CFG is basically contrast. While I get it’s hard to produce same image at different CFG …it can be done. But impossible to reproduce same contrast. It allocates noise to solid lines - contrast - also removing that noise from the pool of changes your prompt can make (essentially managing “creativity”). Lower CFG can often help with over burned Lora merges
@@Not4Talent_AI sure! Happy to help. Yeah I often want CFG high … I go 10-15 usually HOWEVER when I use tons of LORA’s … like 5-10 of them ….. very easy to get messed up images …. I drop the CFG down to 3-6 …. And everything clears up.
Im so sick of watching videos of people telling me about "advanced" techniques, but what they really do is show what openpose controlnet is, or something basic like that. But you actually have taught me like 5 new things in only a few minutes. Thank you, and good job figuring out this stuff!
Outstanding quality for a new channel, easy to follow, straight on point and answering exactly the questions a viewer has at certain points in the video. Keep this going and you will build a successful channel! 👍
Dude, that was neat. I loved that all subject are in chapter and covered with the right amount of time. Your video is easy to use as a cheat sheet. Many tutorial youtubers should learn from that !
Thanks mate. Been trying to learn better prompting these last couple of weeks. I've watched a bunch of videos and read heaps on it. With all that research I've done I still didn't gain as much understanding as I did from this one 12 minute video. Thank you so much. I think yours will be a channel to watch.
Really happy to hear that!! I'll probably make another prompting vid soon enough with some other things that could be interesting as well as a deeper dive into concept bleeding. Thanks!
My dude, you've just made one of the best fucking videos about 'prompt engineering' and all that shit that I've ever seen. Congratulations. Please expand upon this topic with this style of video and IM SURE you'll get a big fat number of subs.
hahahaha thanks for the support!! I do have in mind expanding on this, (as I tried making a short video on it but I still need to learn how to count to 60, made it 1:34 min so it cant be 1 short). And really glad you found it good!
Mate, what a great job you did! I've watched lots of videos on this and yours is top notch! Will be recommending and keeping an eye out on anything else you do!
I've seen so many of these videos and many of them have useful information. But the creator is either going off on tangents or using unclear language, sometimes it can feel a little inaccessible. Other creators don't do a very good job of visualizing what they are trying to explain. Point is there are a ton of inaccessible and undigestible prompt guides. Yours is exactly what I needed. It goes over information I already know but it's done in a way that is quick and concise, so It serves as a refresher. When I come across information that I haven't heard before, it's easy to understand. Your visuals are showing exactly the function you're trying to explain. The language you're using is extremely accessible and the way you progress through the information is very well thought out. Bravo and big thanks!
A little tip for you. Use aspect ratio helper. I often make 16:9 images and if you use ratio helper it will do the math for you when you move the sliders around, although it only works in tex2img. You can set it to 1:1, 2:3, 3:4, 4:5 or 16:9. This has saved me so much time, as it will slide both bars when you move one and I no longer have to do the math to figure out what ratio I'm using. Edit: lol, never mind seems you are using something like that, you just didn't mention it.
Yes! I actually use the one you said, but I didnt mention it as I was keeping everything as vanilla as possible xD. I do plan on making a video about the best extensions and this one is going in for sure. Not over complex, but super useful. Ty for the tip though! intention is what matters hahahah
Ooof, this was humbling. I watched it to see if I could pick up some new tricks, not very hopeful about it. At the beginning yeah, I knew everything. But then? A bunch of stuff I had no clue about! Thanks!
@@Not4Talent_AI It does! Sorry for the late reply. I think it's like the 80-20 rule. You can get to the 80% of what you really want to make and it's all fun and games in SD. But the last 20% of the way makes a lotta difference in generations.
You are really diamond person in this world bro when I start to watch your video it gave me relax. Thanks for ur hard work also I want to you keep upload more SD Videos
One of the best tutorial for Stable diffusion so far. I have worked with Stable diffusion for quite a while, and still learning. I can definitely say, this is one of the best out there.
Thank you so much!!! I would dare to say that we are all still learning😂. So much to try and to figure out. As well as so many new things every day hahahaha
The best thing in this video is the use of visual examples, most other channels sometimes explain something but dont give us the real output, and i know how hard is to make videos in such manner, so you earned a sub !!!
This is an extremely useful video, holy shit. I didn't know anything about prompt blending and it looks so useful for controlling the direction your prompt is taking. Awesome work, man
Thanks!!! Really glad it was useful! Btw if you want to look more of that topic up, the actual name is "prompt editing". I have a short (well, two of them bc I messed up the duration xD) that dives a little deeper into that. Even though it doesn't add that much to what I said on this video. And yes, I also think it is really useful specially without controlnet. I'm even thinking about dedicating a video solely to that and some other non-typical pormpting techniques like adding "BREAK" or "AND". Srry for the long comment, just really happy people are finding this helpful xD
Holy shit this was by far the best and most informative video of all the basics I have ever seen, way better than olivios tutorials. You clear up so many things I struggled with when trying to engineer the prompt until it's really what I want. Big thanks!
Earned a sub for sure. You have a very effective way of communicating and showing examples rather than just mentioning abstract concepts is extremely helpful. Plus the content itself is actually useful 😅 I had been struggling to tweak my prompts to get an idea out of my head and into an image, and was still hitting a wall after hours upon hours... I took the concepts you mentioned and finally got the result I wanted in about 20 minutes!
