| CARVIEW |
Navigation Menu
-
Notifications
You must be signed in to change notification settings - Fork 2k
Replies: 67 comments · 147 replies
-
|
only XL? |
Beta Was this translation helpful? Give feedback.
All reactions
-
😕 9
-
|
Please tell me where to place the antelopev2 model in the Forge file path. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
E:\webui_forge\webui\models\insightface\models\antelopev2 |
Beta Was this translation helpful? Give feedback.
All reactions
-
❤️ 1
-
|
Your help is godsent, you're the kindest soul around, thank you so much, blessings will come your way for sure. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
where i put this path |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
|
Beta Was this translation helpful? Give feedback.
All reactions
-
|
It does work but yeah, it loads the models over and over and over which takes like over minute of waiting time on 3090, so each image takes almost 2 minutes to generate cause of loading times, even if you wont change any reference images. |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 5
-
|
The results I'm getting are very similar to the above. For some reason, they look quite "washed out" of color. Not sure why. |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
|
you mean the contrast? i do painting by vallejo and frazetta prompt , maybe try to up the cfg scale to get more contrast, i personally dont like it and i use levels later to bring contrast back.All those are CFG 3 and 20steps |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I also have an 3090RTX (i9-12900K, 32GB RAM) and I generate (with Instant_ID) a 1024x1024 image in 21sec, and with highres fix x2, it takes 1m15s. It's long, but far from 2 minutes. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
|
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 3
-
|
Works great, thank you for your efforts. Things to watch out for at the moment: |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 15 -
❤️ 2
-
|
Haha, no, InstantID only detects face keypoints, and the body pose is random. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
well, my results has something obscuring the face (fingers or warrdobe element or hair ) way way too often compared to not using instantID, especially near lips/jaw which is superstrange and i dont think its a coincidence |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Thanks for sharing @hben35096 . may have meant up to 1024^2, these settings look good for me at (1024,1024) 🍻 |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Could you please guide me on how to fix this error? Thank you. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I'm gettting very strange behaviour if I set the size below 10241024 then controlnet doesn't even work!? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
awesome this time i didnt make gradio and now implemented :D if face input not changed you really should cache it i did this for IP Adapter Face ID and it did speed up huge. caching the calculated face input vectors |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
We do have preprocessor result cache here: sd-webui-controlnet/scripts/global_state.py Lines 22 to 49 in a67d040
--controlnet-loglevel DEBUG, you will see whether the cached preprocessor result is used.
|
Beta Was this translation helpful? Give feedback.
All reactions
-
|
so if models are cached, why it takes so long and models are reloaded each time it generates ? Nothing should be reloaded if nothing was changed. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
If you are talking about load of ControlNet/IP-Adapter models, you should set model cache size to a value > 1. (By default that config is 1. As we are using 2 models together, the cache size of 1 means we miss the cache every time.) |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 11
-
|
DAMN! This is it, can you make it default controlnet setting ? Now i can generate non stop with like 8 seconds per image, its great. i dont think people know this and it changes A LOT from 2 minutes to 8 seconds. |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 3
-
|
this info super important thanks |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 2
-
|
100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:43<00:00, 2.19s/it] |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Please use SDXL base model. Currently there is no instant id for sd15. |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
|
Are there plans for sd15 support? |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 7
-
|
You can file a request in instant id repo here: https://github.com/InstantID/InstantID. I don't have the resource to train model. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
There's this
Seems that the pre-processor preview works when clicking the 💥( Tested that preview appears to be working with the following pairs:
Expand for
|
Beta Was this translation helpful? Give feedback.
All reactions
-
|
*** Error running process: D:\Webui1111\Webui\stable-diffusion-portable-main\extensions\sd-webui-controlnet\scripts\controlnet.py |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I get an assertion error using all the same settings as above with an SDXL model of course.
