STOP Using LORA'S, Use THIS Instead || Train Flux Models With Lockr

Поділитися
Вставка
  • Опубліковано 17 лис 2024

КОМЕНТАРІ • 39

  • @gueberninja
    @gueberninja Місяць тому +24

    I COULD'NT SEE !!!! BECAUSE YOU PUT THE ENDING VIDEOS TO WATCH AT THE END !!! RIGHT ON TOP OF THEM!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

  • @ImAlecPonce
    @ImAlecPonce Місяць тому +16

    So basically, stop using loras, so you can use a loras. well, ok

    • @hendrysconovianus8546
      @hendrysconovianus8546 Місяць тому +2

      i'm also confused at first but i asked GPT then whats the difference between LoRa'S? it's modular system... which is if you are train to certain part of object (let's say a head) it's more executeable later for full body shot. other than LoRa, you also need to create a dataset with half body shot & full body shot... for examples, in LoRA if your dataset only created for head/close up picture it might seem the generated image will incompatiable with the results...
      that's for my opinion, but please correct me if i am wrong

  • @m.salerno
    @m.salerno Місяць тому +3

    hey mate, thanks for the video. Do you beleive that creating images on tensor using flux is better than creating with Fooocus?

    • @brainprojectx.
      @brainprojectx.  Місяць тому +3

      Yes I think so
      Flux images are more realistic

  • @alexvoorhees
    @alexvoorhees Місяць тому

    Awesome! I see my *Lyra Echo* model in TensorArt - amazing!

  • @FilmSpook
    @FilmSpook Місяць тому +1

    Big Thanks, again, my Friend! Your info is pure gold. Cheers

  • @gumvue.studio
    @gumvue.studio Місяць тому +1

    thank you for the video, questions, can I get generate loras here and download them and use localy?

    • @brainprojectx.
      @brainprojectx.  Місяць тому +1

      Yes you can

    • @gumvue.studio
      @gumvue.studio Місяць тому

      @@brainprojectx. yes,so I can train and sell the loras? is that something ok? I mean, do we have any copyrigt or commercial restrictions?

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      @@gumvue.studio lol never even thought about that, i'll do a research

    • @gumvue.studio
      @gumvue.studio Місяць тому

      @@brainprojectx. yes, cose its like a huge deal , if can create set up characters.. and seell them, I have gpts that can hundle the creations of story and characters. so its like all go in one pack. but not sure if then not get problem wit this kind of generators, that website .. maybe not alow to sell the loras, not sure.

    • @TomSmith-yh9ju
      @TomSmith-yh9ju Місяць тому

      I downloaded every 10 Lora Steps of my SDXL Lora, and used them on fooocus. If you work with low VRAM, and get the info ... loading in LOW VRAM mode, that could mean, that not all the layers of your Lora are being loaded, and you get different results. The 440 MB TA SDXL Lora, vs. the 220 MB civitai SDXL Lora is a question, based on your graphiccard. Thats the reason, that you run first on TA online. 1 Generation of 1 pic cost 8 credits, thats 4x Thank you on someone else pics on TA.

  • @WhySoBroke
    @WhySoBroke Місяць тому +1

    I tried it but is super slow... more than 8 hours!! Just train regular Flux Lora on Civitai in minutes

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      Hmm, yh but the results are worth it

    • @TomSmith-yh9ju
      @TomSmith-yh9ju Місяць тому

      on civitai, I did a SDXL Lora with about 30 pics, and get a Lora with 230 MB - With TA, I did a SDXL Lora with 15 pics, and get a Lora with 420 MB.

  • @Beauty.and.FashionPhotographer
    @Beauty.and.FashionPhotographer 29 днів тому +3

    Clickbait title and useless video,.... he is selling a paid online service.... this can be done for free locally in so many, many ways...

    • @brainprojectx.
      @brainprojectx.  28 днів тому

      Yh but you know
      Not everyone has high end gpu
      Or you want to give them yours??

    • @Beauty.and.FashionPhotographer
      @Beauty.and.FashionPhotographer 28 днів тому

      @@brainprojectx. no need for a high end RTX at all, as you suggest here..... i run stuff on my mac with Pinokio...works just fine...free .....

  • @Bigmac_cat
    @Bigmac_cat Місяць тому +1

    Just so yall don’t waste yalls time you have to pay it is not free

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      How??
      The literally give you free credits everyday

    • @Bigmac_cat
      @Bigmac_cat Місяць тому

      @@brainprojectx. when I try to do this and I press start training now it says “you are not a Pro member yet, enjoy membership privileges after activation”

    • @Axel.0-0
      @Axel.0-0 20 днів тому

      the website now down lol

  • @javimar69
    @javimar69 Місяць тому

    Awesome stuff!! Is there a way to run these model form replicate?

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      Hmm I'm not sure
      It's tensor I use though

  • @SoundChaser_
    @SoundChaser_ Місяць тому +1

    3:58 ... what I see are ads all over the screen. Failure

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      Ads Lol??
      Did you see well?

    • @nitsharma
      @nitsharma Місяць тому

      @@brainprojectx. He means the end screen overlay appears at the end of the video, so we can't see the final output.

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      @@nitsharma Ohh, i understand

  • @mal-avcisi9783
    @mal-avcisi9783 Місяць тому +1

    are you joking with us ?

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      What do mean?

    • @mal-avcisi9783
      @mal-avcisi9783 Місяць тому

      @@brainprojectx. are you jokin wit us ?

    • @brainprojectx.
      @brainprojectx.  Місяць тому

      @@mal-avcisi9783 lol

    • @TomSmith-yh9ju
      @TomSmith-yh9ju Місяць тому

      it can cause frustration, if you see the training-results, and all trainingsteps are not so close to the own dataset. The main problem is, to exclude the pics from your dataset, that cause the main difficulties. For TA, I assume, forget funny mimic, forget bodyshape, forget upscaled images, and always ... try again. If you know the old face movie of picasa software, this face morphing between the pics of your dataset, and other persons, personally, my female Lora ended up with Rihanna eyes in 90% of generations, and only 5 % came close to my dataset. Also male attributes become dominant, if I choose a full body perspective. ... so, ... lets say, its all about the selection of the dataset pics.

  • @quizcartel.
    @quizcartel. Місяць тому

    Bro, can you make a video on how to make an Instagram reel of an AI influencer go viral? I'm uploading reels of my AI influencer, but they're not getting much attention

    • @brainprojectx.
      @brainprojectx.  Місяць тому +2

      oh soon bro, ive been too busy on youtube
      stay tuned, i found a mentor to work with