Your videos are so good! This one is a game-changer for me! Thanks so much! You definitely deserve more visibility for the raw quality of the videos you create!
Embarking on a journey of storytelling and video experimentation, VideoGPT seamlessly integrates into my workflow, refining my content with its subtle enhancements.
8:17 I like using prompt switching for hair colour and ethnic blending. For example '[Red|Blonde|Blonde] Hair' gives a strawberry blonde and doing something like [Swedish|Norwegian] gives a more exotic results than either word alone.
thanks for this video! it was very detailed and had lot of things I didn't know. I've been using stable diffusion since December and I really didn't know about prompt blending and other things you showed in the end. Can you create a more detailed separate video for the concepts in the end that is much slower so that more people like me can understand prompt blending? thanks a ton :) Subbed!
Yess!! Was planning on doing that. Also adding some other stuff. I still need to experiment a little more to fully be able to explain the functioning though. It will come though! Thanks for the kind comment!
Stable diffusion would overtake stuff like Dall-E/Midjourney right out of the gate if their prompting was just made easier. To this day it is hard for me to actually want to use stable over those other programs, particularly for work related stuff. It sucks having to sit there grinding through dozens of prompts in hopes it might kinda, sorta, almost spit something out tht looks slightly like the thing im looking for.
I'm not sure if I agree. when it comes to prompting, yes. Midjourney, for example, can give 1000% better images with a simpler prompt. But Stable diffusion has SO many tools to give infinite control over the result you are looking for. It takes a little more work, but it is definitely possible to have EXACTLY what you want. Midjrouney is way better at just creating really nice looking stuff out of the bat though, as well as having way better anatomy and over all composition
@@Not4Talent_AI I say this as someone who uses AI art in a fast paced work environment. The extra steps and complexity of the near paragraph long pro/negative prompts jam up the work cycle. Particularly when dealing with multiple clients who need multiple different types of art pieces incorporated into their projects. Midjourney/Dall-E are simply easier to use, with less steps required, and much less experimenting with new prompts. At the end of the day the extra features are nice but mass adoption will go to the program that strikes the balance between output and simplicity. Every single tool that has been introduced based off of the stable model needs experimenting with before one can use it with confidence. Not practice but straight up experimenting with to find its limitations. Midjourney is just a handful of buttons you can feed a single line of key words to and you get 4 specific examples with minimal input needed to choose and generate based off of any generated image in the exact same page. Sure theres a lot you cant do, but the many fields that look towards AI mass adoption already have tools in place, that are not experimental to achieve what Midjourney lacks. With those tools comes the confidence I spoke of earlier.
@@SoullessAIMusic Yeah, you are completely right on this. I still think that with experience that can change a little, but it is true that most people will go to the easiest option. Professionally I would use MJ and SD at the same time, but if what you need is speed, nothing beats midjourney atm
Pretty damn good video for a new account. It's a good overview, but I will say, if someone is just learning, there's too much information presented too fast. I've been messing around with AI art for a little over a month and already been introduced to all these concepts, so it was easy to follow. But a new person will struggle with how fast everything moves. All the information is amazing and accurate, but you might consider making a mini series explaining each subject / idea a little slower to gain more subs. Just a suggestion! Keep up the great content!
Thanks a ton, this is just what I needed!! Love all the positive comments but I also had a feeling that maybe it was too much. (Mainly with the next video). After publishing it today, I'll try to get a better balance to have good information for everyone but not so much that it becomes overwhelming. Really helpful feedback, appriciate it!!!
I prompt like that, it isn't necessary a must follow rule. Just what I found out works best in my case. Haven't done a ultra study on best prompt formats XD ty though!
Most of the basics are, references (in other pages, CivitAI is only for SD I think), AR affecting the result, the seed being super useful, the style list, and the fact that all words have a certain Concept Bleeding (major or minor depending on each model, this can change even in-between stable diffusion models). Some main differences I've seen is that, much like stable Diffusion 2.1, midjourney (v5) really likes natural language instead of tokenizing. So try typing prompts like talking to a human artist. Also midjourney is more "artistic" and I find it harder to control the result. And the final tips I give that are used to "combat" concept bleeding, can't really be used in midjourney. Even though midjourney has it's own list of cool things most people don't use, here is a page with all that information: docs.midjourney.com/docs/multi-prompts Hope it helped!
Curious about where you've seen double colons documented; seems they do something in midjourney, but aren't not documented in the official SD prompt guide or anywhere I can find. [text:num] grammar also isn't in the guide, which only describes [text] to apply a 0.9 scale factor to the default text weight and prompt blending with [text1:text2:step].
Hi! Srry I forgot to respond. Here is where I found all the "prompt editing" info. I will add this link in the description as well, thanks for reminding me! github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing
@@Not4Talent_AI Great, many thanks! I'd inadvertently been referencing an unofficial guide that's decent and with recent revisions, but evidently it's missing a few details. I think I might have looked at the wiki way back when getting started, but promptly forgot about it (no pun intended!)
