You can download the cartoon model from here civitai.com/models/297501?modelVersionId=357959 You can try other cartoon models, but this one worked ok for me, juggernautxl v9 kind of works but not all the time You can add after the subject in the prompt professional 3D design of a cute and adorable cartoon character with big eyes, pixar art style or professional 3D design of a cartoon character, pixar art style or mention the type of the object like: portrait of a monster, white and red, creature, professional 3d design of a cartoon creature, pixar art style for the negative you can find it also in the model page is: (worst quality, low quality, normal quality, lowres, low details, oversaturated, undersaturated, overexposed, underexposed, grayscale, bw, bad photo, bad photography, bad art:1.4), (watermark, signature, text font, username, error, logo, words, letters, digits, autograph, trademark, name:1.2)
Man, all your tutorials are awesome in terms of comprehensive explanation, brevity, and without disturbing the follower with nonsense. Thank you for sharing your knowledge with us.🙏🏻
Absolutely mind-blown by the possibilities this opens up for content creation! Your step-by-step guide makes it seem so easy and the results are just stunning. Can't wait to try this out myself, thank you!
some great tips, thanks. This in combination with Instant-ID in Forge seems to work well. Using InsightFace(InstantID) as controlnet preprocessor and ip-adapter-instant-id as cnet model. Seems to allow for a bit more freedom with denoising strength while still keeping face features.
Can you tell me what video in the playlist I can look at to learn how to replace a background. I generated an image but want to change the background, I’m having trouble getting segmentation to work, as I don’t know how to make a mask and use it to change the backgrounds, is there a proper technique?
I didn't have time yet to do research on that, i saw people using it but i didn't need because use Photoshop to quickly remove backgrounds and do mask how i want
Great tutorial! However, I am getting an error that reads. NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. Time taken: 1 min. 45.5 sec.
Did you tried those recommend settings that error gives? Also froge got different updates, the last version isn't the stable version, might want to switch to different fork or to a stable version. I used one version that start with 29
I think is not about which interface you use but about the checkpoint model, some models are trained with more cartoon images or in a cartoon style so it is easier to get it in that style, so any ui you use should work if you got a good model
@@pixaroma Thanks for the response. Do you have any suggestions for improving eyes and hands? When I try your method in the video, the eyes and hands come out weird most of the time.
Depends on the model, some have better eyes and hand then others but in general AI is still struggles sometimes with that, also cartoon eyes are all kind of categories so that might also affect, and with hand is a general problem on all AI because you can do so many combinations of poses with hands so it still do mistakes
Can you tell me how your computer is configured to be running SD?Because I see the rendering speed is very fast, I want to buy a computer to do the work you are doing! thanks
For the speed i get like 5 second for a 1024px image, i often speed up the video to not wait. I have rtx 4090 video card,with 24gb of vram. So if you want to run stable diffusion i recommend an Nvidia card with at least 8gb of VRAM, more VRAM the faster you will generate. For the rest of the system is not so important i think normal ram also help a little like 8GB of ram or 16. My pc was too expensive but i need it for 3d renders and stuff so probably you can go with not so expensive pc. My config - CPU Intel Core i9-13900KF (3.0GHz, 36MB, LGA1700) box - GPU GIGABYTE AORUS GeForce RTX 4090 MASTER 24GB GDDR6X 384-bit- Motherboard GIGABYTE Z790 UD LGA 1700 Intel Socket LGA 1700 - 128 GB RAM Corsair Vengeance, DIMM, DDR5, 64GB (4x32gb), CL40, 5200Mhz- SSD Samsung 980 PRO, 2TB, M.2 - SSD WD Blue, 2TB, M2 2280- Case ASUS TUF Gaming GT501 White Edition, Mid-Tower, White- Cooler Procesor Corsair iCUE H150i ELITE CAPELLIX Liquid- PSU Gigabyte AORUS P1200W 80+ PLATINUM MODULAR, 1200W
@@hahuyson I am not at pc to test it, but I sdxl does better images in 1024px and only v1.5 does better on 512px so since the sdxl models appeared I didnt generate lower then 1024px anymore, but probably way faster, but also depends on models and numbers of steps, with some models like lightning and hyper you can generate way faster
tested this morning on sdxl model juggernaut X with 20 sampling steps: on 512x512pxTime taken: 1.4 sec. | For 768x768px Time taken: 2.0 sec. | 1024x1024px Time taken: 3.6 sec.
Hi, I'm facing some problems in generating a image, basically, at the beginning, it's ok, but in the end, the image is pixelated, blur, a terrible image .
Let me ask you something, is there a way of transforming a photo of a dog or a person, and changing the pose, and the environment, without changing the personnel characteristics? For example, my dog died in February, and I'd like to see a photo-cartoon of him running in the sky with clouds...
I will check it out, i was assuming it is transferring style and here it just switches from photo to cartoon, plus is faster, and the control net still has errors on image size that is not divisible with 64.
