L3: Latent Upscaling in ComfyUI - Comfy Academy

Поделиться
HTML-код
  • Опубликовано: 1 янв 2025

Комментарии • 81

  • @qianmax4146
    @qianmax4146 2 дня назад

    This is very helpful and well organized. Thanks for your wonderful video!

  • @giantbonsai8950
    @giantbonsai8950 5 месяцев назад +6

    Going through these one by one. Thank you. Very clear and concise directions and explanations.

  • @icedzinnia
    @icedzinnia 11 месяцев назад +6

    Excellent! Clarified a number of things for me. Plus, I've heard we get better final results by doing an upscale in latent first but i couldnt figure out how and i couldnt find that random video i watched that stated this and showed it. So this quick tutorial is very helpful!

  • @caseyj789456
    @caseyj789456 8 месяцев назад +1

    Dear Olivio thanks for so enlighten videos! greetings from thailand. (I ll keep watching until the last! )
    (Upscale 5:44)

  • @ojciecvaader9279
    @ojciecvaader9279 9 месяцев назад +2

    I really appreciate your lessons. Super job, great explanations. Thank you

  • @Maisonier
    @Maisonier 11 месяцев назад +7

    Dude you are awesome. Thank you for sharing all this.

  • @amarmayekar7795
    @amarmayekar7795 4 месяца назад +2

    Thank you. Very clear and concise directions and explanations.

  • @justthinkaboutit7983
    @justthinkaboutit7983 9 месяцев назад +1

    Great lessons. I'm following along everything you do.

  • @kacperwozniak2982
    @kacperwozniak2982 4 месяца назад +2

    Man, I haven't finished the tutorial yet but this tutorial is bonkers! Great job, thanks a lot!

  • @MaximusOne
    @MaximusOne 27 дней назад

    These tutorials are excellent! Thank you Olivio!

  • @gwcstudio
    @gwcstudio 9 месяцев назад

    your enthusiasm is contagious!

  • @XuRZaL
    @XuRZaL 3 месяца назад

    Thank you, this has been so helpful. ComfyUI is very user friendly once you get past the learning curve (which you have shown me), and you have absolutely made that curve so much shorter. On to L4!!!

  • @lucasfrancelino5141
    @lucasfrancelino5141 Месяц назад

    This helped me so much, thank you!!!!!

  • @BrawnyAi
    @BrawnyAi 7 месяцев назад

    The step to do another i2i was key compared to other tutorials. Thanks!

  • @devon9374
    @devon9374 3 месяца назад

    Man... I code, but I love a node based workflow sometimes.
    This, Unreal Engine Blueprints, DiVinci Resolve. Chef kiss 👌

  • @Alex-yi3yw
    @Alex-yi3yw 10 месяцев назад

    Thank you very much for this amazing lesson!

  • @maxfxgr
    @maxfxgr 11 месяцев назад

    Let's go another masterpiece! greetings from Greece!

  • @1lllllllll1
    @1lllllllll1 Месяц назад

    5:33 nowadays, use preview bridge node to interrupt at the preview, then re-queue to finish if you like it. Or use interrupt node.

  • @majinvegeta5393
    @majinvegeta5393 10 месяцев назад

    Thanks for your lesson, it helps a lot, thanks !

  • @mibo747
    @mibo747 11 месяцев назад

    Many thanks for explanation!
    THATS PRICELESS!

  • @martinianogarcia9167
    @martinianogarcia9167 11 месяцев назад

    This series is great Olivio!

  • @VIKclips
    @VIKclips 11 месяцев назад +2

    how do I transfer image from the first workflow to the second, when i bypass 1st one, i have image that i want to upsclae, but there is no way to do that...

  • @AIKnowledge2Go
    @AIKnowledge2Go 11 месяцев назад +5

    Hi Olivio, thank you for that insightfull video about latent upscaling. Did you know you can just Edit the "run_nvidia_gpu.bat" file with "--preview-method auto" on the end? This will show a preview in the in the K-Sampler. Happy creating.

    • @amritbanerjee
      @amritbanerjee 10 месяцев назад +2

      This is comment is more productive then the last 10 videos I have watched.

  • @vmiyeli
    @vmiyeli 6 месяцев назад

    Thank you Olivio for your tremendous work ! Can you inform if you know a way to manage regions in the image with different denoise ? Indeed, the latent upscale is changing some details of the image. I would like to keep specific region with unchanged regions. Thank you very much !

  • @r__b6095
    @r__b6095 8 месяцев назад

    Great tutorial. I'm looking forward to getting to grips with comfy using this.
    Though I'm afraid the method gave some ungodly results though haha

  • @eddieyu80
    @eddieyu80 3 месяца назад +1

    "Upscale latent" before "VAE decode" is much better than upscale the output image with more detail. This lesson integrates Upscale and KSampler, one more step for beginner.

  • @TheNicoFX
    @TheNicoFX 8 месяцев назад

    you are a legend ! thank you so much

  • @hamidthephysicist6376
    @hamidthephysicist6376 7 месяцев назад

    Thank you very much sir.

