#### Links from my Video #### You NEED this lora: civitai.com/models/796382/ultrarealistic-lora-project huggingface.co/gemasai/4x_NMKD-Siax_200k/tree/main www.reddit.com/r/StableDiffusion/comments/1fwa99o/some_paparazzi_style_photos/
I believe VAE and Text Encoders are already embedded in dev-bnb-nf4-v2. I tried removing them from the prompt and Forge produces exactly the same image.
Because the images have more realistic lighting, the anatomical mistakes make them even more bizarre. It's worse than the uncanny valley sensation. But I agree with you, Olivio. All of this should be free (considering commercial licensing issues, image theft laws, and other risks), so running it locally is the best option for those who will use it.
I haven't been following local image generation stuff for a while, and sorry for asking a question that has probably been answered a million times already, but what are the V-RAM requirements to make good looking images with Flux lately. I think the last time I checked the absolute minimum was 12GB?
My beef with ultimate sd upscale is that there is no option to render more than 1 tile at once. (Based on the documentation i found online) This is an issue since flux takes more time moving data and loras to and from the VRAM than the actual render. 4 is the optimal amount if you want to generate plenty of images, and tile upscaling.
Added a set of high resolution and high res with hires.fix. ( it is not uncommon for hires.fix to alter the token count to produce different results, but this is an easy fix in parameters to match the before hires.fix image.
Having an episode of prompt ideas with camera settings and positions would be nice. This is where I fall short in my prompting and you probably need some background in photography to really understand if the AI did it right.
I didn't know about SD Ultimate Upscaler so thanks for that. With that hair and beard Hackerman is starting to look a little werewolf-ish this month. He'll soon be the Howlin' Hackerman when he sees an AI moon.
![[Full Episode] The Restaurant War Thailand ศึกพ่อค้าซ่าแม่ค้าแซ่บ Episode 3 | 6 ต.ค. 67](http://i.ytimg.com/vi/fTfIMFVjt_8/mqdefault.jpg)
#### Links from my Video ####
You NEED this lora: civitai.com/models/796382/ultrarealistic-lora-project
huggingface.co/gemasai/4x_NMKD-Siax_200k/tree/main
www.reddit.com/r/StableDiffusion/comments/1fwa99o/some_paparazzi_style_photos/
Hi 👋
Thank you
Please share the file that is in your video, but the author deleted it - clip_vision_l
@@BIMMERRUS I think it is this model.safetensor file here, but you need to rename it: huggingface.co/openai/clip-vit-large-patch14/tree/main
"im broke but im free so i render locally " amazing
"I'm broke, but I'm free. So, I render locally." put it on a t-shirt!
I love that you're letting your hair down.
Great prompt engineering!
Heilige 80er Granate! Diese Perücke ist ne Waffe, Olivio. Bro is having style for sure!
Vokuhila (vorne kurz, hinten lang)
He’s back!
I am broke
because i bought a 4090 and a 13900k
@@TR-707 guess what...you are not the only one 😁
I believe VAE and Text Encoders are already embedded in dev-bnb-nf4-v2. I tried removing them from the prompt and Forge produces exactly the same image.
Olivio with the 80's wig? Instalike!
I just got clip l but not clip vision l, where do I find it
Because the images have more realistic lighting, the anatomical mistakes make them even more bizarre. It's worse than the uncanny valley sensation.
But I agree with you, Olivio. All of this should be free (considering commercial licensing issues, image theft laws, and other risks), so running it locally is the best option for those who will use it.
NICE!
nice hair (again)
what is the model from the guy in bottom right ? i like the details from hair and bear
I haven't been following local image generation stuff for a while, and sorry for asking a question that has probably been answered a million times already, but what are the V-RAM requirements to make good looking images with Flux lately. I think the last time I checked the absolute minimum was 12GB?
@@CaptainPanick I have 12 GB of VRAM. It works well (1 to 2 minutes by image) but I use GGUF version of Flux.
I have on my laptop a 3060 with 6Gb vram and I have no problems running Flux with Forge
My beef with ultimate sd upscale is that there is no option to render more than 1 tile at once. (Based on the documentation i found online)
This is an issue since flux takes more time moving data and loras to and from the VRAM than the actual render.
4 is the optimal amount if you want to generate plenty of images, and tile upscaling.
Added a set of high resolution and high res with hires.fix. ( it is not uncommon for hires.fix to alter the token count to produce different results, but this is an easy fix in parameters to match the before hires.fix image.
Rocking the hair
Great tips for the prompt!
How to turn on VAE text encoder, UI selection Flux SDxL option, on Stable Diffusion?
Traumhaft! Danke!
0:25 I thought the last video was for 1 percenters. 4 cents per video. Outrageous..
How to install flux locally though?
I'm pretty sure it's available on Pinokio.
Having an episode of prompt ideas with camera settings and positions would be nice. This is where I fall short in my prompting and you probably need some background in photography to really understand if the AI did it right.
Olivio, I've watched this video several times and I feel like there's something different about you. Can't quite put my finger on it though.
Except for the curse of the dreadful " Cleft chin" that Flux D loves to put it everywhere.
Train your own Lora and cleft chin is gone
@@Edacuatica the issue I'm having with flux is that there's a 70% chance that the subject will put their hand in a fake skirt pocket.
People want cameltoes? We.ll give them face cameltoes!
where do i drop the 4x NMKD?
ESRGAN folder /facepalm
Hello Olivio, where can I get the VAE encoder t5 xxl fp16.safetensors? I haven't found it anywhere. Thank you
1:07 and the hands are…..
they can be fixed later
@@ShubzGhuman Same time ist korrekt
I didn't know about SD Ultimate Upscaler so thanks for that.
With that hair and beard Hackerman is starting to look a little werewolf-ish this month. He'll soon be the Howlin' Hackerman when he sees an AI moon.
It makes the same female face over and over
Background environments look great but the skin looks soapy/blurred and is lacking detail.
SUBSCRICE.....AYE AYE
You look like a Peeping Tom lol 😂 😂