Until I make a new video for forge you can try this one th-cam.com/video/IBNuALJuOgw/w-d-xo.html Use Juggernaut XL v9 instead and for control net model you can find here, try canny huggingface.co/lllyasviel/sd_control_collection/tree/main just make sure you are using image size with width and height that is divisible with 64, forge has a bug that gives an error if is not in those sizes, so test with 1024x1024px first to see if works
I didn't try in the new version they changed a lot of things since i did the video maybe they changed some settings, depending on the image and the prompt also
most of the nonetype error can be fixed by Enable Pad prompt/negative prompt in Settings -> Optimizations or to Set the Width and Height to be multiple of 64
probably something changed in the interface since I did the video, not sure, if is the same model same settings, must be the new interface maybe they updated. I mean you see how I did it in the video..
I usually got that when i used a width and height that was not divisible with 64, so try 1024*1024 to make sure is not because of that. Then it can be from other things also since is a general error that is caused by other errors, sometimes you can see in command Window more details
I was using the forge with commit that start with 29, maybe the updated version doesn't work anymore. I have been switching to comfy UI to get latest updates faster and am trying to recreate all the workflows from forge to comfyui
You can download the cartoon model from here
civitai.com/models/297501?modelVersionId=357959
You can try other cartoon models, but this one worked ok for me, juggernautxl v9 kind of works but not all the time
You can add after the subject in the prompt
professional 3D design of a cute and adorable cartoon character with big eyes, pixar art style
or
professional 3D design of a cartoon character, pixar art style
or mention the type of the object like: portrait of a monster, white and red, creature, professional 3d design of a cartoon creature, pixar art style
for the negative you can find it also in the model page is:
(worst quality, low quality, normal quality, lowres, low details, oversaturated, undersaturated, overexposed, underexposed, grayscale, bw, bad photo, bad photography, bad art:1.4), (watermark, signature, text font, username, error, logo, words, letters, digits, autograph, trademark, name:1.2)
Awesome, cheers my friend.
Man, all your tutorials are awesome in terms of comprehensive explanation, brevity, and without disturbing the follower with nonsense. Thank you for sharing your knowledge with us.🙏🏻
Absolutely mind-blown by the possibilities this opens up for content creation! Your step-by-step guide makes it seem so easy and the results are just stunning. Can't wait to try this out myself, thank you!
Very informative and simple to fallow, thank you.
this is what I was looking for! Perfect
This worked perfectly. Thank you!
some great tips, thanks. This in combination with Instant-ID in Forge seems to work well. Using InsightFace(InstantID) as controlnet preprocessor and ip-adapter-instant-id as cnet model. Seems to allow for a bit more freedom with denoising strength while still keeping face features.
Nice tutorial, dude! Thank you so much!
Can you tell me what video in the playlist I can look at to learn how to replace a background. I generated an image but want to change the background, I’m having trouble getting segmentation to work, as I don’t know how to make a mask and use it to change the backgrounds, is there a proper technique?
I didn't have time yet to do research on that, i saw people using it but i didn't need because use Photoshop to quickly remove backgrounds and do mask how i want
@ I got it to work, but you are right, the cuts aren’t as detailed and clean, maybe I’m missing something. But, yup, photoshop is the way to go 👍
Great tutorial! However, I am getting an error that reads.
NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 1 min. 45.5 sec.
Did you tried those recommend settings that error gives? Also froge got different updates, the last version isn't the stable version, might want to switch to different fork or to a stable version. I used one version that start with 29
@@pixaroma Thank you for your response. I've since upgraded my GPU and am not experiencing the issue anymore.
Is there a way to have the image take on the style of a LoRA?
I didn't try it on forge but i did something similar on comfyui with lora
Just found your youtube channel, great video! Will def check out the rest..
How can I get this style like I see that each photo you used has worked
Use a stable diffusion model or a lora model that knows cartoon or is trained on cartoon, check episode 18 also to see how i did in comfyui.
@ I’m watching it but I have one question does it work for more than one person in a picture? Like for example a family portrait or a couple portrait
Only if you try, theoretically yes but us hard to keep the face consistent for the person, it will be different, you can describe each person maybe
@@pixaroma I keep getting this message each time I queue ( AIO_PREPROCESSOR ) I don’t get it it’s downloaded why I keep getting that
that was great! I need a turorial for cartoon to realistic photo :) I tried the same the other way around but somehow it always looks trash
thanks
Which one do you suggest for turning real photos into Disney/Pixar style: Fooocus, Forge, or Automatic1111?
I think is not about which interface you use but about the checkpoint model, some models are trained with more cartoon images or in a cartoon style so it is easier to get it in that style, so any ui you use should work if you got a good model
@@pixaroma Thanks for the response. Do you have any suggestions for improving eyes and hands? When I try your method in the video, the eyes and hands come out weird most of the time.
Depends on the model, some have better eyes and hand then others but in general AI is still struggles sometimes with that, also cartoon eyes are all kind of categories so that might also affect, and with hand is a general problem on all AI because you can do so many combinations of poses with hands so it still do mistakes
Can you tell me how your computer is configured to be running SD?Because I see the rendering speed is very fast, I want to buy a computer to do the work you are doing! thanks
For the speed i get like 5 second for a 1024px image, i often speed up the video to not wait. I have rtx 4090 video card,with 24gb of vram. So if you want to run stable diffusion i recommend an Nvidia card with at least 8gb of VRAM, more VRAM the faster you will generate. For the rest of the system is not so important i think normal ram also help a little like 8GB of ram or 16. My pc was too expensive but i need it for 3d renders and stuff so probably you can go with not so expensive pc.
