Update app.py (Continued)

#3
by Yntec - opened
Owner

Hey @John6666 it's not letting me post on https://huggingface.co/spaces/Yntec/blitz_diffusion/discussions/2 anymore, so I opened a new thread.

I tried updating https://huggingface.co/spaces/Yntec/ToyWorldXL and use John6666/YntecLight but this is what I see:

ToyWorldXL Dark

It seems the dark colors are leaking? Can you make its light and dark color the light ones so this doesn't happen?

it's not letting me post on https://huggingface.co/spaces/Yntec/blitz_diffusion/discussions/2 anymore,

Eh? Is there such a phenomenon? It must be a malfunction of the function to prevent forum vandalism.

It seems the dark colors are leaking? Can you make its light and dark color the light ones so this doesn't happen?

It's weird that it leaks, but I can't help you with that, so I'll have a look.

This time the culprit was not the theme, it was the css in app.py.
If you remove it completely, the style of the title will be broken, but the color specification is unnecessary.
If you fix it like this, the title will not be broken and the dark color will disappear.

css="""
.gradio-container {background-image: linear-gradient(#254150, #1e2f40, #182634) !important;
 color: #ffaa66 !important; font-family: 'IBM Plex Sans', sans-serif !important;}
h1 {font-size: 6em; color: #ffc99f; margin-top: 30px; margin-bottom: 30px;
 text-shadow: 3px 3px 0 rgba(0, 0, 0, 1) !important;}
h3 {color: #ffc99f; !important;}
h4 {display: inline-block; color: #ffffff !important;}
.wrapper img {font-size: 98% !important; white-space: nowrap !important; text-align: center !important;
display: inline-block !important; color: #ffffff !important;}
.wrapper {color: #ffffff !important;}
.gr-box {background-image: linear-gradient(#182634, #1e2f40, #254150) !important;
 border-top-color: #000000 !important; border-right-color: #ffffff !important;
 border-bottom-color: #ffffff !important; border-left-color: #000000 !important;}
"""

↓

css="""
h1 {font-size: 6em; color: #ffc99f; margin-top: 30px; margin-bottom: 30px;
 text-shadow: 3px 3px 0 rgba(0, 0, 0, 1) !important;}
h3 {color: #ffc99f; !important;}
h4 {display: inline-block; color: #ffffff !important;}
.wrapper img {font-size: 98% !important; white-space: nowrap !important; text-align: center !important;
display: inline-block !important; color: #ffffff !important;}
.wrapper {color: #ffffff !important;}
.gr-box {border-top-color: #000000 !important; border-right-color: #ffffff !important;
 border-bottom-color: #ffffff !important; border-left-color: #000000 !important;}
"""

Normally, I would be excited to upgrade, but in the case of Windows and Gradio, it's chilling.
However, it does not seem to be as drastic a change as when they dumped 3. I mean, external.py looks really almost unchanged, so I guess it's just that the version went up to 4.9 and they decided to go to 5.

https://huggingface.co/spaces/saq1b/gradio-5-beta
https://github.com/gradio-app/gradio/blob/gradio%405.0.0-beta.3/gradio/external.py

Owner

Eh? Is there such a phenomenon? It must be a malfunction of the function to prevent forum vandalism.

Try it, go here: https://huggingface.co/spaces/Yntec/blitz_diffusion/discussions/2 go to the bottom where you'd post a message. Click the "Preview" tab, everywhere else you get a "nothing to preview" message, here it does nothing. If you upload/paste an image, it won't appear, and the "Comment" button doesn't work. It was so scary! What? I can't post messages anymore?? But it was just that page...

If you fix it like this, the title will not be broken and the dark color will disappear.

Thanks! That worked! I had no idea from where the colors were coming from.

they decided to go to 5

Oh no, it's like a Pandora's box and when you open it it's full of nightmares. I'm sure. They jump a number when old spaces are not backward compatible. I couldn't even jump all my spaces to Gradio 4 and it's getting killed? When will higher numbers just mean more fun and abilities instead of more hours of work testing code changes in a space?

Hopefully you're right and it's basically the same thing and it'll just work, I'll cross my fingers!

https://huggingface.co/spaces/Yntec/blitz_diffusion/discussions/2#66f2b0d20c2d19b3a55ec11f
here. It's a mysterious phenomenon.

That worked!

Good!

I'll cross my fingers!

Seriously, it's the only way. I'm the worst at making GUI's and I'm slightly traumatized, so I'm not sure I'm going to get a github account and commit to it...
If we only observe extenal.py regularly, well, we won't be left behind.
So I'm thinking that I'll get a scrollbar with all the elements in the theme.πŸ€ͺ

Owner

It's a mysterious phenomenon.

Thanks for the test, so it's only me, I'm glad everywhere else still works.

I'm not sure I'm going to get a github account and commit to it...