@@makers_lab hahhahaha Hope this one helps then!! It has some interesting stuff in it, even though not too explained. Fortunately, what I hope to be today's (if not tomorrows) video, will cover that :3
Excelente video, saludos desde Perú, gracias por toda la información, llevaba mucho sin comprender varias de las cosas que mencionas, aunque aún me queda la duda sobre para que sirven poner "enter" en los prompt generando dos oraciones distintas.
@@Not4Talent_AI Oh lo siento si no me expliqué bien, me refiero a que he visto algunos prompts en los que hay dos lineas separadas (una arriba de la otra)
@@ososimiospub6406 ahhh Vale, mala mia, srry. Que yo sepa no hace absolutamente nada XD He estado probando por si acaso, pero las imagenes no cambian en nada si generas con enter o sin enter. quizá estén intentando separar el prompt en chuncks, pero eso se hace usando "BREAK". si sirve para algo yo no he podido ver para qué, pero interesante de todos modos. Merci!
Prompt blending work with cloth? for example, I want a base maid dress mixed with combat clothing so I'll go with wearing uniform [maid | military], [maid:military:10] but its look doesn't to be so effective.
I dont think it will, no. At best you could get a dress with camouflage colors. The best way to mix parts in a good way is to use inpainting. You could try, however, something like [Maid dress : military clothing : 6] and play arround with it. Maybe even go crazy, use controlnet open pose and prompt for something like woman wearing a dress AND military clothing. That will probably create some nightmare stuff but doest hurt to test 😂😂
Hi, I'm working through your videos but I just cannot fin the part, why you have those options between "seed" and "script", how can I get them and how important are they? I'm at the very beginning to understand and work with SD, but I look forward to watch all your guides! Thank you so much for your work!!! :D
hi!! I need to do a video on it probably hahhahha those are extensions, most are extra stuff that just helps in some specific tasks. One of them is controlnet, which I think is the best one of all and I've done a vid on it. You can get most of them by going to the -Extensions- tab and -available- hitting -load from- and then looking up the names and reading what they do to know if you are interested. Then just click -install- and, after it loaded, hit -check for updates, and apply and restart UI-. That should install them And thank you for watching!!
Like a full on "beginner to master" type of deal? Not really, srry. Don't have time to make a full course. I treat this channel as a course though. (Not exactly sure what you meant, srry😅)
I dunno the prompt blending doesn't really work for me like it does for you. It either puts none of it in the image or both but not blended together or just one
you need to play with the steps a lot. it will also depend on the 2 concepts you are trying to blend. In my case I used a very easy to see example, but it isnt always like that. For instance "[snowy|on fire]" will not make it look like a blue fire or incandent snow. But it will probably try to add a fireplace and snow at the same time. What is it you are trying to blend?
How do you find that generation thingy in 0:26? I even fail to find where to generate the image in CIvitai. All I can do is go to homepage, but generate tab is nowhere.
Oh, if you are generating through webs. All of them have different ways to do stuff. The most similar to what I have is a google colab for automatic1111
thank you for ur video, really appreciate that you share ur knowledge, just confused about [> 1 = less weight than without parentheses], so you mean if setting a word weight > 1, it is actually less weight than other words without adding text weight?? how come? I thought the default text weight is equal to 1, so those texts adding weight over 1 should be more important than other words without modifying
My bad, should have explained that better. I just meant that a word between parentheses doesnt always have more weight. If you ad a value that is less than 1, the weight of that word is less than when it didn't have the parentheses.
Does prompt blending not work anymore or does it need a specific extension? I am sure it worked for me last month, on my new PC it doesn't however. [female:skeleton:0.5] or [female:skeleton:15] for example will always just produce a female, no skeleton anymore. I tried it out with different prompts and steps, never worked anymore :(
It should work still, at least in A1111 and SD 1.5. Havent actually tried it with SDXL But it does work for the normal version. Maybe it's a checkpoint thing. Some checkpoints have bias towards making women, so it could ignore the skeleton part xD Try using 30 steps and [woman | (skeleton:1.2) : 10] or smt like that
@@Not4Talent_AI Yep, my current installation is only a few days old, so everything is fresh and working, except the prompt blending. Tried to figure it out, and found a script on git that is named "Prompt Blending", that's why i was confused in my comment. Tried it, does not work either. Is there something i could do? I could not find any information or setting. Other than the compatibility for old emphasis and its unchecked. Whatever i try the blending just wont work. I have no extensions affecting prompts. Only dream booth and Cozy-Nest UI.
