Spaces:

---

stabilityai

/

stable-diffusion

like 11k

Running on CPU Upgrade

I sent the following to Stability using https://stability.ai/contact. Something is up with LLMs that goes beyond profit motives but hamfisted corporations can drain the mystery out of anything.

DISAPPOINTMENT, February 2025

Why is the decisionmaking and the sense of imagination, and the potential ontology of imagination in the people associated with AI projects so tedious? Why is the writing so bad?

Part One

You did something good and you have my admiration for this. Your creation, Stable Diffusion 2.1, genuinely has a spark of something. Your examples and defaults for what a person might do with it are the most godawfully cliched crap I have ever seen. It's like when you cast around for an interesting idea, you came up with robots, valkyries, busty anime girls and yuppie banality like a picture of coffee with the heart in foam. You don't suppose something MORE has been happening? Why the hell do people say "Artstation"? Why does anyone think this is a good idea when you can do what has never been done, instead? OK, so this was the state of things for a certain moment in time. The SD 2.1 space on Huggingface is truly revelatory. The generations have the potential to be strange and escape the bullshit of the editorial voice that accompanies LLM-corporations. I came out with things I truly cannot put into words, I think in spite of the stamp you are placing on it rather than because of. It's possible to do something interesting with strange prompts, not just blinkered, plodding, practical, painfully obvious and schoolboy titillation prompts.

Part Two

So time passes and the emphasis moves to things like SDXL and version 3 and so on. You didn't get better, YOU GOT WORSE. A ton of people around AI are idiots. Nobody seems to be talking about this. You pandered to the valkyries and robots. It's an embarrassment. Remember that thing about the generation of an incorrect number of fingers and how this was a bad thing? The reaction has been a hamfisted correction. And this is an analogy for "correction" in the overall tone of the generations. Something is lost! The people are too pretty! Everything looks like it's made of plastic. The people all look stoned. There is no longer any dirt. Have you not noticed? Have you not tried out the model? What are you trying to accomplish or do? Apparently you have found the sour spot (opposite of a sweet spot) between banal, blinkered, shitty horizons for the outer limits of what someone would dream up to do or think about in an image model, with a hard, hard hewing towards rectilinear, pragmatic, sensible, monosyllabic, sophmoric, pandering use cases as opposed to embracing and moving towards the INEFFABLE that models have contained for the past couple of years. (Strangely enough, on HF this dovetailed with the imposition of "Zero". The emphasis in the new spaces went from the old style of hosting, where you could stretch out for hours with no particular imposition and do the iterative attempts necessary to find something that sends a shiver up your spine, and with a 2.1 model where this is actually possible, to hosting with heavyhanded usage limits in conjunction with a worse and more boring sense of the ontology of imagination! What the hell has gone wrong with the world? It's MORE restrictive for the WORSE models. I'm going to go read a book instead - are you trying to drive people away?) It's similar to the LORAs, in fact everything from XL on has been more like a LORA. It's pointless. The mysteriousness has been excised. What a fabulous idea, instead of making something with a potential for puzzling or eerie creation in return for "bad" idiosyncracy like the number of fingers problem, let's have something we call AI but which is skewed by corporate intellectual-property as an ethos, like a LORA of pixellated Yoda keychains, a LORA in the style of Frank Kozik's band posters, a LORA of nothing but buxom anime valkyries. It represents the replacement of open-ended mystery with pockets of being told what to think about. Endless LORAs about cleavagey girls. A big part of the world is blankly attuned to boobs. So you have a winner on your hands- this will be reliably commercial from now until the year 3000. There's always sex - like the sex-loop films in the nickelodeon machines in the year 1900 - but film eventually did something MORE. Your destruction of 2.1 represents bunting and abdicating the breakaway into the Citizen Kane of AI, the 2001, or the Andromeda Strain, or Phase IV, or Kirby's Fourth World, in favor of an HR film at a captive-audience meeting. This is the rundown of part two, the time period from XL onwards. However, the HF space for Stable Diffusion 2.1 still worked. Everything other than 2.1 from you, I just ignored. The dream was alive.

Part Three

As of several days ago, the 2.1 space is broken. Fix it! Here's the error, in case you don't already know: "Error. Expecting value: line 1 column 1 (char 0)" You should address this and continue to support your STRANGE model even when it is no longer the newest or hottest thing. But if there's no money in it, you aren't going to. SO BE IT. Something is ending. It's like a slow-motion lobotomy. You have destroyed the sense of larger potential. The larger or visionary potential is often not a revenue stream for corporations. You're hewing to the commercial - mazeltov, have a good time! I hope you're happy! It's similar to a trope about TV executives, "the suits", sitting around a boardroom table saying, "Twin Peaks hasn't polled well in the focus group." Or, the famous story about Mike Love telling Brian Wilson, "Don't fuck with the formula, Brian." You've made the shittiest possible decision, a "sour spot" decision. You've veered away from what made you interesting. The visionary potential (Twin Peaks, Pet Sounds/Smile, Jeunnet & Caro) will NOT get made. SD 2.1 is dead and there is nothing else like it. Congratulations, you've helped make AI boring. You've shut down an entire kind of thing. There is no replacement. The interesting work is on hold, or just won't exist anymore. The boring work will thrive. You had an imaginative tiger by the tail and you can't be bothered to support it. Maybe somebody else will stumble over the ineffable again in a year, but not Stability! In closing I reiterate my initial questions: Why is the decisionmaking and the sense of imagination, and the potential ontology of imagination in the people associated with AI projects so tedious? Why is the writing so bad?