Software Engineering Courses (SECourses)•2y ago

hey dumb question, 4090 comes with a power connector that feeds in 4 power cables from PSU but this

hey dumb question, 4090 comes with a power connector that feeds in 4 power cables from PSU but this PC i'm trying to ram it in only has 3 to feed into it

will it work with 3 or is this a retarded question, asked by an idiot

FFurkan Gözükara SECourses RMBG is best as far as i know

Neo•7/26/24, 3:47 AM

I use v1.4 of RMBG, it's very good but sometimes leaves a lot of pixelation in the cropping around the foreground objects and sometimes cuts off hair.

Both of these projects are supposed to improve on that. I've had good success with online demos in comparison.

FFabricatedgirls hey dumb question, 4090 comes with a power connector that feeds in 4 power cable...

J2_Nedo•7/26/24, 5:16 AM

what PSU is that? I doubt it will work

FFabricatedgirls hey dumb question, 4090 comes with a power connector that feeds in 4 power cable...

daroth•7/26/24, 5:28 AM

you don't need 4. 3 is enough. especially if you undervolt the gpu to like 80%

Mattheus Chediak•7/26/24, 11:01 AM

hey i didnt find the linux IDM VTON with gradio and runpod on the patreon, can someone give me some supoprt

__pirate_king__•7/26/24, 11:48 AM

Hey guys, from this code -> https://huggingface.co/spaces/multimodalart/Ip-Adapter-FaceID/blob/main/app.py
I need the below part of the code to use juggernaut-x-v10 and ip-adapter-v2-sdxl models, how can I do this?

original code: (from the above link)

base_model_path = "SG161222/Realistic_Vision_V4.0_noVAE"
vae_model_path = "stabilityai/sd-vae-ft-mse"
image_encoder_path = "laion/CLIP-ViT-H-14-laion2B-s32B-b79K"
ip_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid_sd15.bin", repo_type="model")
ip_plus_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid-plusv2_sd15.bin", repo_type="model")

base_model_path = "SG161222/Realistic_Vision_V4.0_noVAE"
vae_model_path = "stabilityai/sd-vae-ft-mse"
image_encoder_path = "laion/CLIP-ViT-H-14-laion2B-s32B-b79K"
ip_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid_sd15.bin", repo_type="model")
ip_plus_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid-plusv2_sd15.bin", repo_type="model")

My changes:

base_model_path = "RunDiffusion/Juggernaut-X-v10"
vae_model_path = "stabilityai/sd-vae-ft-mse"
image_encoder_path = "laion/CLIP-ViT-H-14-laion2B-s32B-b79K"
ip_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid_sdxl.bin", repo_type="model")
ip_plus_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid-plusv2_sdxl.bin", repo_type="model")

base_model_path = "RunDiffusion/Juggernaut-X-v10"
vae_model_path = "stabilityai/sd-vae-ft-mse"
image_encoder_path = "laion/CLIP-ViT-H-14-laion2B-s32B-b79K"
ip_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid_sdxl.bin", repo_type="model")
ip_plus_ckpt = hf_hub_download(repo_id="h94/IP-Adapter-FaceID", filename="ip-adapter-faceid-plusv2_sdxl.bin", repo_type="model")

Error:

line 80, in generate_image
    image = ip_model_plus.generate(
  File "/venv/lib/python3.10/site-packages/ip_adapter/ip_adapter_faceid.py", line 384, in generate
    prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
ValueError: too many values to unpack (expected 2)

line 80, in generate_image
    image = ip_model_plus.generate(
  File "/venv/lib/python3.10/site-packages/ip_adapter/ip_adapter_faceid.py", line 384, in generate
    prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
ValueError: too many values to unpack (expected 2)

Any help is appreciated.

Mattheus Chediak•7/26/24, 1:07 PM

like, can i get the github idm-vton repository create a dockerfile and create and endpoint in runpod that runs this dockerfile and just make inference using postman with the respective endpoints for the runpod ? would that work ?

FFurkan Gözükara SECourses https://github.com/THUDM/CogVLM2

Sipriyani•7/26/24, 3:11 PM

I think this works amazing, it makes so detailed image captioning. I tried with a few pics, and gpt4 vs Llma3.. I think Llma3 is trained way better than gpt. in every picture of me, llama gave better prompting.

gpt4o

The image shows an individual standing in a parking lot. The person is wearing a black t-shirt and grey pants, and has a watch on the left wrist. The face is obscured by a brown rectangle, likely for privacy reasons. In the background, there are parked cars and trees, suggesting this might be near a park or a commercial area.

vs
Lma3

The image shows a man standing in a parking lot. He is wearing a black t-shirt and olive green cargo pants. He has a short haircut and a beard. He is looking directly at the camera with a neutral expression. In the background, there are several cars parked in the lot, and there is a structure that appears to be a part of a building or a canopy. The sky is overcast, suggesting it might be a cloudy day.

UnrealState•7/26/24, 4:10 PM

Question about Rope Pearl. How do I record videos when using the webcam?

dsienra•7/26/24, 4:42 PM

I'm testing a SD3 full finetune with included kohya_ss SD3 preset, and as you can see is training at 512x512 resolution, even then it consumes my entire vram, is fast but I don't know how will be the quality training at 512x512, the preset comes configured at 512 I think because 1024x1024 will require a lot of vram

dsienra•7/26/24, 4:43 PM

I will share later if the test ends with good results

UUnrealState Question about Rope Pearl. How do I record videos when using the webcam?