|
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Where you able to fix this? I've got the same error |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
|
The same error. Did you find the solution for that? |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
|
Try another XL model |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
|
i've got this error with Pony Diffusion V6 XL checkpoint |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Working fine. Thank you. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Can you link to the implementation code to InstandID's replicate demo? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
https://replicate.com/zsxkib/instant-id though im not sure i understand what he means. Uploading the same image for the control would do the trick already. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I don't think the demo site supports generating random head position when you don't specify a pose. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Currently it seems the second image is defining not only the position but also the size of the head in the image. It would be good, if the very well working face reproduction could be used freely like in face-id or an edit function could be used :) |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Does it work also in img2img? More or less like "roop" ? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Thanks for your work! However, after playing around with this for a bit, something seems off compared to using InstantId through the new Diffusers pipeline. Let me demonstrate with an example (note, however, that the same happens with every image combination I've tried). I'm using a picture of (Daniel Radcliffe as) Harry Potter in ControlUnit 0 (ip-adapter) and a picture of Obama in ControlUnit 1. Here's the result with the "Control Weight" of Unit 0 set equal to 1.0:
Obviously, the quality is rather poor. The CFG scale is set to 4.0 and I use the euler-a sampler. Now, if I turn the "Control Weight" of Unit 0 down to 0.0, I get this:
This looks just slightly less like Radcliffe's Potter than the first image, but the quality is obviously vastly better. If I try values of Unit 0's "Control Weight" between 0.0 and 1.0, the quality is always degraded (although not as much as at 1.0) – even at low values like 0.2, the output image is noticeably blurry. If I keep the "Control Weight" at 1.0 but decrease the CFG scale to 3.0 or 2.0, the quality improves somewhat but the blurriness is still there (and the contrast gets worse):
Across several different input image combinations I've found that I get the best compromise between quality and identity fidelity when ControlUnit 0's "Control Weight" is 0.0 or a very low value like 0.05 or 0.1, and the CFG scale is as high as I can push it without getting too much contrast (usually around 4.0). That's somewhat odd, and I don't experience the same when using InstantId directly – the best combination of weights is then something like 0.8 and 0.8. Any thoughts? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
20-30 seconds on a g5 ec2 instance and a rtx4090 rig. both with 24gb vram. Generation times are quite a bit high with instantid. Only problem was forgetting to put the models in the proper folder, other than that there are 2 different comfy node set. ZHO was a little more flexible but both are far more restricted than the a1111 versions since they are just diffusers wrappers. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Excuse me, what is the cause of this clarity problem in the webui and is there a solution |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Hi! Can you show me the prompt (or model) for this amazing retro pink style? Many thanks!! |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Prompt: Image Width: 960 px |
Beta Was this translation helpful? Give feedback.
All reactions
-
❤️ 1
-
|
Man its been a while lol. But images should contain metada, try to drop them on comfy or a1111. |
Beta Was this translation helpful? Give feedback.
All reactions
-
❤️ 1
-
|
How to deal with multiple faces in one picture? i tested it on the webui, and instantid only handled one face |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Seems to download the onnx files and then complains about the onnxruntime. I tried deleting and forcing it to reinstall the onnx files and the onnxruntime from the venv, but no luck. Thoughts? Downloading: "https://huggingface.co/DIAMONIK7777/antelopev2/resolve/main/1k3d68.onnx" to D:\stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads\insightface\models\antelopev2\1k3d68.onnx 100%|███████████████████████████████████████████████████████████████████████████████| 137M/137M [00:10<00:00, 14.0MB/s] 100%|█████████████████████████████████████████████████████████████████████████████| 4.80M/4.80M [00:00<00:00, 12.8MB/s] 100%|█████████████████████████████████████████████████████████████████████████████| 1.26M/1.26M [00:00<00:00, 8.26MB/s] 100%|███████████████████████████████████████████████████████████████████████████████| 249M/249M [00:19<00:00, 13.6MB/s] 100%|█████████████████████████████████████████████████████████████████████████████| 16.1M/16.1M [00:01<00:00, 13.1MB/s] Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}} Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}} Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}} Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}} Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}} |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
it couldnt detect face in your input image |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
your face must have neck visible, otherwise it wont be detected |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
|
Great feature, thanks a lot !! |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Hello, I have this error in Automatic1111 for Mac M2 *** Error running process: /Users/cinesaibai/pinokio/api/sd-webui.pinokio.git/automatic1111/extensions/sd-webui-controlnet/scripts/controlnet.py |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
ModuleNotFoundError: No module named 'insightface' |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I understood, but I have installed everything and I find automatic1111/extensions/sd-webui-controlnet/annotator/downloads/insightface/models with all the files |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Please read sd-webui-reactor README about install insightface on Windows - https://github.com/Gourieff/sd-webui-reactor |
Beta Was this translation helpful? Give feedback.
All reactions
-
Beta Was this translation helpful? Give feedback.
