Post "All Patreon Sharings and Scripts" by "Furkan Gözükara" from patreon

Furkan Gözükara

this already exists in WAN first last frame presets workflows. we dont have atm but i plan to make

Hockey

Is it possible we can get this? https://github.com/raindrop313/ComfyUI-WanVideoStartEndFrames

Furkan Gözükara

hi in which app?

Julius

Hi, how can I save my personnal presets on SwarmUI

Furkan Gözükara

that sounds interesting. i will check if i can make. perhaps with newest AI tech we may make that easily. but best side of gradio apps that they are based on official pipelines so they are best quality and most accurate implementation

James Charleston II

you build a lot of gradio based apps, which is very cool. Have you though about doing a small series of how to take a hashed out complex comfyUI workflow, and build a self contained single course gradio based application like you do. I love that you create and share and I am very happy to test and use, but I would lover to learn how to do something similar for my workflows. I love comfyUI, but would love to be able to have my workflows as apps, so that making new ones does not break them.. Also, would be cool to have option to pull in models we may already have from a central model storage.

Furkan Gözükara

nice will check

Alexandr Mishuk

https://github.com/microsoft/VibeVoice what about it ?

Furkan Gözükara

the models doesnt support 10-20 yet but it would be fast since it wont need block swap

Robert Arsene

How fast do you think we can generate videos with Wan 2.2 and an RTX Pro 6000 Blackwell? 😊 Or could we get a one shot 10-20 second video generation?

Furkan Gözükara

https://youtu.be/8cMIwS9qo4M this video very up to date follow this 21:32 Part 2: Massed Compute Cloud GPU Tutorial 22:03 Massed Compute - Deploying a GPU Instance (H100) 23:40 Massed Compute - Setting Up the ThinLinc Client & Shared Folder 25:07 Massed Compute - Connecting to the Remote Machine via ThinLinc 26:06 Massed Compute - Transferring Files to the Instance 27:04 Massed Compute - Step 1: Installing ComfyUI 27:39 Massed Compute - Step 2: Installing MultiTalk Nodes 28:11 Massed Compute - Step 3: Downloading Models with Ultra-Fast Speed 30:22 Massed Compute - Step 4: Launching ComfyUI & First Generation 32:45 Massed Compute - Accessing the Remote ComfyUI from Your Local Browser 35:07 Massed Compute - Downloading Generated Videos to Your Local Computer 36:08 Massed Compute - Advanced: Integrating with the Pre-installed SwarmUI 38:06 Massed Compute - Crucial: How to Stop Billing by Deleting the Instance 38:33 Part 3: RunPod Cloud GPU Tutorial

Walker4k

Regards running on MC, what is the best process?

Furkan Gözükara

it works with our existing config and kohya perfect. i also plan to make a comparison video hopefully

Tan

Checking out the new Flux.1 KREA. Can't wait to see how it finetunes.

Furkan Gözükara

yes i plan to do ty for reminding

cool1

There's a new AI image editor called HiDream-E1-1 out that seems similar to Flux Kontext and BAGEL. Though it has a better license than Flux Kontext (it looks like it can be used commercially, like BAGEL can, unlike Flux Kontext). And I assume it might be better quality than BAGEL or that it can run or 24 GB GPUs easier than BAGEL can. If it would be runnable on 24 GB GPUs (though if it goes over into RAM that would be okay too though ideally it wouldn't be too slow)could you please create an easy install for it (eg. with a gui)?

Furkan Gözükara

we have hidream in swarmui already to use this video shows but i plan to investigate further : https://youtu.be/fTzlQ0tjxj0

Furkan Gözükara

yes please remind me in few days

Bartosz Polecki

Hi Mate, will you update your Joy Caption setup to beta one model version ? Im curious how good it is.

Civitaier

Hey bro, have you read about the new AI image generator HiDreamAI? They say it’s better than Flux, with a lot more parameters and better instruction interpretation. Hopefully, you’ll be able to include it in your materials and tools in the near future as well, so we can learn how to work properly with that model and even train our own models. greetings, bro!

Furkan Gözükara

i am waiting 2.5 they said they will publish

Furkan Gözükara

i saw this but this is only useful for blurry video. you still need it?

