580 isnt listed. i am on the latest driver and not using igpu right now

[cinder geode]

other dependencies installed just fine except comfyui. i keep getting the same error

[torn solstice]

That driver was bad. Update to one released yesterday.

Things to try if persists

1. Restart ComfyUI backend from app settings. Wait 2 minutes
2. disable iGPU if you have one. Restart AI Playground

[scarlet pebble]

Hello, I'm new here and just wanted to start by saying thank you for the AI-Playground!

It has really made getting started with local AI on Windows 11 a lot easier for me. So far, everything is working wonderfully. The graphics card (Arc A770 16GB) is recognized, and basically, everything is running exactly where it should (CPU & GPU).

I do have two questions, however:

1. In the console, I'm getting an error stating that the OpenVINO telemetry is outdated and that I should update it using uv sync. However, within the AI-Playground, this doesn't work the way I am used to.
2. When I try to integrate OpenVINO nodes into a workflow in ComfyUI, it fails to link my local C++ compiler.

Is it worth it for me to dive deeper into this, or is this just a tradeoff of the convenient plug-and-play nature of the AI-Playground?

[cinder geode]

doesnt work. i cant even get comfyui installed. i disabled my igpu long time ago

[mint geyser]

15 B580 PCs used as AI Playground outreach machines all performed really well

I had to edit some files to hide the generation options and lock down the safety checking

It went so well the school asked for extra sessions

There was some filming and pictures of the session. Once these have been vetted I may post some here.

[cinder geode]

i finally managed to get it to work. started a fresh new installation on D: and not inside subfolders

[cinder geode]

wan vace still not working on my machine. ltx runs fine but it struggles with >170 frames on my b580

[tulip field]

I got comfyui updated which was a real pain to do its now 0.15.1 im now having issues with models being in differently named folders

[torn solstice]

I'll check the node issue but on your side check the ComfyUI settings and be sure you have reserve-vram there. Wan VACE is very GPU and compute intensive.

[torn solstice]

AI Playground 3.0 is tested on 0.10.0. Updating ahead of our release updates for newer versions of ComfyUI is usually not an issue, but if ComfyUI makes significant changes it could be things break

[exotic mortar]

What triggers a new release of AIP? Is there a generally fixed cadence or some other criteria?

[neat stream]

Are those custom models still broken on ui (3beta)? (only "preinstalled" are shown on list)

[glad portal]

Has anyone got LTX-2.3 to work on ComfUI yet? It states that it requires the latest ComfyUI to be installed (0.10.0 is installed at this time)

[torn solstice]

Quarterly releases., then releases to fix or update features. Usually 4-6 weeks between releases

[neat stream]

@torn solstice so we are waiting 4-6week to fix on that "custom models" selection? , (also there is no cfg-slider and lora selection is missing.)

So basically workflow is use AIPG to generate "flower", open comfyui, and use that interface to change Settings /generate rest of the images.. So now AIPG is only used to start comfyui 🫤

[exotic mortar]

Is AI really helping if we're still on old release cycles? I appreciate the app immensely, simply some food for thought, even when taking into consideration the testing and review by people

[torn solstice]

Issue is something is holding up the release. Should be next week. Manual mode is fixed with ability to add sd1.5 and sdxl checkpoints with LoRAs

[torn solstice]

New release is out.
SD1.5 and SDXL CivitAI checkpoints and LoRAs now can be selected and run in Manual mode with ability to select preferred Scheduler

[thorn pollen]

I installed the new release and now the ComfyUI backend won't function. It says 'failed' when i load the AI Playground app, and when i try to repair it just fails repeatedly

[torn solstice]

Can you send a screenshot. Is it saying "Failed" from the Backend Installation Manager

and what's your hardware, and driver

and usual thing, if a dGPU, make sure to install with iGPU disabled

A test. Quit AI Playground. Go to AI Playground/resources. Rename the ComfyUI folder ie "ComfyUI-bak"
Start AI Playground it should say it is not installed. Check the settings and see if the version is set to v0.17.0 If not change it so it is. Then from there select to install that version

[thorn pollen]

This worked. Thank you. I backed up all of my custom workflows and files before the install so I'm back up and running again

[exotic mortar]

Do the sd1.5/sdxl checkpoints need to go into AppData/Local/Programs/AI Playground/resources/models/stable_diffusion/checkpoints ?