All reactions
-
|
It works absolute best if your input face have half shadow on it so the volume of the face is read better by ipadapter , with flat light stallone was unrecognizable |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 8
-
|
Amazing results, just how did you got that output?? checkpoint, did you add lora or just prompt, or was because of the control_instantID |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I think this was nightvision xl but now i use helloworld sdxl from civit, xl is getting better but more slowly than 1.5, no refiner movie still of man wearing tank top, covered in blood standing on street full of intestines meat blood ,exploding cars ,backlight ,haze |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 5
-
|
I uninstalled the CN on my computer and removed it from the https://github.com/huchenlei/sd-webui-controlnet/tree/fix_insight?tab=readme -OV file has re downloaded the CN update, and also re downloaded two processors and models, but the problem still exists. May I ask if the steps are incorrect. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I tried this out but the results look wonky. Without the ControlNet (which has been updated), it (SDXL with its accompanying VAE) can generate images fine. Let me know what I should do. watercolor painting I also got these if they may affect the result: |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
It ran pretty well for me. I used one image Looks better than anything I've done with ReActor. |
Beta Was this translation helpful? Give feedback.
All reactions
-
😄 1 -
❤️ 1
-
|
Now that I've worked with this a bit more, using all the same stuff. I think your going to have a heck of a time getting a more posable head and neck with only those 5 facial keypoints. You'll also need a way measure and alter lighting/light source, which is it's own area of expertise when it comes to cgi work in films and animated films. For fun, I layered the: facial keypoints, over an opacity altered version of my AI generated image, over the original image I got from a random google search. Here's the result: Anyway, just wanted to share my thoughts. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Is there still no way to edit the keypoints or even create keypoints from scratch without using a reference image? The is an 'edit' button which opens Photopea - but I couldn't figure out what this is supposed to do, since Photopea doesn't seem to treat the keypoints as editable vectors - and even if it did - how would I save the edit so that it is then used by A1111 again? Am I missing a major thing here? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
They said it in the main post: |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Yes, I read this - but it is a bit confusing that there is already an 'edit' button which in fact does nothing useful - without even a hint or a tooltip or - gosh - some kind of documentation. But what do you expect from people who think that the 'Gradio' GUI is 'fun' :-) The whole Open Source AI developer community got the focus completely wrong on 'features' and 'extensions' that are a dime a dozen - instead of taking the most obvious usability aspects into account. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Oh, that's just the controllnet mask. The only thing you can edit about that is by drawing on it like a canvas I think, but that would ruin it. We're pretty lucky to have Gradio & A1111 & StableForgeUI & ComfyUI & StableSwarmUI. Otherwise, we'd have to do this all through CLI or CMD. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Guess it is debatable whether a bad GUI is better than a bunch of intelligent scripts - with good documentation and adequate examples. Being a software developer for decades myself I would very much prefer using image generation AI the way a 'normal' script with function calls works (heck - there is only a very minor 'visualization' part in the AI GUIs - the complete rest is a mere parameter and settings orgy and nested function calls, isn't it?). This would be WAY more transparent than what the GUIs now do - and don't get me started about the chaotic spaghetti nonsense of (Un-)ComfyUI. Look at how you normally build even complex web apps with (script) frameworks - include jQuery, and the stage is already set for most operations. It tells a lot that A1111 to the date isn't even capable to save and reload a project file which includes ALL settings and parameters and referenced file pathnames to recreate the very same image you created before. What are those people thinking who develop this? I have never in decades come across any reasonable GUI which couldn't do this. Or do you know of any system where you have to keep manual annotations on top of a project file to simply continue your work where you left off last week? |
Beta Was this translation helpful? Give feedback.