Philip Heggie

I'd love you to do this video de-blurer please https://github.com/DachunKai/Ev-DeblurVSR

Sérgio Merêces

will be really cool have this one local too: Hunyuan3D 2.0 2MV

Furkan Gözükara

sadly i dont know any good Turkish atm. but if you know any good repo i might make installer

umut yilmaz

Hi, Furkan is there a good text to speech model with Turkish support in your 1 click install archive for 5000 series?

Furkan Gözükara

Thanks. Hopefully I will add to very best one soon

Chris

Hi Furkan, great to see your github is back :) Cheers! Do you know a good free video upscaler for use with framepack?

Furkan Gözükara

100%. and i will look controlnet upscaler sounding good

Alexandr Mishuk

it's clearly worthy of attention. https://github.com/SandAI-org/Magi-1 and this https://huggingface.co/jasperai/Flux.1-dev-Controlnet-Upscaler

Furkan Gözükara

thanks adding to list

Andrea Taini

Hi! I found this project to generate music and add sounds to video. Tried the demo seems great! Can you consider it for adding the project here? :) https://zeyuet.github.io/AudioX/

Furkan Gözükara

yes hopefully will make a tutorial for it soon

Alexandr Mishuk

what about https://github.com/HiDream-ai/HiDream-I1 ?

Furkan Gözükara

perhaps inpainting text may help did you try? here a tutorial : https://youtu.be/hewDdVJEqOQ

Andres Eduardo Damonte

Hi, good afternoon. Which AI model would you recommend to generate images that already include text within them? My goal is to create images where the text is integrated from the beginning, so I don’t have to add it manually afterward. For example, if the image shows a car, I’d like it to already include something like “Edu’s Car”. I have InvokeAI installed, but I haven’t been able to achieve the desired result. Could you guide me on how to do it or suggest a more suitable tool?

Furkan Gözükara

you can try InvokeAI : https://www.patreon.com/posts/112912425

Hockey

I'm having an awful time with inpainting with Flux in SD Forge and Swarm UI. Do we have a better inpaint AI software?

Furkan Gözükara

well there are control nets for this but i am searching for SOTA that will do as we want like OpenAI image model.

Hockey

Is there an AI software we got here that has the ability to alter the style of an image but retaining the structure? For instance, in Photoshop you can apply filters which doesn't alter the structure.

Furkan Gözükara

i dont understand your question can you elaborate?

Taiga

Can you write up a models directory that is remote from your installs so that we don't have to cart the folders from one place to another with each update? We get things working and, these environments are rather fragile, they don't want to try to just 'update'. A shared pool of models would be best.

Furkan Gözükara

hello fixed use the latest zip file please it has working notebook ty

Nam Lê Hoàng

Thank you. Here is my notebook for your ease: https://www.kaggle.com/code/namlh2004/notebook203c8143c4

Furkan Gözükara

checking right now

Nam Lê Hoàng

Hi, I am from Discord server. Here is the my problem: Can you check the Kaggle version of IDM-VTON project again ? https://github.com/FurkanGozukara/IDM-VTON/blob/main/app_VTON.py I followed the instructions in Youtube and Kaggle file in Patreon, it could run 3 weeks ago. But now when I ran, it showed me the error: ValueError: When localhost is not accessible, a shareable link must be created. Please set share=True or check your proxy settings to allow access to localhost. I tried to set share=True, but it didn't work.

Furkan Gözükara

hi latest version is here : https://www.patreon.com/posts/wan-2-1-ultra-as-123105403 i am about to update to next version soon

Milan Šaffek

thanks for reply.. i have installed your 1 click installation package, what is v33 and how can i get it please?

Furkan Gözükara

hello it was broken. just few minutes ago fixed with v33. are you on v33 right now? if you are on v33 what you can do is monitor your VRAM usage is slowly increase number of persistent parameters until it becomes too slow. like increase it with 500m at a time.

Milan Šaffek

Hey, i have also issues with training on rtx 4090. i tried 24g or 16g presets, and all are very slow about (50s /it).. Where i could find Number of Persistent Parameters In Dit ?

