Stable Diffusion 3.5 Large & Turbo Released! And It's Uncensored!

Finally! Stable Diffusion 3.5 Large & Turbo Released! And It's Uncensored! Today I'll cover how to run it with ComfyUI and what settings to start with. I'll also show you some examples on what you can expect. Make sure to check out the links below for everything you need!

⏲Time Stamps
0:00 Intro to SD3.5
0:43 Where to get the model
1:33 SD3.5 Simple text to image workflow and settings
3:18 SD3.5 and Turbo examples
5:37 SD3.5 Text examples
6:15 Inference on an 8GB VRAM GPU
7:09 SD3.5 Training and Lora sighting

**Disclaimer Affiliate Links Below**

📸 Gear I use

🎵 Epidemic Sound

🔦 Find us on:

[MonzonMedia]

aaahhhh crap my bad folks, I made a mistake on the caption for Turbo 2:56, it should've said 4-8 steps the times at 6:15 would be faster for Turbo 8-10 seconds! I must've been thinking of Flux hahhaha!

[labmiked]

Hey there! I'm truly impressed with your approach to sharing insights on the latest AI models. It's refreshing to see content that gets straight to the point, bypassing the unnecessary fluff that often bogs down other YouTube channels with requests to join Patreon, etc. Your hands-on experimentation with Flux and now with SD 3.5 is incredibly valuable, helping both newcomers and seasoned professionals. Keep up the fantastic work, stay awesome, and enjoy the journey! ✌

[SouthbayCreations]

Thank you for the video! Great info! Looking forward to using it more to see what else we can do with it. 🙌🙌

[MonzonMedia]

Have you tried SD3.5 yet and if so...what are your initial thoughts?

[bahethelmy]

Thanks! The large is a base model unlike Flux. Finetuning off SD3.5 will be great. Also, SDXL Loras don't work with SD3.5 but the Loras that will be made for it will make it superior.

[NotThatOlivia]

just use "the fp8 scaled model" made by Comfy to get faster but still incredible quality render times

[pawelthe]

I tried the SD3.5 large model a few hours ago and I think it is more for artistic image generation and not like Flux for realistic ones, maybe I'm wrong, maybe I tested too little, or maybe I chose the wrong settings, but for now I give it an 8 out of 10.

[EskaronVokonen]

Looks promising, can't wait to try on Forge... but I guess I'll have to wait for after the release of the controlnet node for Flux... unless the delay for it is due to SD3.5 compatibility?

[thObserver]

The large model definitely Does Skin better, I'm grabbing that one. Though I'll wait for the custom SD 3.5 models to release.

[vVinchi]

It's very good model for artistic styles, good variety.

[Lykon]

The fp8 that's just a but smaller includes all text encoders, so it's overall much smaller

[Elwaves]

So far, it's okay for a base model. It's definitely miles better than SD3 but that's not exactly difficult.
For positives, it handles celeb names better than Flux, skin is slightly improved and text is okay-ish.
For negatives, model loading is very slower but generation is okay. Hands are often bad while it does boobies, ladies gardens are out. Not tried gentleman vegetables.
A lot will depend on what the community does with it, especially as many are onboard with Flux already.

[Aamir]

Great video, i'm waiting for oct 29th, the medium version will be released then. maybe i can run that on 8gb with ease.

[halilnevzat]

can you share your example workflow json ?

[EvolKulture]

Hi! New to this stuff, but I have no idea what what’s onscreen (the boxes and lines all over the screen), is this not like MJ and Leonardo where you just go to a site and start prompting??

[HikingWithCooper]

I think it's crazy that this started (basically) with a Midjourney prompt. Easy. You type in what you want, it makes it. Now you have to understand node workflows, be able to program in json, know how to problem solve errors such as 'g' (that was literally an error message I got in Comfy yesterday), know what nodes you need and oh so much more.... just to make an image of a person who hopefully has 5 fingers.

[ditheraith]

0:13 we can do women laying on grass with 6 fingers*

[OleFromTheNorth]

Is this finally able to create "A black man with a red shirt shakes the hand of a white man with a blue shirt akd sumglasses on"? On SDXL it almost never got these things correct. Flux was way better at this.

[BenSchewe]

"it gives you Bobas".... what is Bobas?

[jibcot]

Dude, you need to get a better graphics card! :)