@@madrooky1398 oh, my bad on the "prompt blending" part, didnt know there was a script with that name. Srry for the confusion. For the prompt editing part I have no idea why it isnt working. What have you typed atm? Been reading a little and all I found is people that complain about it not working properly, but none saing that it doesnt work at all. Maybe it is the sampler (I read that some work better than others even though I havent had that experience) Or too high/too low sampling steps. Other than that I have no idea, if you have an example maybe we can work it out
@@Not4Talent_AI Well all right i got it. Tricky business. The thing is when you start a simple prompt the sampler has not so much issues merging two concepts. But alone negatives can destroy it. Also the models i use are pretty solid. I only need a few prompts for the last bit of precision but that was it that prevented it working at all. When i removed prompts for anatomy correction i started seeing something happening during sampling, but it remained unstable. Now the trick is you have to include the entire prompt in the merge and balance the concepts. And the model itself can also limit the fun, like cineDiffusion_v3 which look awesome but it cant show full body including feet and so my idea did not fly...lol literally. With another model that allows more freedom it works, but looks much less pretty. Here is the promt, have fun with that piece of engineering... 😆 [F-16 Falcon:(slim girl, flying in the sky (towards camera:1.2), arms spread like wings below the large symmetrical (fighter jet wings:1.2):2.0):12:F-16 Falcon, from the knees on downwards [jet engine afterburner spitting fire:12] , long red hair with ponytail, on the hips hard points attached carrying large missiles parallel to body:34:beautiful face, smile:43] negative: (text, logo, watermark, (close-up:1.2), less than 4k and ultra quality:1.3)
@@madrooky1398 uuuu not currently home but Ill try it for sure. Love jets and tanks and stuff 😂. I feel like the prompt overcomplicates stuff with the ((stuff:x)stuff2:2) which is maybe causing some confusion. Also, it will be hard to have that much of a change (between a jet and a person). Keep in mind that Stable diffusion wont take the original thing it generated and change it for another. It will try to include the new thing on the image, but not necesarly change what it already has made. (Kinda hard to explain like this hahaha srry). Ill probably make a video more in depth on this stuff. Dont know if you had great results with it but im definetely interested on how it would look like. It is a really complex result you are going for which I love. Really fun to look for prompts that do a consistent job on complex stuff like that. If you want to make it easier you can create the image of the F16 first and then use the "img2img" tab to change it. The last video I made goes a little more into detail on that. Hope you found the result you where looking for! Srry for the long respons and thanks for the feedback!!
CFG is basically contrast. While I get it’s hard to produce same image at different CFG …it can be done. But impossible to reproduce same contrast.
It allocates noise to solid lines - contrast - also removing that noise from the pool of changes your prompt can make (essentially managing “creativity”).
Lower CFG can often help with over burned Lora merges
oh wow that's really useful info that I didn't know. I'll pin this comment if you don't mind, so everyone can learn a little more .
Thank you so much!
@@Not4Talent_AI sure! Happy to help. Yeah I often want CFG high … I go 10-15 usually
HOWEVER when I use tons of LORA’s … like 5-10 of them ….. very easy to get messed up images …. I drop the CFG down to 3-6 …. And everything clears up.
@@alecubudulecu I'm used to leave it at 7 for everything xD I'll have to get rid of that bad habit now hahhahaha
Im so sick of watching videos of people telling me about "advanced" techniques, but what they really do is show what openpose controlnet is, or something basic like that. But you actually have taught me like 5 new things in only a few minutes. Thank you, and good job figuring out this stuff!
Glad you found it useful!! Thanks!
Same. Big 👍
Thanks for the video
ty for watching!
Outstanding quality for a new channel, easy to follow, straight on point and answering exactly the questions a viewer has at certain points in the video. Keep this going and you will build a successful channel! 👍
Thank you so much!! I' ll certainly try hahhaha
thank you Man!!!!!!!!!!
happy to help!! ty for watching :3
Dude, that was neat.
I loved that all subject are in chapter and covered with the right amount of time. Your video is easy to use as a cheat sheet.
Many tutorial youtubers should learn from that !
Thanks!! Really encouraging to hear such positive feedback :3
Good video, I learned a lot
Glad to heard that, thank you!
I didnt know about some these amazing video great job
Thanks!!
Very cool thank you, this helps a lot .
So glad it does!! Thanks for the kind comment!
A hidden gem of a channel , thanks very much , looking forward to more vids
Thanks for the kind words!! Trying to get one ready for today, lets see if I can hahaha
Thanks mate. Been trying to learn better prompting these last couple of weeks. I've watched a bunch of videos and read heaps on it. With all that research I've done I still didn't gain as much understanding as I did from this one 12 minute video. Thank you so much. I think yours will be a channel to watch.
Really happy to hear that!! I'll probably make another prompting vid soon enough with some other things that could be interesting as well as a deeper dive into concept bleeding.
Thanks!
My dude, you've just made one of the best fucking videos about 'prompt engineering' and all that shit that I've ever seen. Congratulations. Please expand upon this topic with this style of video and IM SURE you'll get a big fat number of subs.
hahahaha thanks for the support!! I do have in mind expanding on this, (as I tried making a short video on it but I still need to learn how to count to 60, made it 1:34 min so it cant be 1 short). And really glad you found it good!
Mate, what a great job you did! I've watched lots of videos on this and yours is top notch! Will be recommending and keeping an eye out on anything else you do!
Thank you so much! I'll surely try to improve on it!! Feedback like this helps a lot (be it positive or negative) :3
you re a Boss Dude!!!!!!!!subscribed
Thank you!!
Exaxtly what this n00b was looking for. Thank you so much! ❤ Also you have a great presentation style and sense of humor.
thank you ahhahaa. Glad that you liked it. Really appreciate the kind comment!
I've seen so many of these videos and many of them have useful information. But the creator is either going off on tangents or using unclear language, sometimes it can feel a little inaccessible. Other creators don't do a very good job of visualizing what they are trying to explain. Point is there are a ton of inaccessible and undigestible prompt guides. Yours is exactly what I needed. It goes over information I already know but it's done in a way that is quick and concise, so It serves as a refresher. When I come across information that I haven't heard before, it's easy to understand. Your visuals are showing exactly the function you're trying to explain. The language you're using is extremely accessible and the way you progress through the information is very well thought out. Bravo and big thanks!
