Train STUNNING LoRAs Locally inside ComfyUI! Low VRAM Flux Tutorial

แชร์
ฝัง
  • เผยแพร่เมื่อ 24 ธ.ค. 2024

ความคิดเห็น • 46

  • @bartosak
    @bartosak หลายเดือนก่อน +1

    I have just rent a GPU and started looking for this kind of tutorial - THANK YOU!! I love how you slowly explain everything

    • @Aiconomist
      @Aiconomist  หลายเดือนก่อน +1

      You're very welcome!

  • @deyakash2504
    @deyakash2504 2 หลายเดือนก่อน +1

    Hi, small products yes we can get great outputs.. Have you tried training a more complex product? Like a bike or a car? Thanks for sharing your learnings :)

  • @anseltrust
    @anseltrust 2 หลายเดือนก่อน +3

    Looks fine, can we train face models on this one like is it working same as the fluxgym?

  • @NoFaithNoPain
    @NoFaithNoPain 2 หลายเดือนก่อน +2

    My question is: If you can have consistant characters, how are consistant outfits achieved? Would it be though a LoRA (so eg you want her in a Space Marine Suit occasionally) wold you make the LoRA and plug it into the consistant AI workflow?

    • @Aiconomist
      @Aiconomist  2 หลายเดือนก่อน

      Yes you can use two LoRAs, one for your consistant character, and the other trained for your outfit. Just slightly reduce the strength of the outfit's LoRA.

  • @lithium534
    @lithium534 2 หลายเดือนก่อน +2

    What video model or service did you use for the last section where they were animated?
    Thanks for the content.

  • @АндрейБогаев-ы2я
    @АндрейБогаев-ы2я 2 หลายเดือนก่อน +2

    Good Thx. Please tell us about mixing LoRa in one stream (Mix LoRa)

  • @rmeta3391
    @rmeta3391 2 หลายเดือนก่อน +3

    I'll give this a shot. Comfy can do anything! I have both kohya and AI-toolkit installed on my RunPod Network storage. The learning curve for kohya, however, is steep. Thanks for the video. I saw your "handbag" video a few days ago and thought it was pretty professional. You should link that here so others can see it. Sure that Video cost you some major RunwayML credits to get your IG model moving as well as she did. Congraz on that video.

    • @Aiconomist
      @Aiconomist  2 หลายเดือนก่อน +1

      Thanks a lot. It's essentially the same thing. In that video, I used AI-Toolkit instead of this one and Kling instead of RunwayML.

    • @eveekiviblog7361
      @eveekiviblog7361 2 หลายเดือนก่อน

      ​@@Aiconomist is ai toolkit process in comfyui?

    • @rmeta3391
      @rmeta3391 2 หลายเดือนก่อน +2

      @@Aiconomist Okay, I did see the credits at the end of that video, Kling did a nice job.

    • @Larimuss
      @Larimuss 2 หลายเดือนก่อน

      Yeh I've been using comfy locally for image gen. It's crazy what it can do. Been trying out lora on runpod. Fluxgym is really quick and easy but it would be good to use comfy then you just have workflows for everything 😂

  • @Nishitbeststuner
    @Nishitbeststuner 29 วันที่ผ่านมา

    If running Runpod, would you rather suggest using this method or the other method using AI-Toolkit in one of your videos?

  • @Beauty.and.FashionPhotographer
    @Beauty.and.FashionPhotographer หลายเดือนก่อน

    convert the sample prompts inputs converting to widget does physically not show up nor exist on my version of comfyui, neither does the popup window look the way it looks here. any idea why that is, and if there is a workaround for converting this to a widget? Google does no show any results when searching for this step, ..... see timestamp 8:58

  • @petera.4707
    @petera.4707 2 หลายเดือนก่อน +1

    Thank you! an exceptionally good tutorial with basic information for a noob like me. you are a great teacher!

    • @Aiconomist
      @Aiconomist  2 หลายเดือนก่อน +1

      You're very welcome!

  • @samsaraAI2025
    @samsaraAI2025 22 วันที่ผ่านมา

    I had a lot or errors and missing nodes when loading the .json for the 1st time in ComfyUI

  • @MsDario66
    @MsDario66 14 วันที่ผ่านมา

    I tried the steps but with the missing nodes update in manager appears: " (IMPORT FAILED) ComfyUI Flux Trainer" ....can I solve it?

