Consistent Characters in Stable diffusion Same Face and Clothes Techniques and tips

[@ClaudioMalagrino]

At last a good video without bad edited "fx" or exhibition of unwanted skills by the presenter! Thank you, I'm following your channel.

[@AI-HowTo]

Thank you.

[@boraicho01]

Good tutorial and explanation , 👍

[@mitch8541]

This is the video I have been looking for!

[@timothywells8589]

Thank you so much for such a clear tutorial!

[@AI-HowTo]

You are welcome!

[@Ai_Vs_Original]

perfect❤

[@widitruth]

Thanks, excellent video

[@christophermoonlightproduction]

That was really good. Thank you.

[@DewanshuHaswani]

you are smart

[@thekotfather]

respect!

[@loquiman]

Very nice video! I have two questions. Are you using SD on your PC or is it online version? What do you mean to use blender to have the same consistency? Like to use some Ai plugin in blender or just model the whole thing? Thx

[@AI-HowTo]

Thanks, Yes I am using SD on my Laptop, Graphics card is RTX 3070, I meant for blender, that 100% consistency cannot be achieved in stable diffusion or diffusion based models, thus the used of software tools (3D tools) such as Blender is the way to go if someone wants 100% accuracy, this is thus far true.

[@loquiman]

@@AI-HowTo Thank you for reply!

[@RockG.o.d]

did you remove the background and generate almost the same background? anyway thanks for the video.

[@aiartbx]

Between prompting, lora, embedding and dreambooth model trained on a single person with single clothing, which method would give you the most consistent results when using high denoising in img2img?

[@AI-HowTo]

I think dreambooth remains always the best option for consistency, after that, with proper prompting, even in img2img i still use after detailer too it is really great there too + controlnet if you are planning on high denoising strength, no need for controlnet though with low denosing.... still I dont think it is possible to achieve high level of consistency for large number of images(such as long video frame sequence) with high denosing strength.

[@danishraza8843]

probably Dreambooth model since it would not know anything else beside what it is trained on. like famous characters, Models know that they generally look like, celebrities and such. use a name as ID and it might help SD focus its efforts to replicate that ID instead of picking samples from general datasets. all the famous female models such as majikmix and such have a general single female face which is so consistent that you can easily tell which model they have used.

[@aiartbx]

@@danishraza8843 so how did they train models like majikmix? Is the face trained on a 3d model?

[@AI-HowTo]

yes, some check points are trained on 3D Models, because you can capture the body from every direction, and have same body/face all the time in all training images, it has low noise in the skin, it is just perfect, then one can merge it with another realistic model with some strength to make the model more realistic ... training the SD will not make it forget its previous knowledge of most subjects, it mainly updates the token that it saw in its training data such as Woman, Clothes...etc. with the new data.

[@aiartbx]

@AI-HowTo interesting. So if I use a realistic model as a base then train it on one 3d character body and face it will output a realistic version of that character? If so would I caption the 3d images as 3d person then when I create a image using it I would ask for a realistic woman?

[@user-oo1sk1oi4e]

This is so helpful video, thank you! My UI is not same as your UI. So which extansion or models do you use? can you share the links that you have?

[@AI-HowTo]

Thanks, A1111 updates every now and then, but the overall look and feel is the same. The A1111 version here is 16.0 I think, you can upgrade to recent version based on this video kzbin.info/www/bejne/bWKrhaeIiL2md8U ... some stuff may not function well there still suh as Mov2Mov and few other extensions. some UI Features can be added from (Settings, User interface, quick settings, sd vae, clip stop, img2img color correct, noise multiplier...etc... Most important extension in A1111 is !After detailer remaining are controlnet, roop...etc. all installed from extension tab.

[@user-oo1sk1oi4e]

Thank you so much@@AI-HowTo

[@DeFirm-]

hai bro, can you make video about changing everything (inpaint anything) except the subject, (include the cloth and pose) please ^^

[@AI-HowTo]

this is exactly what this video does, same subject, different poses/clothes/backgrounds/settings

[@jamesc.2907]

8:16 the ui tells me the batch directory is unfindable. Mine starts with C: and not E:, but overall seems similar to yours. Any ideas? Your video is very informative, thanks

[@AI-HowTo]

You are welcome, I often copy the path from the navigation bar directly, to avoid any issues, the message you are getting means there is a typo in the path (i think).

