ALL PROMPTS FOR THIS VIDEO Also make sure you have use secondary models unchecked, that is something else I did that I forgot to mention, it can cause longer render times (and may use more memory,unsure of that) but it usually provides better results and especially with some of the non default models, some of them seem to have a real hard time when use secondary models is checked. (please do not reply to this,this is pinned and has the prompts that go with the video, but feel free to make a new comment!) IMAGE 1 Prompt- 0: [" a 3d portrait of a female by Bouguereau William-Adolphe:1.5", "artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.7",], --------------------------------------------------------------------------------------------------------------- IMAGE 2 Prompt- 0: [" a portrait of a female elf sorceress by karol bak and Jia Ruan", "artstation,deviantart,fantasy art,rpg portrait:.4",], --------------------------------------------- IMAGE 3 Prompt-
0: [" a portrait of a female by pablo picasso", "blue period,:.3",], ----------------------------------------------------- IMAGE 4 Prompt- 0:[" A painting by Karol Bak",], ------------------------------------------------------ IMAGE 5 Prompt- 0: [" a 3d portrait of a male pirate by Rembrandt and Bouguereau William-Adolphe", "artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.3",], ------------------------------------------------------------------------------------------- IMAGE 6 Prompt- 0: [" a 3d portrait of a female vampire by Kago Shintaro and gerald brom and Horley Alex ", "legs,hands,teeth:-1", "artstation,deviantart,vray render,unreal engine,hyperrealism,volumetric lighting:.3",],
Dude, I just subscribed to you, Your willingness to share your knowledge without any motive impressed me. And the information you shared was very crystal clear. Thank you, Prophet.
@@prophetofthesingularity Definitely an improvement by turning off the secondary model. Though there is still some issues with deformation. thanks for the tip anyway.
["a 3d portrait of a elvira mistress of the dark by Kago Shintaro and gerald brom and Horley Alex ", "legs,hands,teeth:-1", "artstation,deviantart,vray render,unreal engine,hyperrealism,volumetric lighting:.3"], came out really cool
Hello! Awesome video! Thank you :) Please, I need help: i’m using any setting, including yours, no face comes out... scattered eyes, a chin, an ear, the outline of a half face, nothing more. I scrupulously followed the steps but no portrait comes out! Can you help me, please? :(
Make sure you have Use secondary model unchecked, it is turned on by default. It will not work well if that is turned on due to the way the model is set up. This is probably the issue
Would like to see a more in depth guide on how to do weighting with numbers. Lips, nose,: .-4 etc. How do you set it up exactly and how is each number weighted
Sure, so the weights are basically matched against the other prompts that you are using, so if you just have 1 prompt and you give it a weight of 2 and another prompt with no weight (which defaults to 1) the one with the weight of 2 will be used 2x more then the other prompt, I do not really use it with any specific formula, if I think one of the lines of my prompt is contributing too much to the image then I will turn it down, which usually is just giving it a decimal point like a .6 or something, by default all the prompts have a weight of 1 unless you put a weight at the end of it. So what I usually do is have my main prompt at the very top, the prompt that describes the scene and what is in it and artist modifiers, then I will put modifiers below that, usually starting with a 3d modifier, like unreal engine, volumetric light,etc. If I like the 3d effect I will keep it or turn it up or down, sometimes the 3d effect looks horrible on the prompt I used so I might turn it down to a .3 or just get rid of it entirely. So the weights are basically just a way to balance out elements in your prompt and if you split your prompt up into different lines than you can have more micromanagement over each element. But I will definitely make a tutorial on that, this is a good idea, weights are very important also with animations because you can use them to change the scene entirely, like I have done a few where I will have Summer Flowers:1, Winter Snow:0 and then reverse it at keyframe 50 or 100 to Summer Flowers:0 (or -2, negative numbers will actually remove elements) snow:1 Another thing if your prompts equal out to Zero in a line it will cause an error and will not work, so you have to have it be either a negative or a positive number when it adds up the weights.
honestly, the best way to learn I've found is just by trying various settings and tweaking from there. I also wish there was like a "Dummies" cheatsheet for common GAN terms ie "cut number", "batch number", "ViTB32" (and other models, FOV, Rand_Mag, Near plane, etc...)most of the vernacular is the same for various GAN's but if it's not, that could be denoted or excluded).