UnrealState•7/26/24, 5:46 PM

nevermind. i installed obs studio virtual camera and all is well.

GGazer @Dr. Furkan Gözükara do you have any resources regarding setting up the differen...

Furkan Gözükara SECourses•7/26/24, 11:09 PM

you can start each thing on each gpu

Furkan Gözükara SECourses•7/26/24, 11:09 PM

it is easy

Furkan Gözükara SECourses•7/26/24, 11:10 PM

like SET CUDA_VISIBLE_DEVICES=0

Furkan Gözükara SECourses•7/26/24, 11:10 PM

SET CUDA_VISIBLE_DEVICES=1

Furkan Gözükara SECourses•7/26/24, 11:10 PM

each app will run on each gpu

FFabricatedgirls hey dumb question, 4090 comes with a power connector that feeds in 4 power cable...

Furkan Gözükara SECourses•7/26/24, 11:10 PM

it depends how much power psu can give but i think should work

NNeo I use v1.4 of RMBG, it's very good but sometimes leaves a lot of pixelation in t...

Furkan Gözükara SECourses•7/26/24, 11:11 PM

thanks i should make installer and tutorial for them

MMattheus Chediak hey i didnt find the linux IDM VTON with gradio and runpod on the patreon, can s...

Furkan Gözükara SECourses•7/26/24, 11:13 PM

hi it is inside zip file

Furkan Gözükara SECourses•7/26/24, 11:13 PM

extract it and follow instructions please

Mattheus Chediak•7/26/24, 11:13 PM

I got it working doc

Mattheus Chediak•7/26/24, 11:13 PM

Thanks for the help

Mattheus Chediak•7/26/24, 11:14 PM

I supported you guys on patreon

___pirate_king__Hey guys, from this code -> https://huggingface.co/spaces/multimodalart/Ip-Adapt...

Furkan Gözükara SECourses•7/26/24, 11:14 PM

MMattheus Chediak Thanks for the help

Furkan Gözükara SECourses•7/26/24, 11:14 PM

thanks a lot

Mattheus Chediak•7/26/24, 11:14 PM

Thank you

Furkan Gözükara SECourses•7/26/24, 11:14 PM

patreon rank given

___pirate_king__Hey guys, from this code -> https://huggingface.co/spaces/multimodalart/Ip-Adapt...

Furkan Gözükara SECourses•7/26/24, 11:14 PM

did you try this

Furkan Gözükara SECourses•7/26/24, 11:14 PM

our gradio has export diffusers

Furkan Gözükara SECourses•7/26/24, 11:14 PM

try juggernaught export

SSipriyani I think this works amazing, it makes so detailed image captioning. I tried wit...

Furkan Gözükara SECourses•7/26/24, 11:18 PM

CogVLM is supposed to be the state of the art atm

Furkan Gözükara SECourses•7/26/24, 11:18 PM

i am trying to make work on windows

Furkan Gözükara SECourses•7/26/24, 11:18 PM

working but ultra slow

UUnrealState Question about Rope Pearl. How do I record videos when using the webcam?

Furkan Gözükara SECourses•7/26/24, 11:18 PM

good question

Furkan Gözükara SECourses•7/26/24, 11:18 PM

did you try record button?

Furkan Gözükara SECourses•7/26/24, 11:18 PM

as usual

Furkan Gözükara SECourses•7/26/24, 11:18 PM

when you stop it it should save

UUnrealState nevermind. i installed obs studio virtual camera and all is well.

Furkan Gözükara SECourses•7/26/24, 11:18 PM

nice

Ddsienra I'm testing a SD3 full finetune with included kohya_ss SD3 preset, and as you ca...

Furkan Gözükara SECourses•7/26/24, 11:19 PM

when it comes to VRAM

Furkan Gözükara SECourses•7/26/24, 11:19 PM

OneTrainer best

FFurkan Gözükara SECourses when it comes to VRAM

dsienra•7/27/24, 12:12 AM

now I'm training at 1024x1024 and the vram usage is identical what is very wired, the results at 512x512 were not so good, lets see at 1024 now

dsienra•7/27/24, 12:15 AM

I tried onetrainer with sd3 but the results were terrible, the model gets fried and no people likeness at all, the resulting images were terrible full of artifacts

Ddsienra I tried onetrainer with sd3 but the results were terrible, the model gets fried...

Furkan Gözükara SECourses•7/27/24, 12:29 AM

it needs research

Furkan Gözükara SECourses•7/27/24, 12:29 AM

i plan to do but busy got 2 client work right now that i have to finish :d

dsienra•7/27/24, 12:29 AM

the loss looks much better at 1024x1024, in onetrainer the loss graph is all over the place, I think there is a bug in onetrainer and sd3 I'm using the default presets on both onetrainer and kohya_ss, at this moment kohya_ss is much better on sd3, my 512x512 test looked a little undertrained but quite good

Furkan Gözükara SECourses•7/27/24, 12:30 AM

loss rates were never meanginful for me

Furkan Gözükara SECourses•7/27/24, 12:30 AM

when training SD 1.5 or SDXL

Théo•7/27/24, 12:30 AM

guys i have a big problem that started today and i don't know why, my comfyui is very laggy and the UI is super small, any idea ?

Théo•7/27/24, 12:30 AM

i feel like the resolution of the software got altered or something
edit : ok my chrome decided to put comfy ui in 15% scale resolution, super weird, but it fixed it to put it back to 100%