  • @Jaysunn
    @Jaysunn 9 месяцев назад

    Is there a function that pauses the generation while you inspect the lower res image, see if you like, then click "proceed" to upscale it?

  • @xorq
    @xorq 7 месяцев назад

    My question is wouldn't it be equivalent or better to put more noise an more steps (to be equivalent in GPU time) into the ultimate upscaler ?

  • @ndro7068
    @ndro7068 6 месяцев назад

    What is the ultimate upscaler you talk about at the end of the video?

    • @OlivioSarikas
      @OlivioSarikas  6 месяцев назад

      the node is called "ultimate upscale"

    • @ndro7068
      @ndro7068 6 месяцев назад

      @@OlivioSarikas Oh strange. I could not find it. Thank you

  • @sethmunich4062
    @sethmunich4062 9 месяцев назад

    The upscaled images still had wonky eyes and mistakes in the clothing. I tried a different checkpoint and the upscaled image came out blurry.

  • @bakablitz6591
    @bakablitz6591 11 месяцев назад

    yoooo here before the video is even out. now thats early access!

  • @64kernel
    @64kernel 11 месяцев назад +1

    Muchas gracias maestro! Me enganché con Comfyui gracias a ud. y ya no puedo cambiarlo. Excelentes tutoriales, aunque ciertamente no son para principiantes.

    • @kkryptokayden4653
      @kkryptokayden4653 11 месяцев назад

      Igual es muy util y se puede hacer de todo

  • @AyuK-jm1qo
    @AyuK-jm1qo 5 месяцев назад

    where in your workflow do I pick the upscaler, animesharp for example? I cant find it anywhere.

  • @KlausMingo
    @KlausMingo 4 месяца назад

    Why upscale the rendered latent image, instead of using a bigger empty latent image at the start of the workflow?

    • @OlivioSarikas
      @OlivioSarikas  4 месяца назад +1

      because a) the model might not be able to handle that high of a resolution and b) if you upscale a image that already exists the AI can work with and improve the existing details, rather than creating a new image with less details. The 1.5 models are created for low res images. only XL and later models can handle higher resolutions for the initial text 2 image render

  • @nicodemuss89
    @nicodemuss89 11 месяцев назад +1

    Hi there! That upscale can be used on an existing image?

    • @VIKclips
      @VIKclips 11 месяцев назад

      I don't think so, actually this lesson doesn't really provide enough information to upscale an image in the end you end up with nothing.
      For example I follow the tutorial and then i make an image, then I can't keep it, i need to make another image and upscale it.... What is the point in this?

  • @williamcase426
    @williamcase426 10 дней назад

    I wanna be comfy should I learn this UI

  • @nicobay4512
    @nicobay4512 6 месяцев назад

    I have probleme with face, mouth without teeth , horrrible eyes etc. Any tutorial ma man ?

  • @mariokotlar303
    @mariokotlar303 11 месяцев назад

    I'm interested in how this compares to doing the intermediary upscale step in pixel space rather than in latent space. And if doing it in latent space is consistently better, then I'm interested in understanding why, as I thought latent upscaling was inferior to using esrgans in pixel space.

    • @petec737
      @petec737 11 месяцев назад +1

      Latent upscaling IS inferior, no questions about it.

  • @구원자님
    @구원자님 8 месяцев назад

    Please someone tell me how to increase the size of the blue area in ComfyUI. The blue lines. So that my workflows can be saved the next time I open Comfyui, I assume? :)

  • @violentpixelation5486
    @violentpixelation5486 11 месяцев назад

    This confy ui series is so great!!!! 💯👌

  • @sethmunich4062
    @sethmunich4062 9 месяцев назад

    Why don't I have the "TEXT TO IMAGE" bar at the top of my ComfyUI workspace?

    • @myhlow
      @myhlow 3 месяца назад

      Right click and select "Add Group" to create an empty panel to group your nodes.

  • @RicardoCampos-bs6fi
    @RicardoCampos-bs6fi 4 месяца назад

    MASTER

  • @PL-808
    @PL-808 11 месяцев назад

    Do you prefer Comfy because of it’s customizable and for speed? Been using a1111 about a year, and happy to have learned it. Is it bad, not developed anymore? I think the ui is simpler and less cluttered for my eyes, so I rather not change or is it necessary? Been following your channel and all a1111 tutorials, so a bit sad and wondering why no more on this. But also totally understand. Thanks 🙏

  • @Shisgara77
    @Shisgara77 11 месяцев назад

    why so few likes? Thank you, Olivio

    • @us_f4rmer
      @us_f4rmer 11 месяцев назад

      Because no one care´s about, its youtube (duh), it basically is pointless. What even is the value, if you can´t even dislike which would at least tell you if you´re dealing with bs content or not, even if there were just a couple of K likes on a million views video you would know what´s up

  • @the_infotika_Mayabhumi
    @the_infotika_Mayabhumi 11 месяцев назад

    You are a ginius in AI creative world. Thanks for your lesson.
    I have a query. Can we create a visual multi character interaction UI images, where multiple characters are taking each other in a background environment....? Want to know how it made.....
    Thank you.