Also, if you do not have this, I just linked this in the about section of the video, Zippys Disco Diffusion Cheat sheet is a great resource Here is a great little part on weights, and I have used this to turn Summer into Winter in animations, I am currently doing a similar one now, but weights basically, let you decide how much that prompt is going to be in the image, a negative weight will let you take it away from the image so in this example from Zippys 0: ["rocky beach:2", "sky:-1"] will push the image towards a rocky beach while attenuating sky details. Important: the sum of the weights must not equal 0. Or for an animation you could do something like this 0: ["Apple:2", "Orange:-1"], 50:["Apple:-2", "Orange:3"], So, when it first makes the image it will make an apple, when it hits keyframe 50 it will stop making the apple and subtract the apple from the image and turn it into an orange. You can use weights in still images to just make more or less of one element of the image, like if you want more apples than oranges, you could do 0:["Apple:3", "Orange:1"] and that will make 3 times as many apples as oranges, that is the basic explanation of weights. So, they are really for either fine tuning the elements in an image or changing the image in an animation, which is what I usually used them for in the beginning and then after a while I started using them in still images as well for fine tuning modifiers or prompts. docs.google.com/document/d/1l8s7uS2dGqjztYSjPpzlmXLjl5PM3IGkRWI3IiCuK7g/edit
I think you wanted to use vertical symmetry instead of the horizontal one. Usually, horizontal symmetry is a symmetry between the upper part and the bottom part of the image (around the horizontal axis). A face is vertically symmetric (left-right). Sorry if I missunderstood your point, I’m not a native english speaker and a total beginner with Disco Diffusion.
@@mimique_key Yes, the horizontal mirrors the left and right side, the vertical the top/bottom. Ty for visiting my page! I appreciate people from all over the world visiting! I hope the captions are working ok! Another thing I have done is to use a prompt for symmetrical and you can use a weight also to just make it mostly symmetrical like this "symmetrical:.4" or you can also just run it without worrying about symmetry and doing a few batches and you will get some that come out looking good, usually without symmetry one the main things you get is crosseyed people,heh. Even Dall-E does this. (here is a funny video someone posted before that shows that) vimeo.com/727017045?fbclid=IwAR22WjQzdcd4Qv6pzvszqgh-DaRn-ZfRO-1ZgZ-bgFrNHXKWEAe1Feam9G0 One other thing did a lot before the portrait generator is just post edit and mirror the face in the middle or fix the eyeballs.
Have you found much difference/benefit in using cutn_batches: 1 vs 4 or 8? I've been running on 4 lately just as a personal default because someone mentioned it to me a while back but I haven't really experimented with doing it any other way.
Yes, this is a setting that can affect render time quite a bit! You can do a run and compare how long it takes to make each image at 1 vs 4 or 8. I have usually left mine at 2, I used to use it at 3- 4 but after a while I felt like it was not worth the extra render time it takes and started using 2 most of the time. For this run I actually just have it at 1, and lately I have been turning it down along with adjusting the cut_n scheduling, which is something that was always a bit of a black box to me so now I am starting to understand it's effects better and will post a video soon with some examples. For now what I am seeing from the results is that I get more coherent images by turning these down and to get more detail I have been compensating by adding more steps at times. Also I seem to get a lot less of double horizons and multiple objects, things like that,by turning it low and also keeping the clip guidance low but I am still experimenting with it. Should have a video up soon but 4 seems like it adds a lot of render time to me so I quit using it that high and now I am contemplating using just 1 more often.