Furkan Gözükara

yes you are right. lower steps = lower VRAM and faster generation

Roman Fierce

Furkan, I found the reason: if I use different number of steps, I get different VRAM usage. 81 or more use more than 23.5 GB at once. 40 steps use about 10 GB. It is right? I think the problem might be the Number of Persistent Parameters In Dit (VRAM). For 24 GB it is set to 0.

Furkan Gözükara

Wow weird. so now you have idea therefore you can slightly increase according to your system.

Roman Fierce

Hi Furkan! I hope you have a great day. I am having constant issues with current presets. They go over 23.5GB of my VRAM (RTX 4090) and then the generation stops. Do others report the same issues? Is there a way to make the program use less VRAM? Thanks. Update: I switched to 16GB preset and VRAM usage currently is at 19.1GB. My system uses only 1GB and I don't use any AI apps in parallel. That's weird.

Furkan Gözükara

Kaggle currently banned Gradio. But usually I make. For example I was gonna make for Wan 2.1, even prepared, but Gradio not working :(

Philipp Capetian

If the requirements are rather low, will there be a Kaggle Notebook, as well?

Furkan Gözükara

yes i was working but now working on Wan 2.1 even better

cool1

Could you do a tutorial on the Skyreels v1 image2video and text to video please, for Comfyui (local)? With info about setting things up. I tried setting that up but keep getting errors when I try to queue a prompt (size mismatch. So I've not set things up right). Also I only saw a workflow for image2video not the text one too.

Furkan Gözükara

VRAM is more important because when it doesn't fit into VRAM, it will use RAM and it will be many times slower. So go with highest VRAM :) to compare their speed you can run both GPUs on RunPod and compare or look their CUDA core counts

Hockey

Does anyone know if VRAM is more important for speed or the graphic card itself? Only cheap 16 Gb card I can find in a laptop is the Quadro RTX 5000. It's slower than the RTX 3080, the question is how much slower. =S I use a RTX 2060 with 6 Gb VRAM today.

Furkan Gözükara

Yes. Sorry for the delay. I have a lot of real life back-logs but it is a high priority

VelorianX

Are you working on Hunyuan Video Lora training?

Furkan Gözükara

nice. hopefully very soon with more features than official version i will publish

Sylvain

It's less gpu intensive then trellis also

Furkan Gözükara

yes i am following every changes. my tutorial will be based on that hopefully

Jason Dawn

Well that was easy. Due to the UV version delivered with the bmaltais/musubi-tuner-gui, the complete installation of musubi-tuner and the GUI and all Python requirements was done with the three commands: git clone --recursive https://github.com/bmaltais/musubi-tuner-gui.git cd musubi-tuner-gui .\gui.bat

Jason Dawn

Hi, do you have plan to make an installer for https://github.com/kohya-ss/musubi-tuner in combination with this gradio UI: https://civitai.com/articles/10335/hunyuan-video-lora-trainning-with-gui-in-windows Edit: Actually the creator of Kohya_SS GUI started to work on a GUI for Musubi-Tuner: https://github.com/bmaltais/musubi-tuner-gui

Furkan Gözükara

thanks will work on it soon hopefully

George Gostyshev

Tested it today - it's definitely gives more solid results

Furkan Gözükara

great one thanks should make installer. i wonder if better than trellis

George Gostyshev

https://github.com/Tencent/Hunyuan3D-2 Could be great to have one)

Furkan Gözükara

i agree hopefully try to cover

Mike

seconded. i run local LLM's often. I would have benefitted massively from explanations of different settings especially.

Furkan Gözükara

yes i wanted to cover them as well but i never had time until now :D except very few cases

cool1

In the 'about' section it says "I will share the very best tutorials and scripts and installers and guides for:" and one of the things it says is "Large Language Models (LLMs)". Will there be more tutorials on the free local large language models (maybe including those that can do agent type stuff) and local apps for them please (like GPT4ALL, oobabooga text gen etc)? Like doing clever things that might be profitable with them?

