HunYuan DiT vs Pixart Sigma - which is better?

Поделиться
HTML-код
  • Опубликовано: 1 окт 2024
  • Time for a prompt showdown! Try these at home using your favourite model (such as Stable Diffusion 3) and see how they compare to both Pixart Sigma and HunYuan DiT. A wide range of styles are covered, and the prompts are listed below for your copy-and-paste ease :)
    Want to support the channel?
    / nerdyrodent
    / prompt-showdown-106258102
    Update: HunYuan v1.1 is out now too - and it's even better!
    huggingface.co...
    = Prompt Showdown! =
    Negative Prompt:
    many hands, really wobbly, distorted and blurry fingers and hands.
    Positive Prompts:
    1:
    2: A woman sleeping on the grass.
    3: An avocado chair
    4: Anime art style blue rabbit flying through the air wearing a red cape and goggles
    5: Vector-art style business logo, SVG, simple, plain, psychic rodent emoji.
    6: A fantasy-art style rodent mage, level 12
    7: Chibi rodent scientist discovering where he left his pencil
    8: A giraffe engineer, epic Shōnen manga style, high detail Seinen
    9: Kemono style illustration of a rabbit doctor wearing a blue surgeon's gown walking down the path of an old cemetery. Green eyes, brown fur, fine ears, solemn feel. misty path. dignified, austere. fog, gloom. haunted vibe.
    10: Cthulhu stands over a kitten, oil painting style. A classical artwork, vintage, old. The scene is set as the beast towers over the tiny, but dreadful, all-black kitten. The bright summer sky is in high contrast to the dark, evil shadows that lurk beneath. The style is reminiscent of Jacques Stella and Pompeo Batoni. In the background is an ancient fiery temple of doom, hewn from the very rock face itself by the kitten. The image quality is astounding.
    11: A cubist art style kangaroo druid in the forest, cubism, quality artwork, soft pastel shades
    12: In the Chinese ink painting style, a little deer stands leisurely in the lush bamboo forest. The morning light shines through the bamboo leaves, casting mottled light and shadows. The little deer quietly drinks the clear stream water, and the bamboo leaves are swaying in the wind. The whole picture exudes a sense of tranquillity and harmony.
    13: A stone statue of a deer relaxing on top of a colourful bed in an old, Victorian house. Professional photo, bokeh.
    14: A childish doodle of a bad kitten
    15: Paper-cut art style tiger emerging from a bunch of flowers, bold colours, depth, shading, 3d effect, pop-up
    16: A ghostly face is peering through the window into my house from outside! The ghost looks very scary but may be wearing a mask, and a sense of evil can be felt. charcoal art, shading, sketch.
    17: 在一个充满创意的画廊里,一幅引人注目的画作展示了一只时髦的啮齿动物。画作采用了印象派风格,色彩斑斓,笔触细腻,充满生动的光影效果。画中的小鼠戴着复古圆框眼镜,穿着时尚的格子衬衫和牛仔裤,手持一杯咖啡,悠闲地站在一片繁茂的花园中。背景是柔和的色彩斑点和模糊的树影,营造出宁静与优雅的氛围,突显了小鼠的独特魅力。
    == More AI Things! ==
    * Anaconda for MS Windows Beginners - • Anaconda - Python Inst...
    * Installing ComfyUI for Beginners - • Install Stable Diffusi...
    * ComfyUI Workflows for Beginners - • ComfyUI Workflow Creat...
    * Easy Consistent Character in ANY pose - • Reposer = Consistent S...
    * Make an Animated, Talking Avatar - • Create your own animat...

Комментарии • 60

  • @doggod9754
    @doggod9754 3 месяца назад +42

    >"A woman sleeping on the grass"
    oh this is targeted XD

  • @StringerBell
    @StringerBell 3 месяца назад +41

    SD 3 licensing was a big mistake.

    • @PAEz...
      @PAEz... 3 месяца назад

      Its a commercial product now.....sad.
      But Im sure the ai community will keep promoting it for them.