Do you have the secondary model unchecked? That is the most common issue that I have seen. Make sure it is unchecked. The prompts I give should give you something similar to the images I have depending on the settings, but the default settings are usually ok except for the secondary model, that is checked by default I think.
any tips on using the portrait model for animations? I'm getting great results in the first few frames, but as the camera moves in any direction it fails to create new detail. It doesn't do this when i have the default model with the same settings
Yeah, I have one that I have been working on where I have no camera movement,but the keyframe just changes the prompt, so I have kind of a time lapse of a young person growing into and old person which is what I did in one of my tutorials with the seasons of the year, I had a castle change from winter to summer. I have not tried an animation with movement, since it is very specific to just making portraits, I do not think it would be the best model for animations with movements.But would be good for having someone change from a Pirate to an Alien or that kind of thing. I will post my portrait animation sometime, going to post the dailies today maybe I will put it at the end of that.
@@prophetofthesingularity good to know. I managed to get some cool results by rendering a nice portrait frame and using it as the INIT image for an animation with the default model. It doesn’t maintain a clean portrait, but it has an effect that works for what I’m trying to do. Thanks!!
Yeah it got changed so it has to look like the other fields now you cannot just put in the number by itself It has to be in this format [1]*1000 Put whatever ic_pow you want where the 1 is
Are you using the same models also? I have the use secondary model turned off, I know this makes it come out better but also it renders slower so I think it takes more memory but not 100% on that, I will check that out later,but that is a setting that I changed also, aside from the models and the use secondary model, the other main thing I changed is resolution make sure your resolution is set to where you have it taller than wide, or somewhat of a portrait scale.
@@prophetofthesingularity I fixed it, the secondary model was turned on. Missed that when I first watched your video. Anyway, thanks for the reply. Have had some good results.
They are in the pinned comment, hit read more. I can just copy and paste it here too :) IMAGE 1 Prompt- 0: [" a 3d portrait of a female by Bouguereau William-Adolphe:1.5", "artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.7",], --------------------------------------------------------------------------------------------------------------- IMAGE 2 Prompt- 0: [" a portrait of a female elf sorceress by karol bak and Jia Ruan", "artstation,deviantart,fantasy art,rpg portrait:.4",], --------------------------------------------- IMAGE 3 Prompt-
0: [" a portrait of a female by pablo picasso", "blue period,:.3",], ----------------------------------------------------- IMAGE 4 Prompt- 0:[" A painting by Karol Bak",], ------------------------------------------------------ IMAGE 5 Prompt- 0: [" a 3d portrait of a male pirate by Rembrandt and Bouguereau William-Adolphe", "artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.3",], ------------------------------------------------------------------------------------------- IMAGE 6 Prompt- 0: [" a 3d portrait of a female vampire by Kago Shintaro and gerald brom and Horley Alex ", "legs,hands,teeth:-1", "artstation,deviantart,vray render,unreal engine,hyperrealism,volumetric lighting:.
As you note, a lot of very interesting images. However, and like yours, my results have flaws. In particular, eyes don’t match well even with symmetry enabled. I am taking my images into photoshop with a view to correcting the anomalies. Incidentally, the model only does portraits of people. A portrait of an elephant was rendered as the Elephant Man!
Heh, Yes I still get anomalies sometimes as well, I also use layers and post edit sometime to correct, sometimes this line can help but it does take away a bit of the artist style "Zbrush,nose,eyes,face, mouth:.4", this will basically have it copy a zbrush style 3d render of the features and can help reduce anomalies, but it also does take away a bit of the artists style, I use the weight and turn it up or down just depending, I try to keep it as low as possible if I am using it.
