AI Workflows that Accelerate Production & Respect Artists' Process (GDC 2024)

Поделиться
HTML-код
  • Опубликовано: 15 ноя 2024

Комментарии • 33

  • @TheSergeBird
    @TheSergeBird 7 месяцев назад +5

    Fantastic talk. I can tell that one of the goals of this talk was to refute naysayers and appease the creatives that have anxiety over AI. I can only hope that those people actually listen to you guys and adopt a more positive outlook on AI art as a creative tool.

  • @AndyHTu
    @AndyHTu 3 месяца назад

    Enjoyed this interview a lot. Thanks for the constant updates and videos Hippy ! :)

  • @valentinepavliuchenko7309
    @valentinepavliuchenko7309 6 месяцев назад +11

    Artist should start adding clauses to contracts to disallow training AI on their stuff. Or better yet allowing Nightshading their stuff. Otherwise dude like that one will always prefer prompting Peter Mohrbacher style instead of hiring Peter Mohrbacher.

    • @invokeai
      @invokeai  6 месяцев назад +2

      Artists should definitely be training their own models and thinking about how to monetize those official models. We'd love to see a future where this becomes a new revenue stream for artists (eg. granting controlled access to your official model with all the licensing arrangements in place so a business can use them and artists get paid for their use).

    • @valentinepavliuchenko7309
      @valentinepavliuchenko7309 6 месяцев назад +7

      @@invokeai Ah, I want to believe this rosy picture of artists getting nice passive income streams for their models. But sadly what I find more likely and what has happened already and is happening still is that folks just scrape the artworks from the web and use the artists' style for free. Once it's trained, can't prove what's in the black box, right? So Nightshade and Glaze and legal action - that's the way to go

    • @blurtam188
      @blurtam188 6 месяцев назад

      Good luck with that. But the models are already trained - all that's happening now is that their being better executed.

    • @djinghiskhan9199
      @djinghiskhan9199 5 месяцев назад

      @@valentinepavliuchenko7309 If you produce art, your first concern should be people seeing and appreciating your art not the money you can generate from it. This is why Hollywood is garbage.

    • @valentinepavliuchenko7309
      @valentinepavliuchenko7309 5 месяцев назад

      @@djinghiskhan9199 Ah, one of my favorite arguments of AI enthusiasts. Yes, sure, artists don't need to earn money for their work. Don't need to eat too. Just need "appreciation", preferably in the form of stealing said work and training AI on it 😀Nobody needs money anyway, right? Just a satisfaction of a job well done. We should get in touch with Apple and tell them to send us all an iPhone for free. They will have the satisfaction of us using and appreciating it after all. Man, why didn't I ever think about it?

  • @minyaw1234
    @minyaw1234 6 месяцев назад

    Do you support tiled diffusion, tiled vae and controlnet tile resample for a high resolution upscale with details?

  • @Javan_lima
    @Javan_lima 7 месяцев назад +2

    Amazing job!!!

    • @invokeai
      @invokeai  7 месяцев назад +1

      Thank you!!

  • @b.s.design
    @b.s.design 7 месяцев назад +3

    Thank you very much for your work!

    • @invokeai
      @invokeai  7 месяцев назад

      You're welcome!

  • @TR-707
    @TR-707 7 месяцев назад +5

    i just know that when i train a lora on my own stuff then apply it onto a base or tuned model checkpoint and then show it to my mom, i cant find the gall to lie to her face and say i drew that. And once i start explaining how it was generated, thoughts of doubt, lack of interest and dismay enter her and my mind and in the end its just depression and slop.
    Maybe this can help a totally ailing industry that is now resorted to churning out AAAA games that noone buys because they are uncreative and lack the human touch and understanding how to reach their audience.

    • @invokeai
      @invokeai  7 месяцев назад +2

      @TR-707 - Thanks for sharing your concerns. This part may not have come across clearly enough in the talk (and we certainly didn't show an end to end workflow), but most professionals are still drawing the core concept and overpainting/compositing outputs, even when AI is involved in the middle of the process. The more effective LoRAs just make the generation process a more useful, controllable, and consistent tool for that purpose.

  • @education7281
    @education7281 5 месяцев назад

    In your talk, you mentioned that many of your clients believe it's acceptable to train AI using images from the internet, assuming it falls under fair use. This implies that any artwork can be used to train AI models without consent, as existing data scraping practices aren't addressed by the lack of specific clauses against it. Consequently, the only legal progress for artists now is to label new work explicitly as not for AI training on their portfolios and websites to ensure some level of legal protection. However, any of their artwork that has already been used to train AI models cannot be removed from datasets unless they individually opt out from every AI model that used their work without consent.