Their two-way forced identification thing was a deal-breaker for me because I had no idea how to set it up. Perhaps I need a phone and I need to install a two-way app on it and paste their secret code so I can login every time. No thanks.

Also, why is github necessary at all? Why can't model cards fill that role? It's weird that we can upload Gigabytes of image models daily and yet we can't upload some custom code and do away with Github? I guess I don't get it at all, specially where a free ZeroGPU space would only work with the Gradio from github and no other thing, huggingface seems to have the resources to replace github.

Wow, two-step verification. That's a relief, but a pain in the ass...😭

I guess the reason github is used is because everyone uses it, so it's literally useful as a hub. Everyone from the peripheral tools to the end programmer's environment is built on the premise of github. It's the de facto standard.

Windows has been around for decades because it's a superior OS? No. It's more critically important that they didn't abandon the x86 legacy until recently. (The ARM64 version is about to make compatibility obsolete)
Is English such a superior language? Esperanto may have been superior, but it never caught on. I don't speak it.

But it is true that we don't need to stick to github for Gradio.
There is a Gradio community here, and requests can be made there as well. That's fine, except for the super detailed code commits.

Owner
β€’
edited Sep 24

I'm glad English is the de facto standard, I've been able to use it to communicate with many people that only knew their mother language and English, including us. Imagine if I had learned Korean instead, I could have communicated with the pros of Starcraft, but couldn't have communicated with anybody over here. AI translators are supposed to be the answer, but I'd never trust a machine with what I said, last time I tried to speak Japanese with one I wanted to say "very" and it sent the word for "pretty" or "beautiful", and I wouldn't have known if the other person didn't speak English too.

requests can be made there as well

My request would be to support legacy versions of Gradio and make sure old spaces work for people staying on old versions, Gradio 3 spaces are dying daily as they no longer work, the code changes to save them are not trivial and their authors may be long gone, am I supposed to clone all the spaces of allknowingroger with the changes so they work and people can continue using them? I guess I'll just let them all die, but if this happens again by Gradio 6 and Gradio 4 spaces start dying, the downward spiral... Ahem...

lack of interest

Is gonna keep happening as people look for other options (some people get fed up and just buy the hardware and go local) just like they don't go to a store if they don't sell coca cola anymore, it's just one product but another shop may have it. Here it's like there's a big disconnect between Gradio developers and the Spaces ecosystem, it's as if they build it but don't use it so they don't deal with all the problems.

This is caused by the philosophy of "everyone should upgrade to the latest version and make sure to change the code to make it work", I remember comparing it to how ridiculous would it be if you had your classic car in good conditions and were forced to "upgrade" it to a new model because the roads didn't support it anymore, but THEN governments around the world started banning some old cars from getting on the road, so I had to bite my tongue.

I can't read or write Korean either, but I know a little bit about it because it's our neighbor country. It's an alphabet of sorts, and there are some rules to the shape of the letters. It has the same linguistic ancestry as hiragana and katakana in Japanese. So it is simpler than you think.

Japanese or hieroglyphs would be a more appropriate analogy. Japanese is a chimera of, say, Korean, Chinese, and pseudo-English. Although it is debatable which emerged first because of the unfortunate history between Korea and Japan. Japanese is a primitive indigenous language combined with Chinese, and in modern times, Dutch and English were also combined with Japanese. Even English is a chimera of Germanic, Old English, French, and so on. Hieroglyphs are a crazy combination of phonetic and ideographic characters.

one I wanted to say "very" and it sent the word for "pretty" or "beautiful",

The same goes for English. If you don't polish it after translation, it can mean the opposite. Also, you and I are often turned upside down.
That is why I write Japanese in a tone like a dubbed foreign movie. This way, there are fewer mistranslations. There are quite a few, but only relatively few.
In HF, I can get the meaning if I pick up the words, and everyone is very mild-mannered, so it's OK, but on a message board or something, this might cause a fight.

just like they don't go to a store if they don't sell coca cola anymore,

Good analogy.
Many people in HF disregard de facto standards. It can't be helped that, unlike social networking sites and bulletin boards, their own desire is not to stand out or make money itself, but that path is the road to recluse or hermit.
Most people in the world are not, so they leave in silence or have already left. There is no such thing as a consumer who says goodbye. They just shut up and are gone before you know it.

it's as if they build it but don't use it so they don't deal with all the problems.

Maybe that is 99% correct.
They are not in the position of Spaces maintainers.

Software that users create themselves, like the free software of 20 years ago, tends to be the most well-designed, but most users are not that advanced in programming. At that time, that culture was barely viable because it was still possible to make something useful with small-scale development.

Perhaps this is a problem that generative AI is solving in the long run. Once the users themselves no longer need to code, the rest is purely a matter of ideas and intent. Willingness is the hard part, though, in a sober way.