Thanks for the tips. I will incorporate them in my rendering. With regard to your earlier suggestion of activating horizontal symmetry, a better solution is inserting “ symmetrical and photo- realistic” within the prompt. Distortions are minimised or eliminated and the artistic content is not seriously diminished. Regards, Peter U
First of all thanks for your efforts explain what is going in DD. Portraits for women or young men is like OK, but for a man with beard and/or mustache it's horrible. I tried use my photo as init image and for me it doesn't work. It takes wrinkles, unshaving, cap, age but it's not "autoportrait". Thanks to G-d that I didn't tried to use photo of my wife for experiments. BR
When you use a init image it can blur it quite a bit, you could try to skip more steps to preserve the image better, but it does blur out small details, the finer the detail the more steps you would have to skip to preserve it.
No that one I do not know it is usually a memory error if it crashes but that one does not look like it, if a runtime restart does not work maybe just copy the default notebook again.
ALL PROMPTS FOR THIS VIDEO
Also make sure you have use secondary models unchecked, that is something else I did
that I forgot to mention, it can cause longer render times (and may use more memory,unsure of that)
but it usually provides better results and especially with some of the non default models, some
of them seem to have a real hard time when use secondary models is checked.
(please do not reply to this,this is pinned and has the prompts
that go with the video, but feel free to make a new comment!)
IMAGE 1 Prompt-
0: [" a 3d portrait of a female by Bouguereau William-Adolphe:1.5",
"artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.7",],
---------------------------------------------------------------------------------------------------------------
IMAGE 2 Prompt-
0: [" a portrait of a female elf sorceress by karol bak and Jia Ruan",
"artstation,deviantart,fantasy art,rpg portrait:.4",],
---------------------------------------------
IMAGE 3 Prompt-
0: [" a portrait of a female by pablo picasso",
"blue period,:.3",],
-----------------------------------------------------
IMAGE 4 Prompt-
0:[" A painting by Karol Bak",],
------------------------------------------------------
IMAGE 5 Prompt-
0: [" a 3d portrait of a male pirate by Rembrandt and Bouguereau William-Adolphe",
"artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.3",],
-------------------------------------------------------------------------------------------
IMAGE 6 Prompt-
0: [" a 3d portrait of a female vampire by Kago Shintaro and gerald brom and Horley Alex ",
"legs,hands,teeth:-1",
"artstation,deviantart,vray render,unreal engine,hyperrealism,volumetric lighting:.3",],
Dude, I just subscribed to you, Your willingness to share your knowledge without any motive impressed me. And the information you shared was very crystal clear. Thank you, Prophet.
Ty! Glad to hear! I am going to do some portraits for the weekly images this week I think!
Thank you again for another awesome vid
Beautiful! The biggest weakness of DD is fixed now, at least for close-ups. Glad I slacked off on getting 5.5 installed.. thanks for the demo!
I'd say it's still a fair way off. Unfortunately most of the results still look incredibly mutated and deformed.. from my experience anyway.
@@daveread4969 Try turning off secondary model and check your resolution make sure it is set to a portrait scale (taller than wide)
@@prophetofthesingularity Definitely an improvement by turning off the secondary model. Though there is still some issues with deformation. thanks for the tip anyway.
This was most helpful, thank you. Well presented and your voice is excellent.
Ty! My own voice sounds weird to me like everyone I guess :)
awesome and nice work
Really nice guide, I love the results you have.