[@jamesc.2907]

@@AI-HowTo thanks, fixed it

[@stepphen_ffc5606]

Can you show us a video about good prompts? or is there recommended links about good prompts?

[@AI-HowTo]

will check if i can later. one can learn from citivit ai which has example images that shows the prompts used.... mostly prompting is experimental, sometimes partially based on recommendations related to a specific check point for instance if you check realistic vision civitai.com/models/4201/realistic-vision you can see in description example recommended prompts such as Prompt: RAW photo, subject, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3 ...etc. and example negatives... so does every check point ... I often use simple negatives, and use simple prompts as in this video which i found effective and expand based upon that trying to get something acceptable.

[@Hooooodad]

Great video. Can you share some of your prompts ?

[@AI-HowTo]

Thanks, all prompts appear in the video in details, for the girl, a sample prompt for the face for example is all are simple straight forward prompts. parameters masterpiece, best quality, RAW photo , portrait of a beautiful woman mariasofianblacksmithvereralilly, blue eyes, redhead, lipstick, lips, smile, long hair, intricate details,detailed skin texture, (blush:0.5), (goosebumps:0.5), subsurface scattering, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3, Negative prompt: low quality, worst quality, bad anatomy,bad composition, poor, low effort, blurry Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 1981345104, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, ENSD: 31337, Version: v1.6.0 which i used in the after detailer for the face. simple white shirt prompts is parameters masterpiece, best quality, RAW photo , a beautiful woman mariasofianblacksmithvereralilly, blue eyes, redhead, lipstick, lips, smile, long hair, intricate details,detailed skin texture, (blush:0.5), (goosebumps:0.5), subsurface scattering,cowboy_shot,wearing white shirt, blue jeans, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3, hyper realistic, detailed, empty city street in background Negative prompt: low quality, worst quality, bad anatomy,bad composition, poor, low effort, blurry, ((hat, tatoo)), naked, nude Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 461690003, Size: 512x768, Model hash: ec41bd2a82, Model: photon_v1, ENSD: 31337, ADetailer model: face_yolov8n.pt, ADetailer prompt: "masterpiece, best quality, RAW photo , portrait of a beautiful woman mariasofianblacksmithvereralilly, blue eyes, redhead, lipstick, lips, smile, long hair, intricate details,detailed skin texture, (blush:0.5), (goosebumps:0.5), subsurface scattering, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3,", ADetailer negative prompt: "low quality, worst quality, bad anatomy,bad composition, poor, low effort, blurry", ADetailer confidence: 0.3, ADetailer dilate/erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 23.9.1, Noise multiplier: 0, Version: v1.6.0 one with controlnet parameters masterpiece, best quality, RAW photo , a beautiful woman mariasofianblacksmithvereralilly, blue eyes, redhead, lipstick, lips, smile, long hair, intricate details,detailed skin texture, (blush:0.5), (goosebumps:0.5), subsurface scattering,cowboy_shot,((wearing short yellow winter jacket on top of a white shirt)), wearing blue jeans, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3, hyper realistic, detailed, empty city street in background Negative prompt: low quality, worst quality, bad anatomy,bad composition, poor, low effort, blurry, ((hat, tatoo)), naked, nude, black_jacket, long_jacket,blue_jacket Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 1389853263, Size: 512x768, Model hash: ec41bd2a82, Model: photon_v1, ENSD: 31337, ADetailer model: face_yolov8n.pt, ADetailer prompt: "masterpiece, best quality, RAW photo , portrait of a beautiful woman mariasofianblacksmithvereralilly, blue eyes, redhead, lipstick, lips, smile, long hair, intricate details,detailed skin texture, (blush:0.5), (goosebumps:0.5), subsurface scattering, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3,", ADetailer negative prompt: "low quality, worst quality, bad anatomy,bad composition, poor, low effort, blurry", ADetailer confidence: 0.3, ADetailer dilate/erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 23.9.1, ControlNet 0: "Module: reference_only, Model: None, Weight: 1, Resize Mode: Just Resize, Low Vram: False, Threshold A: 0.8, Guidance Start: 0, Guidance End: 1, Pixel Perfect: False, Control Mode: Balanced", Noise multiplier: 0, Version: v1.6.0

[@dannynguyen123]

Great Video, I manage to get the same face each time, but how can I save the face for future use? Seems like once I close the program I can no longer regenerate the same face again? Any help would be greatly appreciated, keep up the amazing work!