    • @lefourbe5596
      @lefourbe5596 11 месяцев назад

      It IS called region prompt in comfyUi.
      There is an extension in A1111 that Can give you a great deal of "what you want" under each and every Chunk (or Mask) you make.

  • @VIKclips
    @VIKclips 11 месяцев назад +1

    As I understand this is not an example of how to upscale already existing image, but how to upscale images on the go, and doing it this way causes a lot of issues, espacially when cancelling an image in progress, it can glitch the GUI.
    For a person that came here for upscale, i want to see how to upscale an image that I already created.... this is not really helpful

  • @chrisstahl2653
    @chrisstahl2653 Месяц назад

    The real ComfyAI is free. You don't need OpenArt for it.

  • @TheHmmka
    @TheHmmka 11 месяцев назад

    🌿

  • @LouisGedo
    @LouisGedo 11 месяцев назад

    👋

  • @gatomio9739
    @gatomio9739 11 месяцев назад +2

    A1111

  • @None_ya_B
    @None_ya_B 10 месяцев назад

    No thanks

  • @DivinityIsPurity
    @DivinityIsPurity 11 месяцев назад +13

    Thumbs up for A1111 only channel.

    • @mongini1
      @mongini1 11 месяцев назад +9

      Why should I use an inferior interface? Comfy is faster, uses less VRAM, can do way more...

    • @lefourbe5596
      @lefourbe5596 11 месяцев назад +24

      Thumbs UP for comfyUi only channel

    • @DivinityIsPurity
      @DivinityIsPurity 11 месяцев назад +1

      @@mongini1 Boooo!! A1111 doesn't have spiderweb like interface, plus low vram doesn't bother me.

    • @mongini1
      @mongini1 11 месяцев назад

      @@PixelsVerwisselaar if they learned how to get A1111 going they're willing to learn. Otherwise they wouldn't generate anything. People who don't want to learn use Midjourney ;)

    • @MaddJakd
      @MaddJakd 9 месяцев назад +1

      ​@PixelsVerwisselaar is that so bad though. For the longest time, A1111 was what we had, and that was the interface used to make LOADS of neat and even provocative stuff.
      Not all tools are meant to be used by everyone. Nothing wrong with that.

  • @MeMine-zu1mg
    @MeMine-zu1mg 11 месяцев назад +1

    What are you talking about artistic expression? You are telling a computer what to do hwo is that artistic expression the closest thing you get to being either artistic or expressing yourself is the prompt you write. Bypassing the years and decades that actual artists spent practicing and practicing. Look I have no problem with image generation even though it's putting artists out of work because that's life. Technology comes along and up-ends industries. But I do have a problem with people writing promises and calling themselves artists, forgetting that you didn't actually create that image, or video, or story. And TBH it's kind of a slap in the face to the people who put in the work and the time, and the sweat, and sometimes even blood and tears to become masters of their craft. So please, don't talk as though you're artists.

    • @J3R3MI6
      @J3R3MI6 11 месяцев назад +2

      Don’t worry bro… some form of social ip made from AI + blockchain (or similar tech) will make it so Artist will be able to get paid for AI using their art for generating images and video.

    • @tracelistener
      @tracelistener 11 месяцев назад +2

      Human expression is art whether the tool you used is a paintbrush or computer. Imagination is still needed and the end result can still be felt by the viewer.

    • @pj2345-v4x
      @pj2345-v4x 3 месяца назад

      Virtue signal somewhere else.

  • @gingercholo
    @gingercholo 11 месяцев назад

    "my workshop" you barely know how this software works and it shows.

  • @zizhdizzabagus456
    @zizhdizzabagus456 9 месяцев назад

    launch wokfow doesnt really work, asking me to join discord that i did. still

  • @sickvr7680
    @sickvr7680 Год назад

    Gracias Olivioooooooo

  • @rojodiver3344
    @rojodiver3344 11 месяцев назад

    I appreciate what tou jave been doing for a long time, but this just seems overly complicated compared to just using a better program like Leonardo AI. These images look so fake and seem useless to any commercial output. Leonardo's Photoreal is so much better and they now have real time output so you can see instantly changes.
    It would be useful if there was a less potent AI but which had more control over pose and output, but i jist dont see it in all your videos.

    • @spejarn
      @spejarn 11 месяцев назад +2

      Did you get paid to say that? There's a lot of that going around over the last few months i've noticed, especially from NAI v3, all it does is make sure i'll never even consider using it.

    • @us_f4rmer
      @us_f4rmer 11 месяцев назад

      smells fishy, reminds me on Shilli Vanilli ....

    • @Shisgara77
      @Shisgara77 11 месяцев назад +1

      in Leonardo it is possible to take a reference picture, on its basis to make your own in a similar style (ipadapter), to put it in the necessary pose and location, to make so that in one generation to apply 2 different models (for example photorealistic object in the drawn world), etc.? Maybe I'm wasting my time learning Comfy when there is already such a convenient and simple interface 🤔

    • @bigdaddy5303
      @bigdaddy5303 11 месяцев назад

      Dude absolutely everything you've said is nonsense. You can control absolutely everything in comfyui. Stick with Leonardo, comfyui is clearly too complicated for you.