  • @ChrissyAiven
    @ChrissyAiven 2 หลายเดือนก่อน

    Is it crucial that all training pics have the same size? And can they also be like 4:5?

    • @Aiconomist
      @Aiconomist  หลายเดือนก่อน +1

      Yes it's crucial, the images must be 1024x1024 pixels

    • @ChrissyAiven
      @ChrissyAiven หลายเดือนก่อน

      @@Aiconomist oh ok...now my training is done and I had like 4:5 ratio but biggest side was 1024px. Well, I do some more testing anyways. My results really need an upscale but they do without the lora too. And I use the same ration and sizes for creation.

  • @ywueeee
    @ywueeee 2 หลายเดือนก่อน +1

    are you available for custom script work?

    • @Aiconomist
      @Aiconomist  2 หลายเดือนก่อน

      Yes, you can find my email in the description box

  • @EFQWEFWEFWEFW
    @EFQWEFWEFWEFW 2 หลายเดือนก่อน +1

    Can I use Runpod to use this flow?

    • @Aiconomist
      @Aiconomist  2 หลายเดือนก่อน

      Yes, you can follow the same steps

  • @Thatlazerlaze
    @Thatlazerlaze 2 หลายเดือนก่อน

    How did you do the animation at the end?

  • @vaccine9834
    @vaccine9834 21 วันที่ผ่านมา

    i do the vediosays, but when training, it says a error"InitFluxLoRATraining
    module 'torch' has no attribute 'float8_e4m3fnuz'" could anybody told me how to solve this? thanks a lot

  • @yerysh
    @yerysh 18 วันที่ผ่านมา

    and where to get clip_i и text encoder, please!

  • @baingeamain5018
    @baingeamain5018 2 หลายเดือนก่อน +1

    Does it work for people ? Like I have unfortunately a 8 gb card and I would like to make Lora of people

    • @petertremblay3725
      @petertremblay3725 15 วันที่ผ่านมา +1

      Of course you can but it is going to take longer.

  • @piyush_kashyap23
    @piyush_kashyap23 2 หลายเดือนก่อน

    Can I try this in Google colab free version ?

  • @AmyWarren-u2s
    @AmyWarren-u2s 2 หลายเดือนก่อน

    Nice work 😊

  • @cedricweber4141
    @cedricweber4141 2 หลายเดือนก่อน +2

    Can someone tell me if this works on Mac?

    • @AnnieDalgaux-s6m
      @AnnieDalgaux-s6m 8 วันที่ผ่านมา

      just tried using the comfyUI manager, 'install from GitHub', I can see the nodes after updating to the new ComfyUI, but did not try it yet.

  • @Musica-qj1nn
    @Musica-qj1nn หลายเดือนก่อน

    guy could someone help me plese, why do i have this error: Warning:
    Missing Node Types
    When loading the graph, the following node types were not found:
    VisualizeLoss
    UploadToHuggingFace
    TrainDatasetAdd
    FluxTrainValidate
    FluxTrainValidationSettings
    FluxTrainSave
    OptimizerConfig
    FluxTrainEnd
    FluxTrainLoop
    FluxTrainModelSelect
    TrainDatasetGeneralConfig
    InitFluxLoRATraining
    No selected item
    Nodes that have failed to load will show as red on the graph.

    • @mohammadhussain4547
      @mohammadhussain4547 16 วันที่ผ่านมา +1

      Same issue with me. Did you find any solution for it?

  • @yasin6904
    @yasin6904 หลายเดือนก่อน

    "I don't recommend using the portable version" - how come?

  • @ronbere
    @ronbere 2 หลายเดือนก่อน

    i train my loras with it...good but i can't resume

  • @guillaumeboucher7769
    @guillaumeboucher7769 2 หลายเดือนก่อน

    thanks for the tutorial, the images look really underwhelming. Many features are not matching on the original bag. Its really discouraging to see that type of results.

  • @Anelo96667
    @Anelo96667 2 หลายเดือนก่อน

    This is useful contents.
    But I don't think I'd actually use it for a product detail page.
    The products need to match 100% accurately.

    • @SenoSbds
      @SenoSbds หลายเดือนก่อน

      Just wait.. It will improving

  • @mneyanmels
    @mneyanmels หลายเดือนก่อน

    ValueError, is not a directory ........

  • @guijiao
    @guijiao 2 หลายเดือนก่อน

    学一下