Super happy to read this, thank you so much!!!
one of the most helpful videos, thanks
glad it helped!!
Thanks a lot.
ty for watching!
Hi everyone, in 3:02 I made a typo, it should be
man i literally learned so much from this. i knew some things but especially the "prompt bleeding" stuff was so helpful
glad it helped!!!
straight to the point, thanks
Ty for the kind words!
Thank you! I really appreciate you filming this for us!
thank you for watching! hope it helped
Very cool video. It was a surprise when i saw that you have only 44 subscribers... The quality is outstanding and I learned a lot, thank you!
Tyty!! Super happy you liked it and found it useful :3
A little tip for you. Use aspect ratio helper. I often make 16:9 images and if you use ratio helper it will do the math for you when you move the sliders around, although it only works in tex2img. You can set it to 1:1, 2:3, 3:4, 4:5 or 16:9. This has saved me so much time, as it will slide both bars when you move one and I no longer have to do the math to figure out what ratio I'm using.
Edit: lol, never mind seems you are using something like that, you just didn't mention it.
Yes! I actually use the one you said, but I didnt mention it as I was keeping everything as vanilla as possible xD.
I do plan on making a video about the best extensions and this one is going in for sure. Not over complex, but super useful.
Ty for the tip though! intention is what matters hahahah
Ooof, this was humbling. I watched it to see if I could pick up some new tricks, not very hopeful about it. At the beginning yeah, I knew everything. But then? A bunch of stuff I had no clue about! Thanks!
Glad it gelped!!
Underrated content
hahahahaha thanks!!
thanks
Ty for watching!!
Amazing channel, great work! thanks for the prompt mixing trick.
Glad you liked it!! Thanks!!
Amma do a video going more in depth soon enough, hope it helps :3
@@Not4Talent_AI It does! Sorry for the late reply.
I think it's like the 80-20 rule. You can get to the 80% of what you really want to make and it's all fun and games in SD.
But the last 20% of the way makes a lotta difference in generations.
@@natsuschiffer8316 Yeah, it is a really big difference just with that 20% ahahahha
You are really diamond person in this world bro when I start to watch your video it gave me relax. Thanks for ur hard work also I want to you keep upload more SD Videos
thank you so much for the super kind comment!! I will keep posting for sure with an audience like you :3
@@Not4Talent_AI Your welcome sir
Neat and very useful, thanks a bunch !! 👍👍👍
Thanks for the kind comment !! Hope to keep providing useful info !
awesome video bro, subbed
thank you!!
One of the best tutorial for Stable diffusion so far. I have worked with Stable diffusion for quite a while, and still learning. I can definitely say, this is one of the best out there.
Thank you so much!!!
I would dare to say that we are all still learning😂. So much to try and to figure out. As well as so many new things every day hahahaha
The best thing in this video is the use of visual examples, most other channels sometimes explain something but dont give us the real output, and i know how hard is to make videos in such manner, so you earned a sub !!!
Thank you so much!!
(and yeah, most of the time I spend on videos is usually experimenting and finding examples for stuff XD)
Easy subs bro, thank you for the the teaching. And please keep goin
Will keep going for sure!! And with all this support ain't no way I'm stopping hahaha
Thank you so much!
Bro thank you so much, this was excellent
Thanks!!!
woah! bro awesome tutorial been looking for this finally a saviour arrived. dude make more on prompting tutorials. thanks for the tutorial.
So glad you liked it!!! I'll keep making more for sure, there is still a TON to talk about :3 Any suggestion is very much appreciated!
This is an extremely useful video, holy shit. I didn't know anything about prompt blending and it looks so useful for controlling the direction your prompt is taking. Awesome work, man
Thanks!!! Really glad it was useful!
Btw if you want to look more of that topic up, the actual name is "prompt editing". I have a short (well, two of them bc I messed up the duration xD) that dives a little deeper into that. Even though it doesn't add that much to what I said on this video.
And yes, I also think it is really useful specially without controlnet. I'm even thinking about dedicating a video solely to that and some other non-typical pormpting techniques like adding "BREAK" or "AND".
Srry for the long comment, just really happy people are finding this helpful xD
@@Not4Talent_AI nice! I'll check them out
Even though I prefer ComfyUI, this video helps me a lot. Thank you!
comfy is great, but I havent used it yet for a lack of time T_T. Happy to hear it was helpful still tho, tyty!
nice bro ty
glad it helped!!
This is gold...thanks a lot!
ty for watching!
This was super helpful. Didn't know about the blending yet. Drives creativity to a whole new level.
Glad it helped!!
Holy shit this was by far the best and most informative video of all the basics I have ever seen, way better than olivios tutorials. You clear up so many things I struggled with when trying to engineer the prompt until it's really what I want. Big thanks!
Love that you found it useful!! thanks for the kind comment!
TY for the video.
Definitely some cool stuff I haven't seen before.
Thanks!!! Hope it helped
Earned a sub for sure. You have a very effective way of communicating and showing examples rather than just mentioning abstract concepts is extremely helpful. Plus the content itself is actually useful 😅
I had been struggling to tweak my prompts to get an idea out of my head and into an image, and was still hitting a wall after hours upon hours... I took the concepts you mentioned and finally got the result I wanted in about 20 minutes!