One solution is to have a Steve Jobs. Although he died because of an occult coffee enema. If led by a leader with a philosophy, programmers will do a good job as ordered.
In generative AI, we have such people in library developers and model trainers, but not yet in management.

Owner

Around here we have the curious case of having two Spanish languages. The one from Spain, from where it originated, but most people in the world that speak Spanish are in the americas, so when there's a movie or TV series that needs dubbing in Spanish, they make their own one in Spain, and we make our own one at Latin America, though, some places like Venezuela or Colombia make their third version of some things. We try to make a version that everyone would understand no matter where they live in latin america, but sometimes home versions like Candy Candy were only dubbed in Argentina and the rest of the world used that. It became an iconic interpretation because of excellent translation and voice acting, I can't imagine the characters with different voices and they were so emotive it's easy to get tear eyed remembering it.

That means for most things you get at least two choices of what dubbing to watch, the latin american version or the one from spain. Here at Mexico we used to be the best and absolute obvious option because we had voice actors able to improvise and change the dialogues if the original jokes didn't work, and create ones that would become memes still alive today. Then the voice actors strike happened and basically everyone was fired to hire cheaper alternatives, a cartoon like Dragon Ball Z Kai had entirely different voices for the characters.

And under the new ruling they became very strict about what the voice actors could say, first, they have to stick to a translation of what the original character said, even if nobody gets it, and second, they should make a voice as close as possible to the original character. The first seasons of the Simpson were incredible because of the natural voices and improvised dialogues that would manage to make tense moments hilarious and improve further the original episodes, 10 seasons later some original voices were gone and Marge and Homer were sounding like their original counter-parts, she with a raspy voice and him like an idiot (though, the character himself became a "flanderized" idiot as well.)

It was terrible and now, even though it's an unpopular opinion around here, I prefer to watch content using the Spanish dub from Spain instead of what my country makes, I was never convinced by Disney's Frozen, it turned out to be poor voice dubbing and on recent watching I switched to the dub from Spain and found it was a great movie, even Olaf was funny instead of annoying, I hadn't realized how important is both what the characters say and what voice they use to say it, even if it's the exact same movie. I have no idea how dubs to Japanese are like.

their own desire is not to stand out or make money itself, but that path is the road to recluse or hermit.

Oh yeah, I include myself about my models and spaces, I have never gone outside of HF to promote them, maybe they could have been used more if I made a reddit post telling people about them or something, but somehow I like the way things remain kind of niche, I've seen people at Civitai get daily questions and comments about their models, I don't think I'd be able to handle that if sometimes it takes me 2 days to reply to one of your messages.

Once the users themselves no longer need to code, the rest is purely a matter of ideas and intent.

What I've been wondering is how far the magic can go? I never thought about requesting seeds because I figured if people could do such things in a severless way, what would be the reason to buy an account and set up a ZeroGPU space? But now maybe it's a case of things that could be implemented and are not because nobody has done it? Maybe we could have them?

Like https://dezgo.com/image2image

How close can we get to duplicating their functionality? For a very long time I used them until the 31 models they have felt like a very small offering, I wonder if we could get img2img in a serverless way, such a thing used to feel huge until I realized it'd actually use less work than asking for a picture from scratch because you get the one from the user and the prompt and a few steps to change it. Maybe it's already possible but I never thought to ask.

Over here we have a dialect called Kansai-ben, which is practically a separate language. It is so cross-cultural that even the TV program lineup and major political parties are different. However, since the common language is not incomprehensible, anime, drama, and movies are common.

Then the voice actors strike happened and basically everyone was fired to hire cheaper alternatives, a cartoon like Dragon Ball Z Kai had entirely different voices for the characters.

What a nightmare. Voice actors are important. I'm not saying there aren't replacements, but they never grow up fast enough.
Over here, voice actors are basically not replaced except for those who played Kogoro Mouri in Detective Conan and Ryo Saeba in City Hunter, who were the heads of the voice actor labor union that was busted long before the strike, or those who died due to longevity or social troubles that made them unable to do their jobs.
They themselves are also idols of sorts, so commercially, well, it is unlikely to happen.

The voice actors for Son Goku and Vegeta are still the same as they were back then! I bet no one would be surprised if Goku's voice actress died tomorrow, though. Kaioh died in the middle of Super because of his longevity. Though Kaioh doesn't get much of a role in the series, his death does a lot of damage since he was in charge of narration! Even though the previous synopsis by him occupies 5 minutes of the series. The voice actor for Homer Simpson himself has not been changed and is a living comedian, but the new season of the Simpsons in Japanese itself has long since stopped being made due to rights issues.

