OmniGen New Diffusion Model - You Might Not Need Photoshop Anymore

Поделиться
HTML-код
  • Опубликовано: 28 ноя 2024

Комментарии • 48

  • @kalakala4803
    @kalakala4803 25 дней назад +4

    I like how it can remove an object without a need of segment and masking.❤ Great vid.

  • @inside_fighting
    @inside_fighting 24 дня назад +3

    I can't get omnigen to work. It keeps getting stuck on processing. I have a very fast computer. Im just doing text to image without reference and it still wont work. Any help? I also randomly just get "error" on the processing screen after a while.

    • @inflatestudio
      @inflatestudio 11 дней назад

      same issue if you got solution let me know

  • @BoomBillion
    @BoomBillion 25 дней назад +3

    I tried the stand alone version and it sucks arss. Tried diff promps and it could not do a basic face swap.

  • @insurancecasino5790
    @insurancecasino5790 25 дней назад +1

    I noticed more frame by frame stuff on Civitai. Sadly it's mostly adult stuff, but I hope the process catches on because all low end PC or laptop owners will be able to make high quality vids as long as they can generate images based on video frames. More time and work, but that's the trade off. LOL. Awesome vid by the way.

    • @TheFutureThinker
      @TheFutureThinker  25 дней назад

      Those img2img type video, can be done using one of the good old extension in A1111, Mov2Mov. It was my favourite before AnimateDiff.

    • @TheFutureThinker
      @TheFutureThinker  25 дней назад +2

      Btw, Civitai become like a AI porn site for nerds. LOL a very special niche market porn site it is becoming now, rarely go there now, unleast I need to download AI model files.

    • @insurancecasino5790
      @insurancecasino5790 25 дней назад

      @@TheFutureThinker Word. Most of the requests or replies I got, was for adult stuff. LOL I was just pointing out there was a way do make good vids with low VRAM but it would take more work and hopefully developers work on improving it, and they all just asked if they could make adult stuff. I was like: What about Sci Fi or action movies? Not one response. LOL.

  • @yklandares
    @yklandares 22 дня назад

    Hi and where to put the model and why it is not in the nodes tell me

  • @Meowjesticc
    @Meowjesticc 11 дней назад

    my omnigennode have latent, what shoud i do?

  • @Deadlious
    @Deadlious 25 дней назад +19

    According to my tests, this model is crap. The idea is great, but the model itself did not generate even one usable image. I already spent several hours, to get the thing going and generate tens of images. Smudges, inconsistent physics, crappy hands... I even got some gibberish text on top of one of the images... I tested with and without input images, it's the same... There was one image I was about to say I'm happy with, and then I noticed that it put too many buttons on the coat of a woman, even on places where there shouldn't be buttons.

    • @VinciSid
      @VinciSid 24 дня назад +1

      Nice try Adobe

    • @Deadlious
      @Deadlious 24 дня назад

      @@VinciSid I've got nothing to do with Adobe. Except for despising them for their "unfriendly" customer polices.

    • @JohnMC-h6h
      @JohnMC-h6h 24 дня назад +1

      I agree, I have been testing tons of different conditions, prompts, styles, etc... The outputs are all poor and useless. Sometimes something "okay" comes out but even then to have it actually useable would require massive amounts of manual work after the fact.

  • @CharlieLee-l9w
    @CharlieLee-l9w 17 дней назад

    I'm following your video exactly and still can't use it because the node is still red with Import Failed. I've asked claude and others to try and figure this out, but it's still not working.

  • @kalakala4803
    @kalakala4803 25 дней назад +1

    01:56 unbelievable how consistent style it can be created.

    • @TheFutureThinker
      @TheFutureThinker  25 дней назад

      it need to be a clear image for reference to achieve that result. That why some people just randomly test it, and saw not getting result like this one.

  • @milycortes3100
    @milycortes3100 24 дня назад

    I am not so sure why, but today is the second day I use the hugging face space for omnigen and got "error" in output.
    I still haven't been able to make it work locally, but still figuring out. Hopefully is a me problem, and this error in their hugging space is not related.
    Glad to see you talking about it! New sub

  • @ThePrime125
    @ThePrime125 24 дня назад

    The architexture is really great

  • @hicks100
    @hicks100 4 дня назад

    Can this be installed on Stable Diffusion WebUI?