["a 3d portrait of a elvira mistress of the dark by Kago Shintaro and gerald brom and Horley Alex ",
"legs,hands,teeth:-1",
"artstation,deviantart,vray render,unreal engine,hyperrealism,volumetric lighting:.3"],
came out really cool
Hello! Awesome video! Thank you :)
Please, I need help: i’m using any setting, including yours, no face comes out... scattered eyes, a chin, an ear, the outline of a half face, nothing more. I scrupulously followed the steps but no portrait comes out! Can you help me, please? :(
Make sure you have Use secondary model unchecked, it is turned on by default. It will not work well if that is turned on due to the way the model is set up. This is probably the issue
Would like to see a more in depth guide on how to do weighting with numbers. Lips, nose,: .-4 etc. How do you set it up exactly and how is each number weighted
Sure, so the weights are basically matched against the other prompts that you are using, so if you just have 1 prompt and you give it a weight of 2 and another prompt with no weight (which defaults to 1) the one with the weight of 2 will be used 2x more then the other prompt, I do not really use it with any specific formula, if I think one of the lines of my prompt is contributing too much to the image then I will turn it down, which usually is just giving it a decimal point like a .6 or something, by default all the prompts have a weight of 1 unless you put a weight at the end of it. So what I usually do is have my main prompt at the very top, the prompt that describes the scene and what is in it and artist modifiers, then I will put modifiers below that, usually starting with a 3d modifier, like unreal engine, volumetric light,etc. If I like the 3d effect I will keep it or turn it up or down, sometimes the 3d effect looks horrible on the prompt I used so I might turn it down to a .3 or just get rid of it entirely. So the weights are basically just a way to balance out elements in your prompt and if you split your prompt up into
different lines than you can have more micromanagement over each element. But I will definitely make a tutorial on that, this is a good idea, weights are very important also with animations because you can use them to change the scene entirely, like I have done a few where I will have
Summer Flowers:1, Winter Snow:0 and then reverse it at keyframe 50 or 100 to Summer Flowers:0 (or -2, negative numbers will actually remove elements) snow:1
Another thing if your prompts equal out to Zero in a line it will cause an error and will not work, so you have to have it be either a negative or a positive number when it adds up the weights.
honestly, the best way to learn I've found is just by trying various settings and tweaking from there. I also wish there was like a "Dummies" cheatsheet for common GAN terms ie "cut number", "batch number", "ViTB32" (and other models, FOV, Rand_Mag, Near plane, etc...)most of the vernacular is the same for various GAN's but if it's not, that could be denoted or excluded).
Also, if you do not have this, I just linked this in the about section of the video, Zippys Disco Diffusion Cheat sheet is a great resource
Here is a great little part on weights, and I have used this to turn Summer into Winter in animations, I am currently doing a similar one now, but weights
basically, let you decide how much that prompt is going to be in the image, a negative weight will let you take it away from the image so in this example from Zippys
0: ["rocky beach:2", "sky:-1"] will push the image towards a rocky beach while attenuating sky details. Important: the sum of the weights must not equal 0.
Or for an animation you could do something like this
0: ["Apple:2", "Orange:-1"],
50:["Apple:-2", "Orange:3"],
So, when it first makes the image it will make an apple, when it hits keyframe 50 it will stop making the apple and subtract the apple from the image and turn it into an orange.
You can use weights in still images to just make more or less of one element of the image, like if you want more apples than oranges, you could do
0:["Apple:3", "Orange:1"] and that will make 3 times as many apples as oranges, that is the basic explanation of weights. So, they are really
for either fine tuning the elements in an image or changing the image in an animation, which is what I usually used them for in the beginning and then
after a while I started using them in still images as well for fine tuning modifiers or prompts.
docs.google.com/document/d/1l8s7uS2dGqjztYSjPpzlmXLjl5PM3IGkRWI3IiCuK7g/edit
I think you wanted to use vertical symmetry instead of the horizontal one. Usually, horizontal symmetry is a symmetry between the upper part and the bottom part of the image (around the horizontal axis). A face is vertically symmetric (left-right).
Sorry if I missunderstood your point, I’m not a native english speaker and a total beginner with Disco Diffusion.
Ok my bad, after some test, it clearly doesn’t work in disco diffusion like I thought.
@@mimique_key Yes, the horizontal mirrors the left and right side, the vertical the top/bottom.
Ty for visiting my page! I appreciate people from all over the world visiting! I hope the captions are working ok!
Another thing I have done is to use a prompt for symmetrical and you can use a weight also to just make it mostly symmetrical
like this
"symmetrical:.4"
or you can also just run it without worrying about symmetry and doing a few batches and you will get some that come out looking good, usually without symmetry one the main things you get is crosseyed people,heh. Even Dall-E does this.
