in

「未来を拓くAI技術:CIVITAIの安定した普及が禁止される理由」

civitai

CIVITAI BANNED STABLE DIFFUSION 3

CivitaiがSD3を禁止した理由は何なのか?AI技術の魅力を伝える記事

最近、CivitaiがSD3を禁止するという決定を下しました。この決定に対して多くの人々が驚いており、その理由について疑問符がつきます。しかし、このような出来事はAI技術の進化と普及において必要不可欠な一歩であると言えます。

AI技術は今や私たちの生活に欠かせない存在となっています。その魅力は、人間以上の計算能力や学習能力を持ち、膨大なデータを素早く分析し、新たな知識や洞察を提供することにあります。また、AIは人間の手では不可能だった複雑なタスクや問題にも取り組むことができます。

特にCivitaiのSD3は、画像処理技術の進化に大きな影響を与えてきました。Stable Diffusion for Beginners PlaylistやWeekly AI Art Challengesなど、様々なコンテンツがAI技術の可能性を示してきました。さらに、ControlNet tutorialやFamous Scenes Remade by ControlNet AIといった動画もその優れた性能を証明しています。

これらの例からもわかるように、AI技術は私たちの生活やビジネスに革新的な変化をもたらす可能性があることを示しています。CivitaiがSD3を禁止した背景には、その可能性への期待や不安が反映されているかもしれません。しかし、重要な点はAI技術自体の魅力であり、その進化が今後どのような展開を見せるか楽しみであり興味深い点でもあります。



動画はこちら

CIVITAI BANNED STABLE DIFFUSION 3 の画像

Written by Sebastian Kamph

コメントを残す

メールアドレスが公開されることはありません。 が付いている欄は必須項目です

GIPHY App Key not set. Please check settings

50 Comments

  1. This is the turning point for StabilityAI after losing Emad as their CEO… Now, StabilityAI will lean towards their investors instead of the community.
    And you know, I know, everyone knows, that if they tried this now and change it back to the way things used to be, they will try again but with a different angle.

  2. Yes, you said it, Sebastian. Civitai is a big player in this. Not sure about others, but without Civitai being on board with all that goes with that, I won't be bothered spending time experimenting with SD3. Civitai obviously knows they have some clout here, so this does put pressure on Stability. This is like the relationship between game publishers and big name gaming content creators… the two sides have a mutually-beneficial relationship. But if one side tries to push their power/control/authority too far, the other side might just walk away, even if doing so hurts them too in the short run. All makes sense, but that does mean that Stability will know they also have a vested interest in fixing this quickly. Give it another week or two, and I bet they will have refined their legal terms to allow the ban to be lifted.

  3. Here's the issue with AI, in general.
    I understand that it takes money to train models, and I would understand a system of donations.
    BUT these model are trained on public data on the pretense of "educational, no profit usage".
    I am sorry for those companies but as long as they don't have permission for every single image in the dataset, any output is by definition public domain and uncopyrightable.
    Anybody can claim ownership as much as he wants, but I will ignore it because it's BS.

  4. I honestly cannot wait till the internet is flooded with open source models and these companies are forced to take whatever scraps they can get….their entire business is built on a foundation of opensource and they can EASILY monetise their work in better ways…instead they pull restrictive crap like this unnecessarily. They know days are numbered as soon as training algorithms become more efficient and affordable ai focused gpus enter the market. Heck, companies like apple are making ai apps native and free even on low powered phones, baked into the os. SD are trying to make their money asap and bail.

  5. Funny how hypocrite model creators can be. Training models on billions of images without paying royalties to artists, photographers etc. But expect users to agree on twisted agreements. Generative AI is in my opinion only fair if the models stay opensource. Also still hoping for Midjourney to get sued into the ground.

  6. I’m glad this is done. They thought SD3 was great and they had the upper hand to demand all rights. Stable Diffusion community models and loras are what makes SDXL great. SD could be the WordPress / Linux of AI in the open source space. I hope it doesn’t end up like Magento which started as an open sourced e-commerce project to only fall into oblivion after being bought by eBay.