My config
- CPU Intel Core i9-13900KF (3.0GHz, 36MB, LGA1700) box - GPU GIGABYTE AORUS GeForce RTX 4090 MASTER 24GB GDDR6X 384-bit- Motherboard GIGABYTE Z790 UD LGA 1700 Intel Socket LGA 1700 - 128 GB RAM Corsair Vengeance, DIMM, DDR5, 64GB (4x32gb), CL40, 5200Mhz- SSD Samsung 980 PRO, 2TB, M.2 - SSD WD Blue, 2TB, M2 2280- Case ASUS TUF Gaming GT501 White Edition, Mid-Tower, White- Cooler Procesor Corsair iCUE H150i ELITE CAPELLIX Liquid- PSU Gigabyte AORUS P1200W 80+ PLATINUM MODULAR, 1200W
@@pixaroma If the photo is 512px, how long will it take?
@@hahuyson I am not at pc to test it, but I sdxl does better images in 1024px and only v1.5 does better on 512px so since the sdxl models appeared I didnt generate lower then 1024px anymore, but probably way faster, but also depends on models and numbers of steps, with some models like lightning and hyper you can generate way faster
@@pixaroma thank you so much!
tested this morning on sdxl model juggernaut X with 20 sampling steps: on 512x512pxTime taken: 1.4 sec. | For 768x768px Time taken: 2.0 sec. | 1024x1024px Time taken: 3.6 sec.
Hi, I'm facing some problems in generating a image, basically, at the beginning, it's ok, but in the end, the image is pixelated, blur, a terrible image .
I think it has something to do with vae, change it to automatic so it takes auto the VAE see if it helps, other things can be cfg scale too big
@@pixaroma It worked!!! Thank a lot, wonderful tutorial
Let me ask you something, is there a way of transforming a photo of a dog or a person, and changing the pose, and the environment, without changing the personnel characteristics? For example, my dog died in February, and I'd like to see a photo-cartoon of him running in the sky with clouds...
Only if you train a lora maybe with it. Or maybe you can try with in painting to leave the head and inpaint anything else. Is quite difficult
@@pixaroma Thank you, i'll try.
Why don't you just use the new IP Adapter 2 style transfer? It's much superior and consistent than that
I will check it out, i was assuming it is transferring style and here it just switches from photo to cartoon, plus is faster, and the control net still has errors on image size that is not divisible with 64.
What photo would you use as a reference image?
Depends on what image you want to make it look cartoon
@StringerBell I saw only comfy ui workflows, do you know if it works with automatic1111 or forge or where i can find the models for download?
Please guide me to create architectural images from sketches
Until I make a new video for forge you can try this one th-cam.com/video/IBNuALJuOgw/w-d-xo.html Use Juggernaut XL v9 instead and for control net model you can find here, try canny huggingface.co/lllyasviel/sd_control_collection/tree/main just make sure you are using image size with width and height that is divisible with 64, forge has a bug that gives an error if is not in those sizes, so test with 1024x1024px first to see if works
I'm using forge as wel, downloaded the model, used the same settings but my img2img pictures just don't become cartoony, it's driving me crazy
I didn't try in the new version they changed a lot of things since i did the video maybe they changed some settings, depending on the image and the prompt also
@@pixaroma I found it, in the UI settings my forge was still on flux. Once I changed that it worked! Thx for the tutorial!
Anyone knows how to fix this error? TypeError: 'NoneType' object is not iterable
are you using image width and height that is not divisible with 64? doest it work with 1024x1024px or on other size get that message?
most of the nonetype error can be fixed by Enable Pad prompt/negative prompt in Settings -> Optimizations
or to Set the Width and Height to be multiple of 64
@@pixaroma Thanks, i will try.
What kind of software to use the download file? Im so sorry, im newbie in Ai
Check this video in how to install th-cam.com/video/BFSDsMz_uE0/w-d-xo.html
Im making the exact step by step but my image isnt far from the cartoon as yours :(
probably something changed in the interface since I did the video, not sure, if is the same model same settings, must be the new interface maybe they updated. I mean you see how I did it in the video..
TypeError: 'NoneType' object is not iterable, any ideas?
I usually got that when i used a width and height that was not divisible with 64, so try 1024*1024 to make sure is not because of that. Then it can be from other things also since is a general error that is caused by other errors, sometimes you can see in command Window more details
@@pixaroma I was working with 1024 x 1024 along with IMG2IMG and still had it.
I was using the forge with commit that start with 29, maybe the updated version doesn't work anymore. I have been switching to comfy UI to get latest updates faster and am trying to recreate all the workflows from forge to comfyui
@@pixaroma Thank you so much for the info. Will look into it tomorrow :)