I put some there but AIP doesn't recognize/show them.

[torn solstice]

Hot fix release out - updates OVMS and Llama.cpp that had recent version fixes
https://github.com/intel/AI-Playground/releases/tag/v3.0.3-beta_hf

[exotic mortar]

Checking back to see if anyone knows how to find out what folder AIP looks into for image checkpoints

[hard forum]

Arc B390 configured to 36gb shared, Unable to perform any image gen: Error 40 #1243956384052285560

Anybody have any ideas? This is fresh win11 and the new panther lake XPS 16 with 64g memory. Shared memory currently at default (36gb shared) but issue persists even when configured to 58. Appears unrelated to model size as it occurs with the fast preset which is quite small.

[torn solstice]

What does Backemd manager say for the ComfyUI component. Is it Running?

[hard forum]

Thanks for the reply. Here’s exactly what I’m seeing (see attached screenshot):

*Fresh boot → Open AI Playground
*Wait for ComfyUI to start
*Switch to “Image Generation” (smallest preset)
*Enter no prompt → error pops up

System details
Windows 11
Core i7‑358h (B390 selected)
64 GB RAM (default 36 GB shared)
Draft default set to 0.25 MP

Thanks again and let me know if you need more info or something specific tested.

[tranquil elk]

Hello @hard forum , could you check what graphics driver you are using? could you update driver to latest 101.8629 from https://www.intel.com/content/www/us/en/download/785597/intel-arc-graphics-windows.html ?

[hard forum]

Thanks, I'm running 32.0.101.8622. (it has been updated recently) Will update to 32.0.101.8629 and report back. Any other ideas?

[hard forum]

Ok:
Rebooted -> Installed the update -> Rebooted Again -> Ran AI playground -> waited for comfy UI to start -> Switched to Image gen (Draft 0.25mp) -> Entered Dog Prompt

I can provide the console log if it would help. This system is pretty darn clean, nothing crazy going on aside from me disabling ondrive, recall, and a few telemetry options which really should not affect this.

I can see this is now a different error on the new driver. The same error is coming up constantly in the driver log as well. Is the B390 perhaps not supported/whitelisted in the version of ComfyUI that has shipped with AI Playground?

[torn solstice]

B390 is supported. You'll see its the gpu in my last 4 videos.
Things that will help.

1. Is ComfyUI status "Running" if Running then it installed correctly
2. Tell us about the system. Is the B390 an iGPU in a laptop, any other hardware
3. Do you have another instance of Python/Pytorch or Git installed

4. The specific driver you're using

[hard forum]

1. Yes it's running (screenshot in my first response) verified in console as well.
2. It's a Dell Xps 16, nothing fancy going on here I'm not using any external gpu or Thunderbolt devices.
3. Nope
4. Driver version is 32.0.101.8629 for my last post and and was 32.0.101.8622 for the first post.

Would a console log be helpful? The console shows 300+ warnings and 5 errors.

[tranquil elk]

could you show the full log of the last error message ? “runtime error: the program was built for xxxxx”

Also what AI Playground version are you using?

[hard forum]

@tranquil elk AI Playground version is 3.0.3.0. Here is the last error you requested (as a txt file).

[tranquil elk]

we just released 3.1.0-alpha today, could you give the new build a try? it has pytorch 2.11 integrated

[hard forum]

sure ill give it a go

[hard forum]

Should I uninstall AI playground first or just update?

[tranquil elk]

yes, if possible. ensure to backup your downloaded models first tho

[hard forum]

Uninstalled with the keep models folder option. And reinstalled the new version. It seems to be working properly now. I wonder what the issue was on the previous installation?

[tranquil elk]

just by looking at the error message it seems to be a driver issue with pytorch 2.10