[@AI-HowTo]

Thanks, it is great to know. to get same face, you need to use same prompt in the after detailer, and preferrably the same seed in some cases... you can drag and drop an image that you generated before (PNG files) onto the prompt, and it will give you back its prompt, then you can click on the blude arrow under Generate button and it will fill all details such as seed/after detailer info etc so that you can use it again.

[@vektor3dx539]

How do you get such great looking hands?

[@AI-HowTo]

by chance, some checkpoints also produce better hands than others ... too many negatives in some cases may also affect this in a bad way ... sometimes hands could also be corrected with after detailer second tab and adding better hands lora there.

[@gilgameshgaming4210]

what if we refreshed the page, does after detailer going to save the prompt to generate consistent character? if not how do we do that?

[@AI-HowTo]

if you click on the blue icon(with arror) under Generate Button it will automatically bring back the recent prompt you used even after Refresh, including the seed and After detailer prompt.... if you want the prompt you used from an older image you generate, you can drag drop the PNG image into the Prompt and it will give you the whole prompt again, click the blue icon and it populates all the fields in A1111.

[@gilgameshgaming4210]

@@AI-HowTo Thank you! that's a game changer for me! but i tried to drag drop a photo edited in roop and it didn't work, is there a fix?

[@AI-HowTo]

I think this might be because in A1111 Roop rewrites the PNG file meta data, because Roop applies face morphing after the image is generated so the PNG info is possibly lost, this doesnt Happen with ReActor however, I removed Roop when I installed ReActor which is like a minor upgrade to Roop... some guys reports getting errors when they tried installing ReActor though (Roop must be removed before trying)... so the fix is to use ReActor for isntance.

[@gilgameshgaming4210]

@@AI-HowTo thank you for the answer. I tried to install reactor but it didn't work, i will try to reinstall it after removing roop

[@stanleywtang]

You missed the part that you set the batch count to 30 in order to get the same face on all 30 images

[@AI-HowTo]

for controlnet, we only use batch count 1, and images generated are based on number of pose pictures used in the batch controlnet directory, for other examples, I think i used batch 30 for some, 12 for others, some i missed showing in the video possibly to shorted the video I think.

[@mach_j]

Hi, excellent videos! However what I need is I think a mix of several techniques. Is there a way to contact your for more details? I'd like to have a call if you're open to something like that. Thank you.

[@AI-HowTo]

Thank you. and sorry, not taking private or business communications at the time being... in SD there are ways to some something, all of them are mix of different techniques, SD alone cannot do much... so we have to rely certain work flows or on other tools if necessary to complete any given task.

[@user-gq2bq3zf1f]

If you already have other photos as jpgs, how do you change the skirts in one photo, and then apply those skirts to the models in the other jpgs?

[@AI-HowTo]

This cannot be done automatically, inpainting is necssary to do this, for example you can mask the skirt in the first photo , then feed it to IP Adapter, and mask the skirt in the other photo and apply inpainting to it, I will publish a video about chaning clothes based on a reference image as soon as i can, and automatic cloth changes based on prompts in my next video.... basically there is no too simple way to do this, and even then, the results will no be very accurate because SD is not so much about accuracy, it has chaotic outputs, but approximation can be done using control nets.

[@Hshjshshjsj72727]

For better results we should use what? Splinter and what else did you say

[@AI-HowTo]

I think, for perfect consistency, Blender (3D Software) is the right tool, stable diffusion is not the one for perfect consistency of objects/faces/clothes, while one can still achieve good resutls with it.

