I think most of this is technically possible with the automatic1111 version, obviously this has a much better user experience. There's a model called clipseg which I believe this uses. There are some extensions for the a1111 webui that uses clipseg as well. - for masks you can use the txt2mask extension - generate images with upscaling by default - variations is using the same seed again with extra variation - style is predefined "in style of blah blah" with prompt strength using the same seed Again, obviously freewayML has a much better design and user experience than gradio. I kinda personally wish a1111 puts the gradio ui to rest and adopts a more conventional web frontend that uses the api locally, then I imagine web developers will start contributing a lot more and improving the over all user experience. The gradio python stuff is too rigid for ux and especially visual design improvements.
Is there a website that replicates mid journey? I find Discord very hard to use. Right now I'm using playground AI and I love it but it seems to come back with triggered safety issues a lot of things even for anything. Thanks so much for any answers
I think it's because you were using DPM2 a. I've never had much luck using this sampler, I assume it was created for some academic reason? DPM adaptive and even standard PLMS are better. When you use Euler a you get similar results to FreewayML, this is really the only sampler I use. Nothing I saw here couldn't be done in the Automatic1111 Web UI with little effort. I think the enhance feature is zooming in and re-doing parts of the image using img2img, the same as the Web UI highres. fix (I use this to upscale my images with more detail). The styles likely are just words added to the prompt with weights based on your selection, at least the cinematic test you did was defiantly not based on a custom model. The outpainting issues show that they are not using the outpainting model because you won't get picture frames with that model. I don't see anything new or special here.
I thought it was some kind of a painting assistant, which it doesn't seem to be. I would really be happy about a way to simply decide every single detail manually and prevent the A.I. from making preposterous mistakes. If I could literally get a gold material and paint gold all over a canvas and get realistic shades out the box without swapping back and forth between colors on the palette like traditional software assisted painting, it would be amazing. Clip Studio have such a feature where you can paint repetitive patterns along a Bezier curve but it's really rudimentary.
My problem with Midjourney is having to generate everything on Discord with your stuff getting lost in an sea of other stuff. It’s sort of like me writing on Microsoft Word, but other people's writing keeps getting added to the document as I go. Or like having a phone call in a room full of other people also having a phone call.
I feel like I've just given my email to yet another company that does something not that different to what I can do for free. I mean, sure I could have watched you use it, but this is just a sponsor channel at this point so I figured I didn't need to and just decided try it out myself.
I hate how all of these sites are taking advantage of a model that is originally open source for all, yet leave it to the lowlife's & greedy humans out there to use it and make money off of others with.
I think the credit system is a deal breaker. I just love generating a bunch of crappy images for fun. I wouldn't mind credit systems as much if it guaranteed it could fix hands.
@@MattVidPro also can you check out MJ4 i been noticing their bots are always down. Today they stopped all comments in the 'premium help' chat because it was flooded with users screaming "why is it saying interaction failed" Theres a status tab u can see where it states how many times the bot has been down :(. Waaaaay too many times for $50
they are a bit sneaky with the pricing: on the base model you get 250 credits for $8 per month and after those 250 they charge you $0,05 per credit. same for their unlimited plan > 750 credits (aka 1750 images) for $25 per month and after that $0,04 per credit.
Midjourney is best but unfortunately it has censorship of some stupid words like killing or blood and .... I like to make Horror Art How can we get rid of this banned words and make unlimited Art without censorship?
"These are things I had never seen before". C'mon man most tools you were referring to when you said that are literally available in Dreamstudio since months ago. It's just SD's inpainting and img2img presented in a different way. The other tools are a mish mash of tools available on other websites (like ai scaling). I mean it's cool they are trying to itegrate everything in a more coherent way, but the tools themselves are not new.
Quietly vanished off the face of the web at the turn of 2024. Very innovative when it came out - with many unique tools. High quality at the time. Had several updates in its lifetime. Had several workflow bugs. Mildly censored. Customer service (on Discord) responsive and magnanimous. Gradually lost the leading position in the AI race. In my view, it had vastly insufficient marketing/publicity which is why it turned out unprofitable (apparently).
By the end of the year you need to do a top-10 ai websites of 2022!😮
generating variations on masks is a great feature, would like to see other models implement that
After uploading an image on FreewayML you can use the PROMPT button to convert the image to text.
That is a feature that could be really useful. Does it use up a credit?
@@shinyshinythings yes
If he still alive yes he gonna love ai tech
And he'll live forever, as he himself, is an ai
@@apersonthatsucksatnames1133 yes
I think most of this is technically possible with the automatic1111 version, obviously this has a much better user experience.
There's a model called clipseg which I believe this uses. There are some extensions for the a1111 webui that uses clipseg as well.
- for masks you can use the txt2mask extension
- generate images with upscaling by default
- variations is using the same seed again with extra variation
- style is predefined "in style of blah blah" with prompt strength using the same seed
Again, obviously freewayML has a much better design and user experience than gradio. I kinda personally wish a1111 puts the gradio ui to rest and adopts a more conventional web frontend that uses the api locally, then I imagine web developers will start contributing a lot more and improving the over all user experience. The gradio python stuff is too rigid for ux and especially visual design improvements.
Outpainting is the only thing I would have trouble with. For some reason I've never had any luck with it. Not that this has outpainting, but still.
OMGMMGGG YESDSSSSS THE OUTPAINTING IS SOOOO MUCH EASIER!