Furkan Gözükara

yes coming soon hopefully :)

Louis L'herrou

Hi Furkan, please make a tutorial on hunyuan Lora/finetunig :) :)

Furkan Gözükara

nice i will checkout

Philip Heggie

I'd love this 2d to 3d video converter https://github.com/TencentARC/StereoCrafter

Furkan Gözükara

i plan to make a tutorial for this hopefully soon

Chris

Hi Furkan, i thought you also had a tutorial on ltx image to video in swarmui, maybe i am confusing it with some other tutorial. did you do that on youtube about ltx and swarm?

Furkan Gözükara

nice gonna check out

Kappa_Neuro

Hi. Pay attention to this cool image upscaler. The quality is fire. It works on the basis of SD https://github.com/ArcticHare105/S3Diff

Furkan Gözükara

yep it is almost ready. i made some nice upgrades :)

Anshul Gupta

You need to update the Nvidia sana image generator because a new model of sana nvidia avialabnle that is "SANA_1600M_2Kpx_BF16.pth.

Furkan Gözükara

yes we have please read here : https://www.patreon.com/posts/108419878

Philip Heggie

Thanks, Do you have a tutorial to change the hugging face cache folder to a different drive

Furkan Gözükara

surely you can install models on other drives. especially you can change your hugging face cache folder to a different drive . just install python cuda cudnn c++ tools ffmpeg and similar thing on C

Philip Heggie

I'm running out of space on windows c: drive because of AI can I install on a new 1TB F: drive to save space on c: drive please?

Chris

Thank you very much, looking forward!

Furkan Gözükara

i just asked that if possible in SwarmUI

Chris

Hi Furkan, do you know how to get this done in swarmui? Looks like details can be improved greatly with modification of sigma shift: https://github.com/Jonseed/ComfyUI-Detail-Daemon

Furkan Gözükara

your best shot is using ctrl-f in here : https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/Patreon-Posts-Index.md that is what i am using constantly to see. patreon system so bad for a better presentation

gravy

I don't find how things are organized to be very useful. I don't want to access things using chronological order, but by topic / category. Eg flux, sdxl, sd1.5, etc. Stuff is munged together it's an effort to tease out the information I'm looking for within this patreon

Furkan Gözükara

hopefully i am going to investigate this and make auto downloader

Chris

Hi Furkan, do you know if controlnets can be installed and used in swarmui with flux models? there are no models in the dropdown menu of swarmiu controlnet section. do i have to download them or are there none for flux yet?

Furkan Gözükara

yes i didnt describe anything. the bigger image dataset 256 images works best. check out into of this tutorial : https://youtu.be/FvpWy1x5etM

Furkan Gözükara

i don't know txt mlp or other blocks. as far as i know there are 19 double layer and 38 single layer blocks in MMDiT model of the FLUX

Furkan Gözükara

yes i am following it is great one

Kappa_Neuro

Furkan, look what a cool and promising thing! It works with images now, but it will work with videos soon! Maybe you'll even make an 1-click-installer and interface :D https://github.com/wangjiangshan0725/RF-Solver-Edit

Kappa_Neuro

Why didn't you exclude txt, mlp and other blocks when learning on 1 block (layer)?

Kappa_Neuro

Hi, Furkan! Which dataset are you currently using to test training? How did you describe this dataset? I just noticed that there is a problem with the hair color change in your tests, but I don't remember you describing your hair in your dataset.

Furkan Gözükara

i still find pre alpha best thus waiting next version to update it takes huge time the guy changes so much