Extremely happy to read this, thank you so much!!
Glad it was helpful!
Your videos are so good! This one is a game-changer for me! Thanks so much! You definitely deserve more visibility for the raw quality of the videos you create!
thank you so much!! glad it helped hahaha
Very clear and detailed information, amazing job!
Thanks!!!
Great video and explanation!
Glad you liked it!!
it’s the first time I see the blend stuff, nice
Hope it helps! Planning on making a new videoon it soon :3
Embarking on a journey of storytelling and video experimentation, VideoGPT seamlessly integrates into my workflow, refining my content with its subtle enhancements.
8:17 I like using prompt switching for hair colour and ethnic blending.
For example '[Red|Blonde|Blonde] Hair' gives a strawberry blonde and doing something like [Swedish|Norwegian] gives a more exotic results than either word alone.
uuu nice!! I should try that, sounds fun hahaha tyy!
thanks for this video! it was very detailed and had lot of things I didn't know. I've been using stable diffusion since December and I really didn't know about prompt blending and other things you showed in the end. Can you create a more detailed separate video for the concepts in the end that is much slower so that more people like me can understand prompt blending? thanks a ton :) Subbed!
Yess!! Was planning on doing that. Also adding some other stuff. I still need to experiment a little more to fully be able to explain the functioning though. It will come though! Thanks for the kind comment!
Stable diffusion would overtake stuff like Dall-E/Midjourney right out of the gate if their prompting was just made easier. To this day it is hard for me to actually want to use stable over those other programs, particularly for work related stuff. It sucks having to sit there grinding through dozens of prompts in hopes it might kinda, sorta, almost spit something out tht looks slightly like the thing im looking for.
I'm not sure if I agree. when it comes to prompting, yes. Midjourney, for example, can give 1000% better images with a simpler prompt. But Stable diffusion has SO many tools to give infinite control over the result you are looking for. It takes a little more work, but it is definitely possible to have EXACTLY what you want.
Midjrouney is way better at just creating really nice looking stuff out of the bat though, as well as having way better anatomy and over all composition
@@Not4Talent_AI I say this as someone who uses AI art in a fast paced work environment. The extra steps and complexity of the near paragraph long pro/negative prompts jam up the work cycle. Particularly when dealing with multiple clients who need multiple different types of art pieces incorporated into their projects. Midjourney/Dall-E are simply easier to use, with less steps required, and much less experimenting with new prompts. At the end of the day the extra features are nice but mass adoption will go to the program that strikes the balance between output and simplicity. Every single tool that has been introduced based off of the stable model needs experimenting with before one can use it with confidence. Not practice but straight up experimenting with to find its limitations.
Midjourney is just a handful of buttons you can feed a single line of key words to and you get 4 specific examples with minimal input needed to choose and generate based off of any generated image in the exact same page. Sure theres a lot you cant do, but the many fields that look towards AI mass adoption already have tools in place, that are not experimental to achieve what Midjourney lacks. With those tools comes the confidence I spoke of earlier.
@@SoullessAIMusic Yeah, you are completely right on this. I still think that with experience that can change a little, but it is true that most people will go to the easiest option.
Professionally I would use MJ and SD at the same time, but if what you need is speed, nothing beats midjourney atm
That's a good video it's so helpful 🤩🤩 bro keep teaching us like that I'm waiting for the next videos 💪🏼💪🏼
Thank you so much! Will definitely try my best to do so!
the tough part about tutorials for stable diffusion is they go out of date one month after they are made lol
Hahahahhahahaha 1000%
This one I think holds up pretty good, since prompting doesnt change as much. But totally 😂
Pretty damn good video for a new account. It's a good overview, but I will say, if someone is just learning, there's too much information presented too fast. I've been messing around with AI art for a little over a month and already been introduced to all these concepts, so it was easy to follow. But a new person will struggle with how fast everything moves. All the information is amazing and accurate, but you might consider making a mini series explaining each subject / idea a little slower to gain more subs. Just a suggestion! Keep up the great content!
Thanks a ton, this is just what I needed!! Love all the positive comments but I also had a feeling that maybe it was too much.
(Mainly with the next video).
After publishing it today, I'll try to get a better balance to have good information for everyone but not so much that it becomes overwhelming.
Really helpful feedback, appriciate it!!!
Is Prompt blending available scene the very beginning of SD?
Why has no one ever make a UI extension for this?
Yes! It is actually called "prompt editing" in the official SD guides. Just by typing the [] and the correct formatting it should work fine
9:21 you created lady death 😃
Hahahahhaha
Wait a minute! I can highlight a word and simply press Ctrl + up or down to change the weight?! Video worth it for that alone.
Hahhahahahhahaja yup 😂😂😂
tldr for prompt layout:
1. Image type (e.g. painting, photo)
2. Main subject (e.g. girl, cat)
3. Action (e.g. driving, walking)
4. Object/Environment (e.g. car, city)
5. Style (e.g. cyberpunk, watercolour)
I prompt like that, it isn't necessary a must follow rule. Just what I found out works best in my case. Haven't done a ultra study on best prompt formats XD
ty though!
@@Not4Talent_AI it's a good promptflow to not forget any key details
can always swap quality tags to the front tho
How does this apply to midjourney? Are they transferrable concepts?
