Depth-Anything for A1111 and InstantID

[@OlivioSarikas]

#### Links from the Video #### Depth Anything Model Link: github.com/Mikubill/sd-webui-controlnet/discussions/2549 Unofficial implementation of InstantID for ComfyUI (pain to get to run at the moment of video release): github.com/ZHO-ZHO-ZHO/ComfyUI-InstantID depth-anything.github.io/ huggingface.co/spaces/LiheYoung/Depth-Anything github.com/InstantID/InstantID?tab=readme-ov-file huggingface.co/spaces/InstantX/InstantID

[@LouisGedo]

👋

[@haofanwang9914]

Thanks for introducing our InstantID!

[@michaellorne5326]

Thank you Olivio! 3:12 "The first time you going to use this it is going download two different models 1.5gb each" For those of us who are bandwidth challenged, this is a really nice thing to know and I REALLY appreciate the quick FYI. Thank you!

[@Gwenyria]

Yay again something new for automatic1111 :D I still am to lazy to proper learn comfy UI and secretly hope that most of the comfy features come sooner or later also to Automatic1111

[@ANGEL-fg4hv]

Well welll must've hurt coming back you traitor. Automatic 1111 Stan Here 😢

[@AshT8524]

Lol

[@CoconutPete]

never noticed depth anything until I randomly stumbled upon it last night... this should make a difference in some of my images I've been struggling with

[@nikolesfrances1532]

I saw these come out and couldnt figure out how to set it up and been looking into the lack of content and direction with both of these softwares, like a prayer our shepard Olivio riegns down and deliviers! Thank You

[@Steamrick]

For SDXL, the diffusers_xl_depth_full depth model seems to pair well with the preprocessor. I've only tried two image inputs, but they were challenging ones and the other depth models for SDXL didn't seem up to the challenge.

[@JohnVanderbeck]

InstantID looks very interesting and promising. I love anything that gives me more control. Will need to try it out

[@joehogans4494]

Awesome thanks Olivio! Depth Anything Model, it's definitely also a bad hands, leg, body resolver for me as well.

[@Dmitry-n7r]

Depth-Anything is now supported & is the default depth processor in Invoke 3.6.1

[@FredPauling]

Depth Anything model is a huge leap in quality. Can't wait to see more advancements in monocular depth estimation.

[@NuKeBoX-NKBX]

Always on the hype, thanks Oliviioooo! 🐔😘

[@moonduckmaximus6404]

the depth map was not generated with your workflow. i had to take the result image and put it into control net and push the explosion button to generate the depth map

[@gimperita3035]

Thanks for the research, Olivio. As always.

[@benjamininkorea7016]

Very nice video as usual. I'm curious if there's any chance to use the sharper depth map preprocessor with older models or with the old Depth utilities. I'm interested in creating stereo images-- they used to be only so-so, but with improved sharpness, it might be worth trying again.

[@magenta6]

"As soon as I get my hands on it!"

[@Semi-Cyclops]

is it only for 1.5 not XL?

[@Steamrick]

It seems there's only a SD1.5 controlnet model so far. Of course you're welcome to try the preprocessor with any of the existing SDXL Depth models. In a quick check, diffusers_xl_depth_full seems to work pretty well with it.

[@fjccommish]

Stunning? It makes head shots. That's it.

[@paulsheriff]

might want to mention the file name must still be named control_sd15_depth_anything.safetensors or it will not be reconised

[@moviecartoonworld4459]

Thank you for sharing the latest information!!! ^----^

[@MS-gn4gl]

Why would anyone be surprised tiktok has ai projects? They've literally been building some of the most effective realtime AI tools for their filters for at least the last 4 to 5 years.

[@wykydytron]

Filters don't use ai, it's simple face detection algorytm

[@LokeKS]

@@wykydytron face detection is ai, haha

[@EarmWermChannel]

Game changer. Load up an old 3D scene, remove all textures and lights, prompt what you want. Bingo. The only downside is that it runs on 1.5 models only. So using Krea to upres to 4k

[@Vargol]

Hi All, be warned that if you try to do photo or cinematic styles with InstantID you get Getty watermarks on ~ 50% of your images. That's probably the reason those styles are missing from the HF spaces demos.

[@hamedsadeghizadeh6660]

Hi Olivio, thank you for your videos. Is need to download the Depth-anything/checkpoints, if yes, which one and where should i put them? Sorry if my question is so noobie.

[@moonduckmaximus6404]

Finally a good ole fashioned video for automatic 1111 without mentioning .....''comfy ui''

[@8eck]

Just wow, i remember how slow it was a few years ago. I think i was using MiDaS project at that time.

