Why is stable diffusion not as simple as dall-e?

I don't get it. They seem to be going in the wrong direction. They need to focus on making it as easy to use as stuff like dalle or midjourney instead of this tinkering tool. Sure, have an option for tinkering for die hards but also make it easy for the rest of the people.

No one wants to deal with models, loras, vae, texture embedding, checkpoints, negative prompts and like 40 tags to get a good result and all that other shit. Most people don't want to do that shit. I don't want to have to write 40 tags and an entire fricking bible of negative tags just to get a result similar to what one sentence gives on dalle.

You can say "skill issue" or "you're just tech illiterate " but I and most people do not give a frick. Ai is a tool that is designed to make things as easy and fast as possible. If I wanted to put in the effort I'd rather become an actual fricking artist. The reason ChatGPT got 100 mil users so quickly and spread like wildfire in schools and stuff is because of how easy it was. You go on the website, type "write me an essay for ___" and click enter. Boom. It's done. Imagine if chatgpt required the same set up as stable diffusion does. Not even 1% of people would use it. Actually, chatgpt has 100 million users. 1% is way too high. Maybe 0.01% would use it.

SD bros, either streamline the experience and make it very simple to use or someone please try and steal the dall e 3 code

Homeless People Are Sexy Shirt $21.68

Black Rifle Cuck Company, Conservative Humor Shirt $21.68

Homeless People Are Sexy Shirt $21.68

  1. 7 months ago
    Anonymous
  2. 7 months ago
    Anonymous

    Stable Diffusion is the model and system underneath the open source Web UI, the only reason why the GUI is complicated is because that's how it was set up by a random guy, you fricking moron homosexual.
    There are simpler GUI where you just push a button, baboon. Search for it.

  3. 7 months ago
    Anonymous

    >models, loras, vae, texture embedding, checkpoints, negative prompts and like 40 tags
    It's either that or you buy a GPU that has 1tb of VRAM

  4. 7 months ago
    Anonymous

    [...]

    Pictured: Microsoft employees "retiring" MariaDB on Azure in favour of Oracle MySQL

  5. 7 months ago
    Anonymous

    As always, thank freetrannies.
    They might be able to code powerful software, but they will never, EVER, make anything actually usable (see: Linux desktop) or also be women.

  6. 7 months ago
    Anonymous

    As a veteran StableDiffusion user, it's a lot of things. It started out that simple and slowly grew more 'complex' over time. Why? Because it fell on the community to improve it, and it basically turned it into modding skyrim. You have negative prompts to guide it away from undesired outputs, you have LoRAs to guide it towards styles/characters the base model doesn't recognize, Controlnet to attempt to contort it into poses it doesn't understand, various vaes to try and balance deep-frying/saturation, various models for anime/realistic since it's not smart enough to do both, etc.

    The reality is StableDiffusion, as a model, just isn't good. It's just that people have been milking everything they can from it. If anything this shows just how insanely powerful a locally hosted Dall-E 3 could be. StabilityAI has a company has underdelivered compared to their peers. SD 2.x was dogshit and SDXL is also underwhelming. The dataset is less-than-stellar, the prompt comprehension is a joke, the aesthetic is bleh, NSFW relied on the proprietary NAI leak, and the censorship of their new base models has put the community off from working on stuff for them. There isn't much the "SD Bros" can do besides make it even more complex, because they're limited by the shortcomings of the base model. The simplicity of Dall-E prompting speaks to how exponentially better it is a a model.

    Then there are the empty drifters desperate for an identity who latched onto the idea of engineering prompts, of tinkering with knobs and levers being a good thing. They worship the idea of unnecessary complexity and absolutely seethe over the fact that Dall-E gets better results with less tweaking. This mindset has seeped deep into the community where rather than becoming a better AI, SD has become a 'toolbox suite' with hundreds of extensions that still get you the same easily identifiable melted slop in the end.

    tl;dr- even if it was streamlined, stable diffusion won't look better than dall-e

    • 7 months ago
      Anonymous

      Yeah, pretty much what this guy said.

      >who latched onto the idea of engineering prompts, of tinkering with knobs and levers being a good thing
      It's fun to mess with things just like it would be with anything else. Sure, I'd love it to read my thoughts and create things 1:1 right away, but all the tools, extensions, plugins, LORAs, etc make it fun to experiment with. I don't know how people could be on BOT and not enjoy the "I wonder what would happen if I do this and this and then hit this button" experimenting.
      The generals suck, though, so avoid those.