Most of the basics are, references (in other pages, CivitAI is only for SD I think), AR affecting the result, the seed being super useful, the style list, and the fact that all words have a certain Concept Bleeding (major or minor depending on each model, this can change even in-between stable diffusion models).
Some main differences I've seen is that, much like stable Diffusion 2.1, midjourney (v5) really likes natural language instead of tokenizing. So try typing prompts like talking to a human artist.
Also midjourney is more "artistic" and I find it harder to control the result.
And the final tips I give that are used to "combat" concept bleeding, can't really be used in midjourney.
Even though midjourney has it's own list of cool things most people don't use, here is a page with all that information: docs.midjourney.com/docs/multi-prompts
Hope it helped!
Curious about where you've seen double colons documented; seems they do something in midjourney, but aren't not documented in the official SD prompt guide or anywhere I can find. [text:num] grammar also isn't in the guide, which only describes [text] to apply a 0.9 scale factor to the default text weight and prompt blending with [text1:text2:step].
Hi! Srry I forgot to respond. Here is where I found all the "prompt editing" info. I will add this link in the description as well, thanks for reminding me!
github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing
@@Not4Talent_AI Great, many thanks! I'd inadvertently been referencing an unofficial guide that's decent and with recent revisions, but evidently it's missing a few details. I think I might have looked at the wiki way back when getting started, but promptly forgot about it (no pun intended!)
@@makers_lab hahhahaha Hope this one helps then!! It has some interesting stuff in it, even though not too explained. Fortunately, what I hope to be today's (if not tomorrows) video, will cover that :3
Excelente video, saludos desde Perú, gracias por toda la información, llevaba mucho sin comprender varias de las cosas que mencionas, aunque aún me queda la duda sobre para que sirven poner "enter" en los prompt generando dos oraciones distintas.
Gracias a ti!! Hmm a que te refieres con "enter" ? Poner la palabra "ENTER" tal cual? Si es eso no me suena haberlo visto nunca la verdad
@@Not4Talent_AI Oh lo siento si no me expliqué bien, me refiero a que he visto algunos prompts en los que hay dos lineas separadas (una arriba de la otra)
Por ejemplo:
1girl, black dress, outside in sunny day
Magic forest, giant trees
@@ososimiospub6406 ahhh Vale, mala mia, srry. Que yo sepa no hace absolutamente nada XD He estado probando por si acaso, pero las imagenes no cambian en nada si generas con enter o sin enter. quizá estén intentando separar el prompt en chuncks, pero eso se hace usando "BREAK".
si sirve para algo yo no he podido ver para qué, pero interesante de todos modos. Merci!
@@Not4Talent_AI acabo de ver tu publicación sobre "BREAK" y es increíble, grqcias por todo amigo, espero que tu canal crezca mucho más, te lo mereces.
Prompt blending work with cloth? for example, I want a base maid dress mixed with combat clothing so I'll go with wearing uniform [maid | military], [maid:military:10] but its look doesn't to be so effective.
I dont think it will, no. At best you could get a dress with camouflage colors.
The best way to mix parts in a good way is to use inpainting.
You could try, however, something like [Maid dress : military clothing : 6] and play arround with it.
Maybe even go crazy, use controlnet open pose and prompt for something like woman wearing a dress AND military clothing.
That will probably create some nightmare stuff but doest hurt to test 😂😂
Hi,
I'm working through your videos but I just cannot fin the part, why you have those options between "seed" and "script", how can I get them and how important are they?
I'm at the very beginning to understand and work with SD, but I look forward to watch all your guides!
Thank you so much for your work!!! :D
hi!! I need to do a video on it probably hahhahha those are extensions, most are extra stuff that just helps in some specific tasks. One of them is controlnet, which I think is the best one of all and I've done a vid on it.
You can get most of them by going to the -Extensions- tab and -available- hitting -load from- and then looking up the names and reading what they do to know if you are interested.
Then just click -install- and, after it loaded, hit -check for updates, and apply and restart UI-. That should install them
And thank you for watching!!
You have a course?
Like a full on "beginner to master" type of deal? Not really, srry. Don't have time to make a full course. I treat this channel as a course though.
(Not exactly sure what you meant, srry😅)
I dunno the prompt blending doesn't really work for me like it does for you. It either puts none of it in the image or both but not blended together or just one
you need to play with the steps a lot. it will also depend on the 2 concepts you are trying to blend. In my case I used a very easy to see example, but it isnt always like that.
For instance "[snowy|on fire]" will not make it look like a blue fire or incandent snow. But it will probably try to add a fireplace and snow at the same time.
What is it you are trying to blend?
What is the website of the AI algorithm that you use? I havent seen so many sliders and options anywhere? Thanks
Its a local instalation. Name is "automatic1111"
@@Not4Talent_AI Oh, I didn´t know you can do that. Ok, thanks
yep! if you have a good GPU you can install it for free with no censorship or nothing, its amazing tbh @@stuart2777
How do you find that generation thingy in 0:26?
I even fail to find where to generate the image in CIvitai. All I can do is go to homepage, but generate tab is nowhere.