  7. Very interesting. I'm not sure how great Stable Diffusion 3 will be, but I've found that the best-performing model so far is actually version SD1.5. This model has impressed me with its speed and ability to generate decent images, especially when combined with other models through training and LoRas. In contrast, I've been underwhelmed by SDXL and rarely use it. Overall, SD1.5 remains my go-to choice for generating images which when upscaled are good. These current licensing issues are interesting.

  8. I'm mostly upset about the celebrity purge… faceswapping is separate from image generation and its so fcking easy and fast and jesus christ what sense does it make to purge celebrities? it just takes away the fun… and before faceswap we had photo shop. weve been face swapping for fcking 30 years, whatever, I guess this goes right along with a PORN ban… conservatives and dems are both dumb af on this

  9. The new SD3 license basically says use it in your closet or pay us, share it and they pay us too. The derivatives of derivatives rabbit hole can be a Mandelbrot and all of this is even funnier coming from a company that never allowed legal evaluation of its own dataset. While it may seem naive It is in fact regulatory capture… now SOMEBODY will have to start REGISTERING llms right? The Cisco of AI.

  10. I think the part that says derivative models can be recalled or destroyed open Civit to potential lawsuit down the line, both from SAI and from its users. It’s a colossal failure on SAI part even if license is “misunderstood”. If your $20 monthly subscription license takes a lawyer to understand, then the problem is your license.

  11. After comfyanon resigning from SAI and then coming forward with the statements he did (and I'm sure you read about that) I think you're being too charitable with your views on SAI. They're not what they were before anymore. Releasing a poisoned and unfinished model and writing the license the way they have has made it pretty clear they're not our friends. Civit made the correct choice and unless SAI takes some serious steps to fix things like rewriting that license, doing a better job of communicating with the community, and ideally releasing less broken and poisoned weights for the larger models, they're at the end of the road here. There are plenty of less predatory means of monetizing their models that tons of people have pointed out, larger one time fees for model license, or going the blender route, all much better options but instead they've chosen a really nebulous and unfair approach. It's a big shame. I'd be happy to give SAI my money for better open source models, but not under these circumstances.

  12. The concern outlined under 'the Danger' is legit. This is what happened with Monsanto.
    There was a farmer near where I live who had only heirloom crops that he raised himself, had 0 interest in GMO crops. However apparently at some point some Monsanto seed got onto his property and mixed with his seeds.
    He lost everything in a lawsuit because there was no way to distinguish what seeds were or were not descendants of Monsanto crops. Had to destroy all his seeds.

    That's what will happen to CivitAI if they don't take these measures now, based on the current best laymen understanding of the terms.

  13. Their license is unenforceable, not only due to the copyright office refusing to grant copyright to most AI-generated images, but also due to the fact that as time goes by SD3 images posted on the Internet are going to get scraped and incorporated into all of the AI systems of all of the other companies. Everyone should pretend that the license doesn't exist and use SD3 as they please.

  14. 🎯 Key points for quick navigation:

    00:00 🚫 Temporary Ban on Stable Diffusion 3
    – Temporary ban on all SD3-based models on Civii due to licensing clarity issues.
    – Existing SD3 models will be archived; no new models or derivatives allowed on Civii.
    – Legal uncertainties around SD3's license prompted Civii's cautious approach.
    01:09 📜 Licensing Concerns and Community Impact
    – Civii seeks official clarification from Stability AI regarding SD3 licensing.
    – Concerns over AI's control rights on SD3 models and their derivatives.
    – Community discussion on the legal and practical implications of SD3's usage.
    02:47 🌟 Value of SD3 Despite Restrictions
    – Emphasis on continuing experimentation with SD3 despite restrictive licenses.
    – Acknowledgment of mixed results in SD3 outputs; variability in quality observed.
    – Decision framed as protective of community and supportive of creative contributors.

生成AI検索サービス「サーチGPT」試験提供を開始…米オープンAI、グーグルに対抗

「生成AI検索革命:OpenAIの『サーチGPT』がGoogleに挑む!」

Fine-tuned model on GPT 4o-mini can't use vision - API

「最新AIモデルのGPT-4o-miniがビジョン機能を持たない理由とは?」