[@matthallett4126]

There's a Adapter Face ID controlnet you can experiment with in A1111. I tried it with my own face, and it did OK, but the checkpoint you use makes a big difference.

[@Elwaves2925]

You likely can't post a link because it's YT but can you help by pointing me in the right direction of where to look for it? Last time I searched for A1111 and FaceID I found nothing. Cheers.

[@matthallett4126]

@@Elwaves2925 Huggingface / IP Adapter FaceID /

[@bunnystrasse]

Where

[@boyblue80]

you say you must rename the file, but its a folder not a file..so which file within t hat folder must be renamed? or does the folder itself have to be named 'control_sd15_depth_anything' ?

[@anthony_leckie]

Thank you, Olivio.

[@Theexplorographer]

Excellent, thank you.

[@mayorc]

Instant ID is the same tech as that in PhotoMaker from TencenArc or an improvement over that?

[@film-alchemist]

When trying to generate the depth map I'm getting this error: "ValueError: controlnet is enabled but no input image is given" Any ideas?

[@sentinelah7]

Thank you very much!

[@NaitorStudios]

Please make a tutorial about Depth-Anything standalone, specially for video and photo editing outside Stable Diffusion

[@svenvarg6913]

How does this compare with Marigold Depth estimation and does this new depth model provide relative or absolute depth?

[@OurResistance]

Hey. Why is there no way to convert the depth map into a SBS image? I looked everywhere!

[@benjamininkorea7016]

Yeah that's what I want too.

[@kernsanders3973]

cool been waiting for this to come to A1111 EDIT: Ah rats thought from the title that InstantID was now in A1111... Guess the waiting continues The new depth processor is great, anything non realitic/3D was hit and miss with normal depth. This new one certainly feels like there are more hits than miss.

[@borutesufaibutv1115]

Depth color reminds me if lens blur from LrC haha

[@Veselin_Angelov]

Two months ago: IP Adapter comes out. Five days ago: Photomaker comes out! Today: InstantID comes out! However, even after an InstantID extension comes out for A1111, I don't think Photomaker will fall into obsolescence, because I didn't see a feature where ImstantID can merge features. Therefore: Photomaker to "bash" facial features into a new image, and then InstantID to reuse the synthetic face, equals consistent characters from reference? (I wonder if feature-bashed synthetic faces are ethical? Seriously. Any thoughts?)

[@humanbot4]

Does anyone know the notebook for comfy ui. I had tried a bunch of notebooks but they didn't work on the Amazon sage runner?

[@jonmichaelgalindo]

Please post an update for instandID once it's working in A1111 / Comfy! ❤

[@JoeBurnett]

Your tutorials are so good! Thank you!

[@hyrokon5767]

the amount of tab opened 😮 😢

[@ImmacHn]

News on this depth-map for comfy?

[@OlivioSarikas]

Apparently it was updated in controlnet there also, but i didn't have time to check yet.

[@ipwnzuall2000times]

genAI is moving so fast...i haven't gotten my head around segment anything yet and now this

[@sherpya]

we need a comparison with photomaker

[@richardgreaney]

Those of us using CPU-only, this preprocessor won't work. It seems like it requires xformers, which requires an Nvidia GPU :(

[@goliat2606]

I generated woman face that i very like. Is any way to apply this face to rest of images which i will generate? Reactor or other face replace nodes makes face very bluury when image is bigger than 512x512.

[@simpleandfrank]

I need a new PC...

[@Dohpamine_Design]

Me too 😂

[@OlivioSarikas]

I need a clone so i can do more

[@Airbender131090]

2 bad this update broke animatediff in A1111...that was the reason i updated in a first place...

[@mame947]

can comfyui use the depthanything and instantID?

[@PawFromTheBroons]

It's kinda weird of the developer to ask for its own file on GitHub to be renamed to something more sensible, while they could just as well provide said file with its proper name. 😆

[@SoCalGuitarist]

1.5 only? that's disappointing.

[@OlivioSarikas]

yes, but you can upscale in SDXL, so that's kind of a work around

[@saschascharf2932]

Taylor Swift is THE face of AI image generation. ;)

[@jumanjiro1]

Actually it’s an Asian face early on…

[@timtarbet4594]

Is this coming to Comfy?

[@OlivioSarikas]

yes :) always is :)

[@sinayagubi8805]

We need this for Comfy, I'm not comfy with A1111

[@MilesBellas]

oh no..... TT !!!!!

[@artistaartificial5635]

O:

[@Orangeduck]

[@angloland4539]