Oh, if you are generating through webs. All of them have different ways to do stuff. The most similar to what I have is a google colab for automatic1111
thank you for ur video, really appreciate that you share ur knowledge, just confused about [> 1 = less weight than without parentheses], so you mean if setting a word weight > 1, it is actually less weight than other words without adding text weight?? how come? I thought the default text weight is equal to 1, so those texts adding weight over 1 should be more important than other words without modifying
My bad, should have explained that better. I just meant that a word between parentheses doesnt always have more weight. If you ad a value that is less than 1, the weight of that word is less than when it didn't have the parentheses.
Oh wait, I thought I typed
@@Not4Talent_AI lol,thank you for ur reply, I just literally have watched all ur videos already, looking forward for more ,have a good day😊
@@youwang9156 hahha thank you so much for the support man! I have a ton planned 😂
Btw pinned a comment with the
Does prompt blending not work anymore or does it need a specific extension? I am sure it worked for me last month, on my new PC it doesn't however. [female:skeleton:0.5] or [female:skeleton:15] for example will always just produce a female, no skeleton anymore. I tried it out with different prompts and steps, never worked anymore :(
It should work still, at least in A1111 and SD 1.5. Havent actually tried it with SDXL
But it does work for the normal version. Maybe it's a checkpoint thing. Some checkpoints have bias towards making women, so it could ignore the skeleton part xD Try using 30 steps and [woman | (skeleton:1.2) : 10] or smt like that
I also steal prompts to learn
the way to go xD
@@Not4Talent_AI learn from those who already know. Life hack
@@WiseOwl_1408 hahahaha pretty much. Human evolution in a nutshell
this will help so much with the porn images i create! thanks alot
xDDDDDDDD It sure will hahahah
Bro, the prompt blending is a script. It is not working by default.
It shouldn't be I don't think. Let me check.
I don't see anything about it beeing a script, are you using Automatic1111?
@@Not4Talent_AI Yep, my current installation is only a few days old, so everything is fresh and working, except the prompt blending. Tried to figure it out, and found a script on git that is named "Prompt Blending", that's why i was confused in my comment. Tried it, does not work either.
Is there something i could do? I could not find any information or setting. Other than the compatibility for old emphasis and its unchecked. Whatever i try the blending just wont work. I have no extensions affecting prompts. Only dream booth and Cozy-Nest UI.
@@madrooky1398 oh, my bad on the "prompt blending" part, didnt know there was a script with that name. Srry for the confusion.
For the prompt editing part I have no idea why it isnt working. What have you typed atm?
Been reading a little and all I found is people that complain about it not working properly, but none saing that it doesnt work at all.
Maybe it is the sampler (I read that some work better than others even though I havent had that experience)
Or too high/too low sampling steps.
Other than that I have no idea, if you have an example maybe we can work it out
@@Not4Talent_AI Well all right i got it. Tricky business. The thing is when you start a simple prompt the sampler has not so much issues merging two concepts. But alone negatives can destroy it. Also the models i use are pretty solid. I only need a few prompts for the last bit of precision but that was it that prevented it working at all. When i removed prompts for anatomy correction i started seeing something happening during sampling, but it remained unstable. Now the trick is you have to include the entire prompt in the merge and balance the concepts. And the model itself can also limit the fun, like cineDiffusion_v3 which look awesome but it cant show full body including feet and so my idea did not fly...lol literally. With another model that allows more freedom it works, but looks much less pretty. Here is the promt, have fun with that piece of engineering... 😆
[F-16 Falcon:(slim girl, flying in the sky (towards camera:1.2), arms spread like wings below the large symmetrical (fighter jet wings:1.2):2.0):12:F-16 Falcon, from the knees on downwards [jet engine afterburner spitting fire:12] , long red hair with ponytail, on the hips hard points attached carrying large missiles parallel to body:34:beautiful face, smile:43]
negative:
(text, logo, watermark, (close-up:1.2), less than 4k and ultra quality:1.3)
@@madrooky1398 uuuu not currently home but Ill try it for sure. Love jets and tanks and stuff 😂.
I feel like the prompt overcomplicates stuff with the ((stuff:x)stuff2:2) which is maybe causing some confusion. Also, it will be hard to have that much of a change (between a jet and a person). Keep in mind that Stable diffusion wont take the original thing it generated and change it for another. It will try to include the new thing on the image, but not necesarly change what it already has made. (Kinda hard to explain like this hahaha srry). Ill probably make a video more in depth on this stuff.
Dont know if you had great results with it but im definetely interested on how it would look like.
It is a really complex result you are going for which I love. Really fun to look for prompts that do a consistent job on complex stuff like that.
If you want to make it easier you can create the image of the F16 first and then use the "img2img" tab to change it. The last video I made goes a little more into detail on that.
Hope you found the result you where looking for! Srry for the long respons and thanks for the feedback!!
Bro is it completely free ??
As long as you have a decent gpu, yes
Wtf hapened at 2:35
no idea, what do you mean?
Promo-SM 😋
not really sure what this means, but like and heart xD
I lost you veeeery quickly..
damn, srry xD It goes pretty fast tbf
Now we got nobodies blabbering random stuff.
no idea if you mean me, but if you test the random stuff it works xD
I have found this artist reference useful: docs.google.com/spreadsheets/d/1SRqJ7F_6yHVSOeCi3U82aA448TqEGrUlRrLLZ51abLg/edit#gid=2005893444
Nice! thanks for sharing!!