cool1

But isn't that note about "Note: Alpha Two is not a general instruction follower and will not follow prompts outside its training data well. Use this feature with caution." isn't it the same thing as with joycaption alpha 1 as well. (or maybe he just meant the additional instructions he trained it for - the 100K ones weren't enough and varied enough to make it _very_ flexible for that)? In the Joycaption alpha 2 release notes he said he added "100k more captions based on a quick set of instructions I threw together. The instructions are nowhere near comprehensive, so Alpha Two isn't a general instruction follower. But the new Caption Types you'll find in the demo are part of this expansion of capabilities. " But alpha 1 isn't a general instruction follower either outside the training set (which is about 100K less than alpha 2 I think), eg. I don't think you can put in the custom prompt "...write the caption in British English/UK spelling" and have it properly do that (maybe because none of the training captions had it with that spelling). Though you can put in the custom prompt things like who the person is and where the shot is and it sometimes takes that into account when describing the image (in alpha 1. I assume it might do that in alpha 2 too). For alpha 2 in the release notes he said he manually checked the different versions against 15 validation images with human written captions I think (though that seems quite a low number for proper validation, ideally he could have checked a lot more), so for accuracy on those checks, his scores were 22.6 for PreAlpha, 21.5 for Alpha 1 (so alpha 1 was actually slightly less accurate than the prealpha based on those 15 validation images), 22.8 score for Alpha 2. So alpha 2 is, according to him, the most accurate so far (though only 0.2 better than the previous best one - PreAlpha), but also has more custom options (even though it could be better and follow instructions better, maybe if it was trained on more content with more variety of captions).

Furkan Gözükara

ye i know alpha 2 but it is i think even worse than alpha 1. pre alpha still working best. i skipped that version since it is "Note: Alpha Two is not a general instruction follower and will not follow prompts outside its training data well. Use this feature with caution."

Kappa_Neuro

Hi, Furkan! Let's update the model in the image description interface (JoyCaption https://www.patreon.com/posts/110613301) The second alpha version has been released: https://civitai.com/articles/7697/joycaption-alpha-two-release Demo: https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-two Model: https://huggingface.co/fancyfeast/llama-joycaption-alpha-two-hf-llava

Furkan Gözükara

it is hard to train a new language not easy task

Arkadiusz D

I managed install it locally, its easy, also i wanted to train my voice, but Polish language, and as I think, it needs a lot of hours to learn new language as for test i used only 2 minutes of my voice lol, and result ... voice was similar but completely not Polish accent. Or I do not know how to use a vocab list

Furkan Gözükara

i am closely following for a full tutorial with fine tuning

Arkadiusz D

Did you try T5 TTS ? https://github.com/SWivid/F5-TTS

Furkan Gözükara

it is so hard to test and especially installation of this part will be biggest challenge cd submodules/gaustudio-diff-gaussian-rasterization python setup.py install cd ../../ python setup.py develop also they have no demo app :D

Kappa_Neuro

Hi, can you watch this neural network? Is it really possible to install it on Windows? No matter how many times I tried, it doesn't work out. https://github.com/GAP-LAB-CUHK-SZ/gaustudio

Furkan Gözükara

i think swarmui may support this let me ask

Arkadiusz D

OmniGen https://github.com/VectorSpaceLab/OmniGen

Arkadiusz D

I also wish to use but I think its only for Linux ?, not Windows ?

Furkan Gözükara

hopefully yes. currently editing the FLUX fine tuning tutorial

Philip Heggie

Hi are you going to do Hallo2 https://fudan-generative-vision.github.io/hallo2/#/

Furkan Gözükara

yes it is on my todo list hopefully

Arkadiusz D

Did you hear about https://huggingface.co/spaces/ohayonguy/PMRF ??? I wish to know how to install it locally

Furkan Gözükara

yep possible but usually ai picks problems that comes with ai images :D

Patrick Major

this is really pretty cool. some automation combined with this tool and we could easily make proper good sized datasets for flux that expand the range of emotions on a models face.

Furkan Gözükara

I am gonna make Fine Tuning which is better than DoRA :) but If Kohya adds DoRA I plan to research it as well

Furkan Gözükara

awesome ty so much

Philip Heggie

You might want to make an app for expression-editor cog image by fofr Easily Change Facial Expressions On Images https://github.com/fofr/cog-expression-editor https://www.youtube.com/watch?v=DC3sYgj_byc

V Santhosh

Hi, Could you please kindly make a video/tutorial on DoRA character training? I saw astonishing results online using DoRA with flux.

Nobodys Hero

HopeFOLEY lol

Furkan Gözükara

yep coming soon hopefully

Tobe2d

+1

Furkan Gözükara

yes it is on my todo list hopefully

Philip Heggie

Hi can you make an app for Flux PuLID please https://github.com/ToTheBeginning/PuLID https://www.youtube.com/watch?v=NA_NpkIYsEo Flux Pulid - Create Consistent Characters From A Single Image - No Lora Needed!