(here is a funny video someone posted before that shows that)
vimeo.com/727017045?fbclid=IwAR22WjQzdcd4Qv6pzvszqgh-DaRn-ZfRO-1ZgZ-bgFrNHXKWEAe1Feam9G0
One other thing did a lot before the portrait generator is just post edit and mirror the face in the middle or fix the eyeballs.
Thank you
Have you found much difference/benefit in using cutn_batches: 1 vs 4 or 8? I've been running on 4 lately just as a personal default because someone mentioned it to me a while back but I haven't really experimented with doing it any other way.
Yes, this is a setting that can affect render time quite a bit! You can do a run and compare how long it takes to make each image at 1 vs 4 or 8.
I have usually left mine at 2, I used to use it at 3- 4 but after a while I felt like it was not worth the extra render time it takes and started using 2 most of the time. For this run I actually just have it at 1, and lately I have been turning it down along with adjusting the cut_n scheduling, which is something that was always a bit of a black box to me so now I am starting to understand it's effects better and will post a video soon with some examples. For now what I am seeing from the results is that I get more coherent images by turning these down and to get more detail I have been compensating by adding more steps at times. Also I seem to get a lot less of double horizons and multiple objects, things like that,by turning it low and also keeping the clip guidance low but I am still experimenting with it.
Should have a video up soon but 4 seems like it adds a lot of render time to me so I quit using it that high and now I am contemplating using just 1 more often.
Haven't had much luck with good looking portraits on DD.
Do you have the secondary model unchecked? That is the most common issue that I have seen. Make sure it is unchecked.
The prompts I give should give you something similar to the images I have depending on the settings, but the default settings are usually ok except for
the secondary model, that is checked by default I think.
@@prophetofthesingularity Thank you!!! Will check it and try it out!
any tips on using the portrait model for animations? I'm getting great results in the first few frames, but as the camera moves in any direction it fails to create new detail. It doesn't do this when i have the default model with the same settings
Yeah, I have one that I have been working on where I have no camera movement,but the keyframe just changes the prompt, so I have kind of a time lapse of a young person growing into and old person which is what I did in one of my tutorials with the seasons of the year, I had a castle change from winter to summer. I have not tried an animation with movement, since it is very specific to just making portraits, I do not think it would be the best model for animations with movements.But would be good for having someone change from a Pirate to an Alien or that kind of thing. I will post my portrait animation sometime, going to post the dailies today maybe I will put it at the end of that.
@@prophetofthesingularity good to know. I managed to get some cool results by rendering a nice portrait frame and using it as the INIT image for an animation with the default model. It doesn’t maintain a clean portrait, but it has an effect that works for what I’m trying to do. Thanks!!
How do you change the Cut_Ic_Pow Settings in 5.6? Whenever I alter the value it wont do the Run.
Yeah it got changed so it has to look like the other fields now
you cannot just put in the number by itself
It has to be in this format
[1]*1000
Put whatever ic_pow you want where the 1 is
Mine are coming out like a monster from The Thing? Followed your settings?
Are you using the same models also? I have the use secondary model turned off, I know this makes it come out better but also it renders slower so I think it takes more memory but not 100% on that, I will check that out later,but that is a setting that I changed also, aside from the models and the use secondary model, the other main thing I changed is resolution make sure your resolution is set to where you have it taller than wide, or somewhat of a portrait scale.
@@prophetofthesingularity I fixed it, the secondary model was turned on. Missed that when I first watched your video. Anyway, thanks for the reply. Have had some good results.
@@MONGIE30 Great to hear! I always turn that off when I can, it does slow things down a bit but can give better results.
Where are the prompts you said would be posted here?
They are in the pinned comment, hit read more.