All reactions
-
🚀 1
-
|
I found
|
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Has anyone been able to get this working with Oobabooga text generation web ui via the API? Seems there would be some amazing possibilities... |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
After the 1.1.447 update, Instant ID stopped working, it gives an error |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I am facing same issue as well! Need to wait fix |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
New update should fix this. It is really wield why the full qualified name reaches here. Can I confirm the gradio version you folks used? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Thanks, new update fixed the problem. gradio version 3.41.2 |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
There is a quick fix here: #2849. A more proper fix & large refactoring on the way: #2847 |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
|
Hi there! Any tips for reduce the blurriness / improve the quality of the image ? Im using :
Only stuff I imagine can improve the quality is pixel perfect on both unit but my laptop is not powerful... worth to use it if I rent a gpu cloud ? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
hey can we use this instant id in stable diffusion forge |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
为什么我生成的效果这么差 |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
俺也一样 |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
will this model be updated for the FLUX model? |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
All this does not work now, I tried it on different builds of SD. Automatic 1111 version v1.10.0, model juggernautXL_version6Rundiffusion.safetensors [1fe6c7ec54] although I tried another XL model, I tried different resolutions. In the first tab ControlNet Unit 0 [Instant-ID] Preprocessor instant_id_face_embedding, Model ip-adapter_instant_id_sdxl [eb2d3ec0], the second tab ControlNet Unit 1 [Instant-ID] Preprocessor instant_id_face_keypoints, Model control_instant_id_sdxl [c5c25a50] nothing happens, the face does not change. ControlNet was reinstalling. Does it even work for anyone now??? |
Beta Was this translation helpful? Give feedback.
All reactions
-
👎 1
-
|
Forge is now just a preview browser for Flux. They don't give a shit about actually using it in a professional or creative setting. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Why do the characters I generate always appear in the middle of the image without any action, even though I specifically include actions in my prompt? I tried many many times, different prompts, control weight, and so on... but no lucky. Did I miss something? mid shot,bust of A man is playing golf and is about to hit the ball,from front, Steps: 28, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 3, Seed: 2070426131, Size: 1280x768, Model: DreamShaperXL_Turbo_v2_1, VAE: sdxl_vae.safetensors, ControlNet 0: "Module: instant_id_face_embedding, Model: ip-adapter_instant_id_sdxl [eb2d3ec0], Weight: 1.0, Resize Mode: Resize and Fill, Processor Res: 768, Threshold A: 0.5, Threshold B: 0.5, Guidance Start: 0.0, Guidance End: 1.0, Pixel Perfect: True, Control Mode: My prompt is more important", ControlNet 1: "Module: instant_id_face_keypoints, Model: control_instant_id_sdxl [c5c25a50], Weight: 1.0, Resize Mode: Resize and Fill, Processor Res: 512, Threshold A: 0.5, Threshold B: 0.5, Guidance Start: 0.0, Guidance End: 1.0, Pixel Perfect: True, Control Mode: My prompt is more important", Version: 1.10.1 |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
InstantID finds the face and generates what's around with your prompt. If your face ends up always in the middle that's because the head in your image is... always in the middle. Duh. Move the head or the whole image, doesn't matter.
|
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I would like to ask a question. When using Instant ID, both the preprocessor and model have been configured accordingly. Clicking "Allow Preview" will display an error. console output |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Instant ID project
https://github.com/InstantID/InstantID
Instant ID uses a combination of ControlNet and IP-Adapter to control the facial features in the diffusion process. One unique design for Instant ID is that it passes facial embedding from IP-Adapter projection as crossattn input to the ControlNet unet. Normally the crossattn input to the ControlNet unet is prompt's text embedding.
Download models
You need to download following models and put them under
{A1111_root}/models/ControlNetdirectory. It is also required to rename models toip-adapter_instant_id_sdxlandcontrol_instant_id_sdxlso that they can be correctly recognized by the extension.How to use
InstantID takes 2 models on the UI. You should always set the ipadapter model as first model, as the ControlNet model takes the output from the ipadapter model. (ipadapter model should be hooked first)
Unit 0 Setting
You must set ip-adapter unit right before the ControlNet unit. The projected face embedding output of IP-Adapter unit will be used as part of input to the next ControlNet unit.
Unit 1 Setting
The ControlNet unit accepts a keypoint map of 5 facial keypoints. You are not restricted to use the facial keypoints of the same person you used in Unit 0. Here I use a different person's facial keypoints.
CFG
It is recommended to set CFG 4~5 to get best result. Depending on sampling method and base model this number may vary, but generally you need to use CFG scale a little bit less than normal CFG.
Output
Follow-up work
Note
As the insightface's github release currently do not have antelopev2 model, we are downloading from a huggingface mirror https://huggingface.co/DIAMONIK7777/antelopev2. If you are in mainland China and don't have good internet connection to huggingface, you can manually download the model from somewhere else and place them under
extensions/sd-webui-controlnet/annotators/downloads/insightface/models/antelopev2.Known issues
Beta Was this translation helpful? Give feedback.
All reactions