Furkan Gözükara

nice will do hopefully

Philip Heggie

Would you mind making an installer for FoleyCrafter which automatically add synced sound effects to your mp4 videos https://www.youtube.com/watch?v=o8uhRTuqvvY and https://github.com/open-mmlab/foleycrafter

Furkan Gözükara

i published it if you missed : https://www.patreon.com/posts/110060645

Philip Heggie

Thank You. It's also lightning fast when processing very small images too.

Furkan Gözükara

nice. surely i will add hopefully soon

Philip Heggie

I found it worked very well on small images. It alters images less than supir so think it might work good for videos with less difference between frame to frame. Thankyou for adding to your list.

Furkan Gözükara

I tested it and it wanst good but surely i can make. adding to list

Philip Heggie

Would you mind doing windows install with batch image support for GigaGAN upscaler please for AURA-SR GigaGAN Open-Source Super-Resolution Upscaler it's on pinokio without batch support. https://discord.com/channels/1121039057993089076/1144758664746512435/threads/1255601335307468910

Furkan Gözükara

thanks just set your rank. it allows you to have special rank color and access patreon chat

The Strategist

Uhm... what's special rank on discord. My discord name is "igothics"

Furkan Gözükara

yes. this means while installation there was an error so it was incomplete. please reinstall and send me all install logs : monstermmorpg@gmail.com

PirateDiffusion.com

Does anyone know how to fix this error? \gradio_demo.py", line 13, in import einops ModuleNotFoundError: No module named 'einops'

Furkan Gözükara

well maybe they updated

Bas Uterwijk

--share=TRUE seems to work, altghough it gives me a weird gradio hex address that I am not used to...

Bas Uterwijk

"gradio_demo.py: error: argument --share: expected one argument"

Furkan Gözükara

in the start_SUPIR_Windows.bat like this python gradio_demo.py %lowvram% %tiledVAE% %theme% %cpuMove% --open_browser --share

Bas Uterwijk

Furkan Gözükara

we have --share args

Bas Uterwijk

I am trying to ass the --listen flag anywhere in a .bat file but no luck so far. Does anyone know how to access my Supir from another PC on my home network?

Furkan Gözükara

nice

Karl B

Looks like a bug was found in gradio, now you can just add a few lines of code to change oobabooga theme colors on everything easily! I wonder if this can be done in forge/auto1111, would make great tutorial also! https://github.com/oobabooga/text-generation-webui/issues/5731

Furkan Gözükara

I saw that. I should make. I plan to cover voice cloning

Karl B

Any chance you can make an easy one click installer for this voice cloning with oobabooga with mic. Uses alltalk_tts. His workflow in comments. https://www.reddit.com/r/StableDiffusion/comments/1bbtu7l/oobabooga_with_voice_cloning_now_working_great/

Furkan Gözükara

if working you are fine. SUPIR is standalone app right now not extension or module

VRMOTION

Sorry for the nonsense, so I don't need to install today's update? will everything work for me anyway, based on the v22 update for March 5th? I just thought that this update would work in my SD as an add-on module.

Furkan Gözükara

yes you dont need runpod at all. if you have 24 gb gpu you can run everything perfectly

VRMOTION

That is, I don't need to install RUNPOD to make everything work on the computer:?

Furkan Gözükara

100% i have everything for both locally and on cloud

VRMOTION

And is it possible to use it locally in your SD 1.5 without the Internet?

Furkan Gözükara

runpod is a cloud service which you access from browser. however the scripts made for runpod can be used on linux because runpod uses linux as operating system. so you can use runpod even from your mobile phone with browser. that is the difference.

VRMOTION

I'm a little confused. is runpod not for Windows but for Linux? can I install this on windows?

Furkan Gözükara

yes hopefully i will make script for it for both runpod and kaggle. it is a great new web ui.

Anton Kuvshinov

Thanks for your work! Can you please do one more script for runpod https://github.com/lllyasviel/stable-diffusion-webui-forge It's upgrade for auto1111 with new upgrade for controlnet extention