    • @TheFutureThinker
      @TheFutureThinker  4 дня назад

      Unfortunately stable diffusion web UI have not update at all

  • @vincentgautier5036
    @vincentgautier5036 25 дней назад +1

    Hi, it seems to be very intersting, I would like to try it with ComfyUI, I don't understand how to load model and VAE, can you update the readme please ?

    • @ChrissyAiven
      @ChrissyAiven 25 дней назад

      I would like to try it to, but I have comfyUI on pinokio, I think that could be a problem? With CatVTON I have problems too with having to install requirements. :(

  • @Dr.UldenWascht
    @Dr.UldenWascht 25 дней назад +2

    OmniGen offers a convenient way to access a series of features that are typically available separately or through different ComfyUI workflows. It combines tools that function like Insightface, ControlNet, and inpainting, etc, into a single package, which is helpful for users who prefer not to wrestle with the complexities of ComfyUI noodles 😄
    However, this convenience comes at a cost. The model's weights size of ~16 GB (partially loaded into VRAM) makes it either impractical or extremely slow on common GPUs (with 8 GB of VRAM). Additionally, to be completely honest, the functionality provided by OmniGen can be replicated using separate models outside of the package. There's almost nothing new here, as far as I can see.
    Nonetheless, it is a commendable effort to merge everything a regular user needs in one place.

    • @JohnVanderbeck
      @JohnVanderbeck 25 дней назад +2

      It also costs flexibility. I often compare ComfyUI to the software Nuke which I used for many years when I worked as a digital compositor. Having things as separate nodes allows for far more flexibility than just one node that does a whole slew of things.

    • @bushwentto711
      @bushwentto711 25 дней назад

      Bruh ur just mad you spent hours like me on complex comfyui models that literally get out performed by a few nodes hahaha

  • @kkryptokayden4653
    @kkryptokayden4653 25 дней назад

    got it working on AMD (linux). This is such a powerful yet simple node.

  • @Elaina-nn5jd
    @Elaina-nn5jd 20 дней назад

    I dont know, but my comfyui stuck and it wont run.

  • @berhunt
    @berhunt 25 дней назад

    I wished it could. I tried it a lot now and no result are es good compared to what I did with the same picture in photoshop. Even simple stuff, like remove sky leads to a new strange image. Only the example they have in the list are working well.

  • @Aibyda
    @Aibyda 25 дней назад

    How to run .txt in colab or how to open command in colab

  • @snipervicar
    @snipervicar 25 дней назад

    Good video, but it's too slow, how to speed it up in low vram machine.

  • @piracymoney
    @piracymoney 25 дней назад +1

    🔥

  • @ismgroov4094
    @ismgroov4094 25 дней назад

    Respect you, sir. Comfyui node plz, sir.

  • @LuckRenewal
    @LuckRenewal 25 дней назад

    the other day i was just really disappointed at the new adobe ai photo editor. i think omnigen is really a game changer. just the installation process is a bit more tedious than other nodes.

  • @travelwithfaycal
    @travelwithfaycal 25 дней назад

    I tried it, without image (10 minutes), with 1 image ...1 hour 😢

  • @dalecorne-new-mtv
    @dalecorne-new-mtv 25 дней назад +2

    It's pronounced Ah'm Nee Gen.

  • @SasukeGER
    @SasukeGER 25 дней назад

    not that good ... used on image ... asked "replace the head with Image 2" ... and wasnt great ....

  • @Nocare89
    @Nocare89 25 дней назад +1

    Seems it does a poor job of retaining the original image when it does it's 'maskless inpainting' gimmick.

  • @Fluxailab
    @Fluxailab 24 дня назад

    The model does not perform well in hand details, and when editing images, individual text prompts are far less effective than using images as a reference. But I think the model's consistency is excellent when it comes to image fusion.

  • @erleguizamo
    @erleguizamo 25 дней назад

    Now I am out of work

  • @ratside9485
    @ratside9485 25 дней назад +4

    Results are often not good.

    • @TheFutureThinker
      @TheFutureThinker  25 дней назад +1

      Then your prompt structure and input images need to optimize

  • @marcduchamp5512
    @marcduchamp5512 12 дней назад

    Super slow on their site, almost useless, this is why nobody will use it in the long run because it requires too much setup and coding installations. I’ll wait for a superior model and software