I can just copy and paste it here too :)
IMAGE 1 Prompt-
0: [" a 3d portrait of a female by Bouguereau William-Adolphe:1.5",
"artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.7",],
---------------------------------------------------------------------------------------------------------------
IMAGE 2 Prompt-
0: [" a portrait of a female elf sorceress by karol bak and Jia Ruan",
"artstation,deviantart,fantasy art,rpg portrait:.4",],
---------------------------------------------
IMAGE 3 Prompt-
0: [" a portrait of a female by pablo picasso",
"blue period,:.3",],
-----------------------------------------------------
IMAGE 4 Prompt-
0:[" A painting by Karol Bak",],
------------------------------------------------------
IMAGE 5 Prompt-
0: [" a 3d portrait of a male pirate by Rembrandt and Bouguereau William-Adolphe",
"artstation,deviantart,vray render,unreal engine,hyperrealism,photorealism,volumetric lighting:.3",],
-------------------------------------------------------------------------------------------
IMAGE 6 Prompt-
0: [" a 3d portrait of a female vampire by Kago Shintaro and gerald brom and Horley Alex ",
"legs,hands,teeth:-1",
"artstation,deviantart,vray render,unreal engine,hyperrealism,volumetric lighting:.
As you note, a lot of very interesting images. However, and like yours, my results have flaws. In particular, eyes don’t match well even with symmetry enabled. I am taking my images into photoshop with a view to correcting the anomalies. Incidentally, the model only does portraits of people. A portrait of an elephant was rendered as the Elephant Man!
Heh, Yes I still get anomalies sometimes as well, I also use layers and post edit sometime to correct, sometimes this line can help but it does take
away a bit of the artist style
"Zbrush,nose,eyes,face, mouth:.4",
this will basically have it copy a zbrush style 3d render of the features and can help reduce anomalies, but it also does take away a bit of the artists style, I use the weight and turn it up or down just depending, I try to keep it as low as possible if I am using it.
Thanks for the tips. I will incorporate them in my rendering. With regard to your earlier suggestion of activating horizontal symmetry, a better solution is inserting “ symmetrical and photo- realistic” within the prompt. Distortions are minimised or eliminated and the artistic content is not seriously diminished. Regards, Peter U
Did you mean to say you didnt run the custom model? without secondary I cant run it.
To use the portrait generator do not have use secondary model checked, and use the portrait diffusion model,
@@prophetofthesingularity thank you ill start over and try again
Note your winner winner chicken dinner prompts in the colab notebook by commenting them with #
First of all thanks for your efforts explain what is going in DD. Portraits for women or young men is like OK, but for a man with beard and/or mustache it's horrible. I tried use my photo as init image and for me it doesn't work. It takes wrinkles, unshaving, cap, age but it's not "autoportrait". Thanks to G-d that I didn't tried to use photo of my wife for experiments. BR
When you use a init image it can blur it quite a bit, you could try to skip more steps to preserve the image better, but it does blur out small details, the finer the detail the more steps you would have to skip to preserve it.
@@prophetofthesingularity Tnx for advice. I will try. What about "beard and/or mustache"?
it doesn't work for me, any idea ?
(it say this error)
Starting Run: TimeToDisco(113) at frame 0
Prepping model...
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
in ()
205 model.load_state_dict(torch.load(custom_path, map_location='cpu'))
206 else:
--> 207 model.load_state_dict(torch.load(f'{model_path}/{get_model_filename(diffusion_model)}', map_location='cpu'))
208 model.requires_grad_(False).eval().to(device)
209 for name, param in model.named_parameters():
1 frames
/usr/local/lib/python3.7/dist-packages/torch/serialization.py in __init__(self, name_or_buffer)
240 class _open_zipfile_reader(_opener):
241 def __init__(self, name_or_buffer) -> None:
--> 242 super(_open_zipfile_reader, self).__init__(torch._C.PyTorchFileReader(name_or_buffer))
243
244
RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory
No that one I do not know it is usually a memory error if it crashes but that one does not look like it, if a runtime restart does not work maybe just copy the default notebook again.
@@prophetofthesingularity well, I have delete some .pt stuff in my drive, and now it's working ! thx u