The SD model they use it's VERY impressive, i wonder if we will ever get to download it
Is there a website that replicates mid journey? I find Discord very hard to use. Right now I'm using playground AI and I love it but it seems to come back with triggered safety issues a lot of things even for anything. Thanks so much for any answers
just ignore it and keep going. or change prompts and then go back to the one. you want to use. it will work. I love playground too.
I've registered two months ago but too much involved in PixelzAI, I forget to used that. After watching your Edit thumbnail, I opened up again.
Geezer here...
Great vid, sincerest thanks.
I think it's because you were using DPM2 a. I've never had much luck using this sampler, I assume it was created for some academic reason? DPM adaptive and even standard PLMS are better. When you use Euler a you get similar results to FreewayML, this is really the only sampler I use. Nothing I saw here couldn't be done in the Automatic1111 Web UI with little effort. I think the enhance feature is zooming in and re-doing parts of the image using img2img, the same as the Web UI highres. fix (I use this to upscale my images with more detail). The styles likely are just words added to the prompt with weights based on your selection, at least the cinematic test you did was defiantly not based on a custom model. The outpainting issues show that they are not using the outpainting model because you won't get picture frames with that model. I don't see anything new or special here.
There’s a txt to mask script for stable diffusion that can mask what you tell it but only in 1 layer
Pretty cool. A more streamlined approach to stable diffusion inpainting. I get similar results using SD inpainting locally.
Cool, thanks
I thought it was some kind of a painting assistant, which it doesn't seem to be. I would really be happy about a way to simply decide every single detail manually and prevent the A.I. from making preposterous mistakes. If I could literally get a gold material and paint gold all over a canvas and get realistic shades out the box without swapping back and forth between colors on the palette like traditional software assisted painting, it would be amazing. Clip Studio have such a feature where you can paint repetitive patterns along a Bezier curve but it's really rudimentary.
exciting stuff, def gonna check it out
I believe it is other way around. Stability has incorporated runway models?
Really love your aspect…Thanks!!
My pleasure!
Thank you Matt!
Sure thing!
4:41 ouch
Matt did you see that replicate has a midjourney v4?
Thanks.
My problem with Midjourney is having to generate everything on Discord with your stuff getting lost in an sea of other stuff. It’s sort of like me writing on Microsoft Word, but other people's writing keeps getting added to the document as I go. Or like having a phone call in a room full of other people also having a phone call.
FreewayML's AI Editor (site is down)
thank you for the videos, but the problem is how to remove the watermark, waiting for your answer, thanks again, and have a nice day.
Super cool!
I feel like I've just given my email to yet another company that does something not that different to what I can do for free. I mean, sure I could have watched you use it, but this is just a sponsor channel at this point so I figured I didn't need to and just decided try it out myself.
I'd like to know the most coherent stable diffusion model out there.
Automatic have it already custom shapes mask inpainting and even some sd photoshop plugin
☝️😄 Steve job not dead , He sync his soul into Apple chip..
how do you do Personas?
I hate how all of these sites are taking advantage of a model that is originally open source for all, yet leave it to the lowlife's & greedy humans out there to use it and make money off of others with.
I think the credit system is a deal breaker. I just love generating a bunch of crappy images for fun. I wouldn't mind credit systems as much if it guaranteed it could fix hands.
U need a whiteboard behind u for us to keep up with all these apps. Think MJ4 would be top of the leaderboard ;)
interesting idea...
@@MattVidPro also can you check out MJ4 i been noticing their bots are always down. Today they stopped all comments in the 'premium help' chat because it was flooded with users screaming "why is it saying interaction failed"
Theres a status tab u can see where it states how many times the bot has been down :(. Waaaaay too many times for $50
How do I sign in?
Kind.
Neat
I am waiting combination function like in new MJ4
It can produce 1024px. Can any free ai image platform which can produce over 4000px image with high quality.
Not with high quality
You can do this in SD with a script
You can only compare the same results if you sue the same seed
Did this site get deleted ?
Is there a green-screen in your room? Your hand looks a little off, your left I mean
👋
Wow, 10mins, I'm early
they are a bit sneaky with the pricing: on the base model you get 250 credits for $8 per month and after those 250 they charge you $0,05 per credit.
same for their unlimited plan > 750 credits (aka 1750 images) for $25 per month and after that $0,04 per credit.
Right, midjourney will do the same for their $10 plan but not the $30
Midjourney is best but unfortunately it has censorship of some stupid words like killing or blood and .... I like to make Horror Art How can we get rid of this banned words and make unlimited Art without censorship?
This looks like Google's prompt to prompt
"These are things I had never seen before". C'mon man most tools you were referring to when you said that are literally available in Dreamstudio since months ago. It's just SD's inpainting and img2img presented in a different way.
The other tools are a mish mash of tools available on other websites (like ai scaling).
I mean it's cool they are trying to itegrate everything in a more coherent way, but the tools themselves are not new.
I haven’t seen layering yet, or style transfer integrated directly into sd
Quietly vanished off the face of the web at the turn of 2024. Very innovative when it came out - with many unique tools. High quality at the time. Had several updates in its lifetime. Had several workflow bugs. Mildly censored. Customer service (on Discord) responsive and magnanimous. Gradually lost the leading position in the AI race. In my view, it had vastly insufficient marketing/publicity which is why it turned out unprofitable (apparently).
Your website isn't working my dude
Hi
I prefer stable diffusion online. This freeway site harvests your email address. While the other does not.
Hi Kermit! hahahahaha LD
hello back 😄