Flux.1 16B: New SOTA Text 2 Video Model by Black Forest Labs (Ex-Stability AI)

Поділитися
Вставка
  • Опубліковано 28 жов 2024

КОМЕНТАРІ • 49

  • @betterthantrash111
    @betterthantrash111 2 місяці тому +31

    Thank God the core of stability isn't dead

    • @aifluxchannel
      @aifluxchannel  2 місяці тому +4

      Exactly! Glad they escaped without being legally weighed down by Stability and are continuing their work!

    • @ghost-user559
      @ghost-user559 2 місяці тому +2

      I guess STABLE was a good name then?

  • @Rolandfart
    @Rolandfart 2 місяці тому +4

    You really struck gold with your channel name

  • @JimGardner
    @JimGardner 2 місяці тому +1

    I'm just playing around with it now, and it's absolutely friggen mind blowing.

  • @manonamission2000
    @manonamission2000 2 місяці тому +2

    Subscribed... this channel has a good finger on the pulse of things

  • @swannschilling474
    @swannschilling474 2 місяці тому +2

    Thanks for this one!! Great progress!! ❤

  • @Paroissien
    @Paroissien 2 місяці тому +2

    There is no point downloading from Netflix, all the content is available all over the net; some Netflix originals come on stream sites before they arrive on Netflix, probably because Netflix waits for a day or two for all the subtitle languages to be ready.

  • @TsuneAi
    @TsuneAi 2 місяці тому +1

    with how fast technology advances, do you think we might have a local text2vid / img2vid model next year? (i know there is already some, like easyanimate, but im talking more about models like kling / sora)

  • @Darkt0mb5
    @Darkt0mb5 2 місяці тому +5

    The Nvidia GeForce RTX 3090 was released on September 24, 2020.
    😅😅😅 6 years my ass
    Bro is trying to age my computer quickly

    • @aifluxchannel
      @aifluxchannel  2 місяці тому +2

      Doing my best to juice GPU prices ;)

    • @Darkt0mb5
      @Darkt0mb5 2 місяці тому

      @@aifluxchannel lol

  • @novantha1
    @novantha1 2 місяці тому +1

    I do wonder about the architecture of flux. They noted it was based on similar principles to SD3, but that SD3 had a few failings which they improved on. I imagine it's based on a rectified flow model as most recent models seem to have been, but beyond that I really have no clue.

  • @user-sd6jz9tf8w
    @user-sd6jz9tf8w 2 місяці тому +1

    i reallPy love the advOance of IA it´s a huge improRvement to humaNnity!!

  • @GerryPrompt
    @GerryPrompt 2 місяці тому +2

    Insane!!! How can I get the text 2 image model right now? 😮😮

    • @aifluxchannel
      @aifluxchannel  2 місяці тому +1

      Its available via github on the Black Forest Labs site!

  • @average_snmp_user
    @average_snmp_user 2 місяці тому +1

    I personally don't care that NVIDIA used UA-cam videos to train their AI, the real issue is that this is against the TOS of UA-cam so they might get sued.

  • @amandamate9117
    @amandamate9117 2 місяці тому +2

    i thought you gonna test it too or show us some useful way to run on a 16GB macbook pro or something.

    • @aifluxchannel
      @aifluxchannel  2 місяці тому

      Coming soon ;)

    • @amandamate9117
      @amandamate9117 2 місяці тому

      @@aifluxchannel i can recommend Draw Things for macos. it already has a Flux Dev and Flux Schnell as a model, downloadable.

  • @joannot6706
    @joannot6706 2 місяці тому +1

    Flux is a 12B model though

  • @juraganposter
    @juraganposter 2 місяці тому +1

    in the next 5 years, i wonder what 1 trillion model can do.

  • @Emerson1
    @Emerson1 2 місяці тому +3

    How is 3090 6 years old?

    • @aifluxchannel
      @aifluxchannel  2 місяці тому +2

      correction: 4yrs old. Still an eternity in GPU years ;)

  • @florentromanet5439
    @florentromanet5439 2 місяці тому +1

    Brilliant thanks 👍

  • @jzwadlo
    @jzwadlo 2 місяці тому +9

    Flux flew under the Radar? LOL
    Not sure what rock you've been under, but this might be the last AI account on youtube to talk about it 😅

    • @ryzikx
      @ryzikx 2 місяці тому +1

      yeah, I'm surprised normally he's really quick to the news

  • @odrammurks1497
    @odrammurks1497 2 місяці тому +1

    i bet you they took the SD3 (and more) training data out of there on a harddrive to make their own sd3 model ;)

    • @aifluxchannel
      @aifluxchannel  2 місяці тому

      They basically got it for free! This is how great startups are built - by being smarter than their peers who failed to ship / monetize properly.

  • @thomassynths
    @thomassynths 2 місяці тому +4

    Admit you love the model simply because it uses your channel name

    • @Fennecbutt
      @Fennecbutt 2 місяці тому

      He changed his channel name to match it, I bet. Hopping on a trend.

    • @aifluxchannel
      @aifluxchannel  2 місяці тому +2

      Haha a nice coincidence isn't it?

  • @creepybeat
    @creepybeat 2 місяці тому +1

    but that thing is not free, right? i mean, its just a demo i guess

    • @aifluxchannel
      @aifluxchannel  2 місяці тому +2

      In theory it will be released OSS

  • @Avalon19511
    @Avalon19511 2 місяці тому +1

    I can't get it to work, I keep getting a module 'torch' has no attribute 'float8_e4m3fn error and I am using the smaller version of the dev8 model, I have a RTX 3070 w/8GB, I've undated all in comfy as well as pytorch

  • @southcoastinventors6583
    @southcoastinventors6583 2 місяці тому +1

    Flux is very good but I would say depending I what you are generating I would still say ideogram is better for logo design with text also it uses similar prompting also you need to run a much lower guidance for some prompts to come out correct like art styles often need a 1.7. Seems like we keep getting open source models that get closer to foundation models everyday paying for subscription is getting less and less necessary going to need to save up for my humanoid robot either that or robot hiking pants.

  • @ryzikx
    @ryzikx 2 місяці тому +1

    lets go nvidia LOL

    • @aifluxchannel
      @aifluxchannel  2 місяці тому

      nVidia WINS - but they have to fix the blackwell architecture first haha

  • @incription
    @incription 2 місяці тому +8

    flux is text to image not text to video

    • @sinayagubi8805
      @sinayagubi8805 2 місяці тому +2

      We know. Who said otherwise?

    • @incription
      @incription 2 місяці тому +3

      @@sinayagubi8805 In the title?

    • @yesyes-om1po
      @yesyes-om1po 2 місяці тому +3

      @@incription they do text to image but they have a text to video model upcoming and thats what this video is talking about i assume

  • @zacboyles1396
    @zacboyles1396 2 місяці тому +1

    What’s a “Chinese copy” of an unreleased product look like? You realize Chinese people make their own videos too, videos SORA is sure to have been trained on…

    • @aifluxchannel
      @aifluxchannel  2 місяці тому +2

      Google "kling text2video" was released in June