[@Hshjshshjsj72727]

Thank you. Do you have video on that you can add link yet? Also, would it allow me to create photorealistic portraits like for social media, as that is my goal.

[@AI-HowTo]

sorry, Blender is a 3D Free software tool, it is for games and realistic 3D stuff and not AI based.... for social media stuff such as creating realistic pictures, then Stable diffusion is really good to create photo realistic images, you can check other videos in this channel hopefully you find something useful, the only drawback for stable diffusion i think is just the Computing power (good graphics card) which makes it difficult for most people to really dive into it and it's capabilities.

[@AI-HowTo]

kzbin.info/www/bejne/rHKVp2V_qsx-bNk this video shows how one can create a LoRA model that can be used for instance for creating anything related to a specific photo realistic character ... two LoRAs could also be combined to create a new Character with features from both using specific ratio of both ... or one can use the same LoRA with 0.7 weight to get a different variation of the same character... there are no limits on what one can do .... if you were new to Stable diffusion then kzbin.info/www/bejne/iKXNdaatl7h4e7s could be a good starting point ... and you may want to check other creatos content such as www.youtube.com/@sebastiankamph Sebastian has lots of content about creating stuff or www.youtube.com/@OlivioSarikas Olivio start is also fun to watch ... my content contains just limited number of videos and mostly about techniques such as LoRA creation, training, or provides tools and techniques to create certain stuff that other creators may not have covered it properly or in depth.

[@Hshjshshjsj72727]

@@AI-HowTooh excellent, thank you very much 😊

[@ggrrr4237]

i have a T-Shirt design in which there are some certain designs .. but Im not able to get those designs on Models clothes its just clothes that been referred.. Can you help me with that?

[@AI-HowTo]

I am sorry, I am not working in this area at the time being...but to get your clothes shown in the SD model when dressing the character, you need to do Cloth training using a LoRA model such as in this video kzbin.info/www/bejne/rXu7ZZV4qamlb7s which is a time consuming process Alternatively, you can try using something called: Controlnets and use a new Controlnet which is IP Adapter, upload the clothes image and try to describe it, the Controlnet will try to use the clothes you provided onto the dressing of the character, however, this will not provide accurate enough dressing compared to LoRA training... even with LoRA training the accuracy of dressing a certain design is not gauranteed unfortunately.

[@user-tl5lh7yg8k]

Hello, how can I have the same openpose patch as you?

[@AI-HowTo]

from civitai there are plenty of ready to use poses civitai.com/search/models?modelType=Poses&sortBy=models_v2 check something suitable for you, it has many large number of pose collections

[@slookify]

its always the same scene

[@alexlefkowitz]

Hey how can we reach you for business inquiries?

[@AI-HowTo]

sorry, not available at the time being.

[@alexlefkowitz]

@@AI-HowTo Any particular reason why not?

[@rarf2142]

Will I still get the same faces with a different background? I’m looking to make an AI influencer.

[@AI-HowTo]

yes to a very good level of resemblance, using this method, even if you change the background you get similar faces even if angels of the face is slightly different, (or very similar ones that it might be difficult to tell the difference) ... also if you didnt (this might differ depending on which checkpoint you use) the use of LoRA mixing at some ratios will give the same face effect inside after detailer, see here Elon (LoRA) for instance is using different clothes/backgrounds, the girl also is displayed in different clothes but its almost the same face.

[@HPCAT88]

thanks. now let's scam some simps on OF

[@the_one_and_carpool]

hads to dislike because you used elon muskrat

[@AI-HowTo]

funny :), yes, I thought some will do so but put him anyway, I totally get it :)

[@user-jk9zr3sc5h]

All of the faces are a similar angle, is it possible to do left/right facing for a profile shot?

[@AI-HowTo]

correct, most are, i think this is because of the check point and the prompt, check kzbin.info/www/bejne/d6LRe5h9hamaqbs for instance, the second picture it is not same angel, after detailer can inpaint with different angels and preserve the same features, different prompting will produce different face angels ... one can utilize a LoRA for a certain character inside after detailer for a certain subject which is what I often do, a LoRA that is trained on left/right/front/up/down view of the face produces great results from all angels, even in full body shots and complex scenes.