      • 7 months ago
        Anonymous

        The issue isn't with the tools, it's with the AI becoming increasingly reliant on tools and that being viewed as a good thing. Dall-e without tools is still way more powerful than SD with all the extensions in the world. There are some things SD just can't generate no matter how much you try and guide it. You'd be spending so much time in photoshop and inpainting that you might as well draw it yourself. Base model trumps all

        • 7 months ago
          Anonymous

          That's true, although I mostly use it for fetishistic 2D porn so there's an obvious advantage there. Right now people are enjoying DALLE3's test drive, but I wonder about the legal challenges ahead of it as you can generate copyrighted stuff and companies with armies of lawyers like Disney will be eager to wage scorched earth against other corporate giants.
          I see it going the same way the text stuff is with more and more filters added on to it which will lead to degradation with false positives and other unintended effects, causing local models to become comparatively stronger in their niche usage. They won't win with pure quality, but with capabilities that are artificially removed from the otherwise technically superior product.

  7. 7 months ago
    Anonymous

    >You can say "skill issue" or "you're just tech illiterate " but I and most people do not give a frick.
    StabilityAI does also not give a frick about you and all the morons that are throwing temper tantrums 24/7 because SDXL would not generate naked anime toddlers for them. It is a free research product with a permissive license, they really could not care less if millions used it or nobody used it. You are free to use DALLE 3, oh never mind, you ran out of points or got banned again and are now on your 8th account. DALLE 3 will never be leaked for the same reason GPT-4 will never be leaked, good luck trying to run the model locally.

    This is the same as looking at footage of Microsoft Flight Simulator and getting all excited over it, only to then get frustrated when it actually requires you to learn how to fly an aircraft and know ATC communication if you play multiplayer.

  8. 7 months ago
    Anonymous

    >No one wants to deal with models, loras,
    I do. How else can I get it to generate faces its never been trained on?

  9. 7 months ago
    Anonymous

    but I like SD the way it is. as soon as they package it neatly, and make it normie-friendly, I will lose all interest in the tech. enjoy the ride my friend, it won't last forever.

  10. 7 months ago
    Anonymous

    >add as many controls as possible to fine tune everything
    >require near-expert knowledge to create anything good with it
    >great degree of control, enable maximum customization, steep learning curve
    Programmer GUI
    >just a prompt
    >tuned to create impressive things from prompts that are as simple as possible
    >no control or customization a tall, even a kid could use it effectively
    Marketing money maker GUI

    Both are shit.

  11. 7 months ago
    Anonymous

    >waaahhhh I'm too lazy to learn how to use the most easily customizable and uncensored image generation program that will never hit you with "unsafe content" warnings from big corpo jannies
    Not our problem that you have Black person-tier intelligence. Tinkering with settings is exactly what you need to do if you want maximum control over your generation.

  12. 7 months ago
    Anonymous

    because dall-e was designed for consoomer normies whereas the webui was designed for coomer booba enthousiasts

  13. 7 months ago
    Anonymous

    Try this https://github.com/lllyasviel/Fooocus

    • 7 months ago
      Anonymous

      Promising, will use once BOT jumps on it.

  14. 7 months ago
    Anonymous

    It's the same as any open source project or GNU/Linux.
    Endless bickering and in the end deciding on including everything.
    It always lacks someone saying no to things and scrapping things already in development.

  15. 7 months ago
    Anonymous

    >I don't get it. They seem to be going in the wrong direction.
    No dude. You WANT dongles so you can actually get what you want. If Dall-E 3 is giving you exactly what you want you might be a brainlet.

  16. 7 months ago
    Anonymous

    i think its a pretty nice interface. most of the options a first time user would care about are pretty in your face. all the really weird shit is under tabs.
    if youve installed mods on a computer game you can definitely find your way around.

  17. 7 months ago
    Anonymous

    >rowing out of round

    • 7 months ago
      Anonymous
  18. 7 months ago
    Anonymous

    sd bros hold me

    • 7 months ago
      Anonymous

      you should take a shower first

      • 7 months ago
        Anonymous

        God damn that is dogshit

        • 7 months ago
          Anonymous

          so is your mom

Your email address will not be published. Required fields are marked *