In a country where the economy has been in recession or flat in dollar terms except for the richest 5%, the only thing that matters is whether or not anime can be made cheaply. About half of today's famous anime are based on free fiction sites on the Internet. Publishers avoid including political messages in their works because it would destroy the program itself. That may be the case with The Simpsons. Still, it is ironic that of all the major forms of entertainment, anime is the one that carries the most political messages. Drama comes in second.
The girls' anime series Pretty Cure is made by one of the leftist broadcasters and is leftist, so Pretty Cure is in fact one of the most leftist visual works in Japan! And it's cute! Also leftist is the original author's version of Gundam. The robots are cool, and many of the fans are right-wingers who love the military. I am a fan of all of them, but I think I am a leftist.

Documentary films rarely go viral, unlike anime and drama films, mainly because they can't spend $0.01 billion on them. A documentary film made with a lot of money in South Korea was popular for a short while. The BBC probably makes the most leftist reports and documentaries broadcast in Japan. This is not a typo. The BBC first reports on social issues in Japan, then the people involved hold a press conference in Japan, and the Japanese press follows suit, which is the general reporting style today. Sometimes it is the NYT or so.
There are many reasons for this, but anyway, there is no money for Japanese journalists. In the first place, there is no habit of investing in the end constituents. TV stations just want cheap and sellable programs, not news. These days, even disaster reporting often relies on YouTube videos. Except if it happens in Tokyo. Going to a neighborhood for news coverage is cheap. Incidentally, not particularly many right-wing news reports or documentaries are made because of the same lack of money.

The sentence sounds like a lament, but it's strange because it actually becomes rather enjoyable when the devastation is this bad. I was scared during the fall, though.
In short, all I have to do is read them all in English. Except for anime.
Political issues are basically 30 years too late in general, so I don't care about them in the extreme. I never miss a vote, though. The U.S. presidential election is still more important. Any adults know. Some people call this country a colony of the U.S., but it is not so colonized that it is at best a dependency state. Not even half of the economic news is translated to begin with. 90+% of programming information is in English. It's not uncommon to find Korean but not Japanese on the language selection screen today.

but somehow I like the way things remain kind of niche,

I'm in a similar boat.
I originally used Spaces on Command-R Plus without knowing it was HF, then a forum-mate was looking for a T2I model to improve a serverless mobile app, and I created an account on HF to help him with that. Of course, it was also to find the model I wanted for my own use.
Since then, it's become a means to an end, and I haven't advertised it properly. Well, unlike Reddit, or maybe even Reddit, it's so bad in comparison, and so overpopulated for so many reasons, that even if I advertised it, it wouldn't spread much.

I mean, isn't the whole world in a slump for anonymous forums? I like the fact that anonymous forums are not account-based, but because they are not account-based, they have been absolutely powerless against any kind of trolling or malice for over 20 years now.

Currently in the Japanese-speaking world, X is the most active in the anonymous and semi-anonymous world, despite the daily vindictiveness against the impression zombies created by Eron Musk and Eron Musk himself. The anti-generated AI is sitting on X, though, so the only generated AI is usually all on anonymous forums. Of course, I don't have an X account, or rather, I lost my ROM account that I just created somewhere, so I virtually don't have one.

What I've been wondering is how far the magic can go?

Well I envisioned a part of the present and a future where the code is written by a generative AI. Although I'm doing it in full manual, which I'm used to, just like you stick with Winamp.
But VSCode is the only new but great software. Just using it will take your Python and HTML editing skills up a notch. It points out and corrects grammatical errors, and the display is cool.
I guess Microsoft programmers made it for themselves. The quality of such software is truly amazing. Even if it's Microsoft software.

That aside, it's certainly something that can be accomplished with a full range of cloud services. In fact, it exists in this way normally.
https://huggingface.co/docs/huggingface_hub/package_reference/inference_client#huggingface_hub.InferenceClient
If you do this to your README.md, you can probably use it with Gradio's load(). However, the problem is that if you modify README.md, you can only choose either t2i or i2i for now. So if you actually want to try it, duplicate the repo and then modify it.
https://huggingface.co/Yntec/epiCPhotoGasm/edit/main/README.md

pipeline_tag: text-to-image

to

pipeline_tag: image-to-image

https://huggingface.co/spaces/huggingface-projects/repo_duplicator

Serverless i2i CPU Spaces already exist, albeit less so than Zero GPU Spaces.
https://huggingface.co/collections/John6666/spaces-for-image-to-image-video-66ab56d6d456f0408b9ff208
Even the content used in other companies' generated AI services, other than NovelAI, does not have its own proprietary core libraries. They are probably using Diffusers or modified versions of Diffusers internally, and the same thing can be done with Zero GPU space or more paid space or Endpoint API for basically anything.
HF is not a buyer, not a factory, but a developer, so basically anything can be done, but only multimodalart and a few others are willing to make an image generation AI space in HF official.
What a waste. mottainai.

Sign up or log in to comment