Software Engineering Courses (SECourses)•6mo ago

I am using python 3.10 and cuda 12.8. I have the newest version of cudnn installed. onnxruntime-gpu

I am using python 3.10 and cuda 12.8. I have the newest version of cudnn installed. onnxruntime-gpu defaults to 1.17 version. What version should all of those be at so I can utilize my 5070ti and still use reactor? I just tried a massive rabbit hole with Gemini AI and it didn't fix it.

Aan80sPWNstar I am using python 3.10 and cuda 12.8. I have the newest version of cudnn install...

Furkan Gözükara SECourses•6/25/25, 9:49 PM

cudnn latest could be culprit

Furkan Gözükara SECourses•6/25/25, 9:49 PM

that repo is not getting updated

Furkan Gözükara SECourses•6/25/25, 9:49 PM

try cudnn 8.9 also

FFurkan Gözükara SECourses try cudnn 8.9 also

an80sPWNstarOP•6/25/25, 9:50 PM

the wheel or the .exe?

grandtheftdelta•6/25/25, 9:50 PM

@Dr. Furkan Gözükara Dm'd you!

Aan80sPWNstar the wheel or the .exe?

Furkan Gözükara SECourses•6/25/25, 9:52 PM

they are dll files like this

Furkan Gözükara SECourses•6/25/25, 9:52 PM

Ggrandtheftdelta @Dr. Furkan Gözükara Dm'd you!

Furkan Gözükara SECourses•6/25/25, 9:53 PM

replied

FFurkan Gözükara SECourses Click to see attachment

an80sPWNstarOP•6/25/25, 9:53 PM

gotcha. I'll let you know how it goes. You are freaking amazing for being so responsive on here!

Aan80sPWNstar gotcha. I'll let you know how it goes. You are freaking amazing for being so res...

Furkan Gözükara SECourses•6/25/25, 9:53 PM

you are welcome. i usually response in a day but i try to be faster usually

Segundus•6/26/25, 1:52 AM

What's the better choice for extending video right now - SwarmUI / Comfy or the Dr.'s Framepack app?

Curious if there are feature or advantages that ppl are enjoying in one or the other.

SSegundus What's the better choice for extending video right now - SwarmUI / Comfy or the ...

J2_Nedo•6/26/25, 10:27 AM

i endet up using Swarmui with the Wan2.1 Fusion X, its super fast, but i had no luck sofar to extend a video. Doc's Wan2.1 app have a really good extend video option, even with batch rendering! Would be cool to implement WAN2.1 Fusion X into his app. The original WAN2.1 is just super slow.

LTGpro•6/26/25, 11:16 AM

nice place

Inoshi•6/26/25, 11:24 AM

Dr is the new sageattention 2++ better than 2.7.4post1, i dont think they released it yet but if its on the second version doesnt that mean that its still close to the one we currently use?

JJ2_Nedo i endet up using Swarmui with the Wan2.1 Fusion X, its super fast, but i had no ...

Furkan Gözükara SECourses•6/26/25, 1:32 PM

i may look for how to do this in swarm

IInoshi Dr is the new sageattention 2++ better than 2.7.4post1, i dont think they releas...

Furkan Gözükara SECourses•6/26/25, 1:32 PM

2++ which one?

Furkan Gözükara SECourses•6/26/25, 1:33 PM

i think they are working on 3

AiInfluence•6/26/25, 1:33 PM

@Dr. Furkan Gözükara

AAiInfluence @Dr. Furkan Gözükara

Furkan Gözükara SECourses•6/26/25, 1:33 PM

AiInfluence•6/26/25, 1:33 PM

The most SOS

link I ever send you

AiInfluence•6/26/25, 1:33 PM

https://github.com/meigen-ai/multitalk?tab=readme-ov-file

GitHub

GitHub - MeiGen-AI/MultiTalk: Let Them Talk: Audio-Driven Multi-Per...

Let Them Talk: Audio-Driven Multi-Person Conversational Video Generation - MeiGen-AI/MultiTalk

AiInfluence•6/26/25, 1:34 PM

Someone also made a wrspper for comfyui we need tutorial with wan

AiInfluence•6/26/25, 1:34 PM

it's impossible to setup

AiInfluence•6/26/25, 1:34 PM

I'm getting "feature_extraction" errors when loading the Chinese wav2vec

AiInfluence•6/26/25, 1:34 PM

or English (facebook960)

FFurkan Gözükara SECourses 2++ which one?

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•6/26/25, 1:35 PM

3 is only for Blackwell and FP4 support, IIRC and 2++ is some enhancement, and they will not give you access to it probably https://huggingface.co/jt-zhang/SageAttention2_plus

jt-zhang/SageAttention2_plus · Hugging Face

AAiInfluence https://github.com/meigen-ai/multitalk?tab=readme-ov-file

Furkan Gözükara SECourses•6/26/25, 1:46 PM

ye they dont have any optimizations i dont see in repo

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 3 is only for Blackwell and FP4 support, IIRC and 2++ is some enhancement, and t...

Furkan Gözükara SECourses•6/26/25, 1:47 PM

lol this is weird

Furkan Gözükara SECourses•6/26/25, 1:47 PM

i bet it wouldnt compile on windows

Furkan Gözükara SECourses•6/26/25, 1:49 PM

i dont want to deviate from video upscaler almost done and i am extremely busy this week until monday then hopefully will have way more time

Furkan Gözükara SECourses•6/26/25, 1:49 PM

university ending

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•6/26/25, 3:35 PM

@Dr. Furkan Gözükara https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev

black-forest-labs/FLUX.1-Kontext-dev · Hugging Face

Inoshi•6/26/25, 3:36 PM

We need fp8 version

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•6/26/25, 3:38 PM

https://comfyanonymous.github.io/ComfyUI_examples/flux/#flux-kontext-image-editing-model

Inoshi•6/26/25, 3:48 PM

download is too slow though

Flame火•6/26/25, 3:58 PM

hello im trying to train a character lora. i've generated like 100 images of the character but now im confused since i've got this many images how many repeats, epochs, steps to use?

AiInfluence•6/26/25, 5:27 PM

https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev/tree/main

black-forest-labs/FLUX.1-Kontext-dev at main