    • @invokeai
      @invokeai  5 месяцев назад +1

      It's important to differentiate between foundation models and specialized models. Our views are outlined here: www.invoke.com/post/artists-generative-ai-evolving-landscape-copyright. The primary use case we see with Invoke today with model training are studios training LoRA models using their own proprietary content.

  • @thomaskrogh1244
    @thomaskrogh1244 2 месяца назад

    This is can't be the gdc talk. I'm looking for but good anyone can point out the gdc talk about ai use to improve the production pipeline. Thanks.

  • @illusionista
    @illusionista 6 месяцев назад +1

    What is the base model of this tool? I am guessing it's Stable Diffusion and in this case it's full of copyrighted works. Fine tuning is just another layer on top of it.

  • @kcoppa
    @kcoppa 7 месяцев назад +2

    Love this

  • @SonnyNguyen
    @SonnyNguyen 7 месяцев назад +2

    Congratz! Can't wait for the future updates! This tool is very very potential!

  • @USBEN.
    @USBEN. 7 месяцев назад +3

    Good to see this tech reaching to wider audience.

  • @luxuriousmindset1906
    @luxuriousmindset1906 7 месяцев назад +3

    This is great

  • @bcraigcraig4796
    @bcraigcraig4796 7 месяцев назад +1

    You said it was open source but I see you need to pay to use it sorry confuse

    • @invokeai
      @invokeai  7 месяцев назад +2

      Open source can be found in the "Community Edition" link on our site, or by searching for Invoke on github!

  • @beaverson
    @beaverson 6 месяцев назад +1

    Naaaa... F×ck "AI" generated content. Using this tech to do your artwork for you just kills the integrity and respect for artists more in society. I mean, just look how unethically this bs was created. It won't be used as a tool for long, just a future replacement that also creates distrust within the art world.. I'll keep making my own projects without "AI" sludge, thank you.

  • @djinghiskhan9199
    @djinghiskhan9199 5 месяцев назад

    So-called artists demonising AI are the kind that put profits before art and their art reflects that. I'm an Artist that doesn't want commercial art to be a little exclusive club. I want the world to create art with ease and accuracy, and it all should be free. Thankyou Invoke for remaining opensource.

  • @danielkaiser4412
    @danielkaiser4412 7 месяцев назад

    Can you make a video showing how to make AI generated images for print? So images that are not just 1024x1024 or expanded via canvas, but also have a resolution of e.g. 4096x4096 or other aspect ratio, but do not change the original image significantly?

    • @invokeai
      @invokeai  7 месяцев назад +1

      Noted! We support ERSGAN upscaling workflows now, but we'll be making improvements to upscaling in the next couple months. Keep an eye out!

    • @danielkaiser4412
      @danielkaiser4412 5 месяцев назад

      @@invokeai Workflow beta with some fine tuning it is. Might be something for a livestream. Because the upscaling and refining already works very well in the beta. And would take Involke AI to the top for creatives. Adobe Firefly doesn't produce 20k images 😉👍

  • @tomschuelke7955
    @tomschuelke7955 7 месяцев назад +1

    It's realy a great Artist friendly tool, far ahead of the one button tools like midjurney.
    Never the less the images depend strongly on the incorporated Ai models.
    For Art, there's allready great stuff out there, but in my case we still search for models fulfilling our hopes for Architecture.
    In some cases with smal rooms and all day private home architecture it's allready ok, but when it comes to big, international Architecture, you soon understand, that this is not yet realy trained to any moddeln. The degree of architectural style fine-tuning and detail quality still is unreachable apart of first ideas.
    Never the less the tools invoke inkorporates to iteratively improve your Ai images are secound to none.
    So invoke deserver a much bigger comunity. It's realy a well thougt user friendly concept

    • @invokeai
      @invokeai  7 месяцев назад

      Thanks for the nice words! We know a lot of architecture firms have been using Invoke and it will be great to see how the model ecosystem develops.

  • @Bloodsong13T
    @Bloodsong13T 7 месяцев назад +2

    photography is going to kill art! who needs a portrait/landscape painter when you can just take a super accurate true to life photograph?
    digital art is going to kill art! you kidz with your undo! and layers! and non-destructive workflows!
    ai art is going to kill art!
    ...it's just another tool in the shed.
    if you're very worried about having a career as an artist in the future... hey, in my day, "NFTs" were called "oil paintings"! Real Art! the original NFTs!