    • @NevelWong
      @NevelWong 3 месяца назад +3

      I think it was high time. They need money to train their models. I don't think donations would cover it. And there were plenty of services raking in big money from the work they did, while they didn't see any of it. I wish the license was more permissible for personal use, but people need to stop being so entitled.

    • @StringerBell
      @StringerBell 3 месяца назад +17

      ​@@NevelWong It's not an entitlement. The models they release are borderline unusable at their raw state.
      Thanks to the community and countless hours, GPU compute, work and money the community develops additional models, Control Nets, IC Lights and what not and this is what makes the Stability special.
      They singlehandedly deterred anyone who had incentive to develop and fix their product for FREE to lose interest immediately.
      In the end, why I have to lose sleep, money and time developing your product to have to pay YOU for that privilege at the end?
      Sounds dumb, isn't it?

    • @stratos7755
      @stratos7755 3 месяца назад

      @@PAEz... The ai community is pretty pissed because of the extreme censorship in the model.

    • @PAEz...
      @PAEz... 3 месяца назад

      @@stratos7755 I know.

  • @a.........1._.2..__..._.....__
    @a.........1._.2..__..._.....__ 3 месяца назад +21

    I love sd3. Ive always wanted to see piccasso art with realistic humans.

    • @NerdyRodent
      @NerdyRodent  3 месяца назад +1

      Picasso does realism now

  • @vitalis
    @vitalis 3 месяца назад +4

    I’m not even downloading SD3. Absolutely no incentive. I rather stick with SDXL and the community models/loras.

  • @worthstream
    @worthstream 3 месяца назад +11

    RUclipsrs are a big part of the community around these projects, and draw a lot of people that eventually fine tune, train l'ora, control nets, etc.
    I'm not sure requiring a separate license for creators was a good idea.

  • @ayrengreber5738
    @ayrengreber5738 3 месяца назад +8

    Laughs maniacally… I loved the comparison to stable diffusion 3… and the careful review of their license. I was worried everyone would shrug and take SD3 licensing like cascade.

  • @swannschilling474
    @swannschilling474 3 месяца назад +5

    My god, not even using SD3 in the video is actually a very good reaction to the new license...so sad that all we can do is cancel the new model, because its too censored and has this horrible license attached! 😢

  • @rundiffusion
    @rundiffusion 3 месяца назад +3

    Great video! It would have been awesome to test typography and text. Also scene composition. “Next to”, “on top of”, “in the background” etc. a woman holding a sign that says “Freedom” standing next to a police officer with a chat bubble above saying “Pay a license”. 😢

    • @NerdyRodent
      @NerdyRodent  3 месяца назад +1

      Thanks! I did some tests like that in the previous videos, and they’re quite good at composition, not so much at English text. The latest HunYuan is also much better as they’ve fixed the colour issues. Whilst also not a truly open source license, I think it’s reasonable enough for most people to use.

  • @MyAmazingUsername
    @MyAmazingUsername 3 месяца назад +2

    Thanks for the great overview! HunYuan is better at following most prompts you gave it, and better at composition, and was better at proper human faces and hands. But I sadly think its name will hold back its popularity. Its name doesn't sound cool. It sounds confusing. 😅 Edit: Oh it's by Tencent, now the quality makes sense. They are some of the best in the world at this stuff.

    • @NerdyRodent
      @NerdyRodent  3 месяца назад +1

      IP adapter soon! 😃

    • @MyAmazingUsername
      @MyAmazingUsername 3 месяца назад +1

      ​@@NerdyRodentOh yeah true, Tencent are great at creating IPAdapter stuff, which I hope helps the popularity of this model. I can't even remember the name right now to write it again in this comment. Yuanhun something? That's a problem for its popularity. 😅 There's only one or two fine-tunes for it right now. Really hoping it gets more popular soon.

    • @MyAmazingUsername
      @MyAmazingUsername 3 месяца назад

      ​@@NerdyRodentThe difficult name strikes again. I couldn't remember its name and literally had to check here again. I don't think the name was a good idea. 😁

  • @KalLif-k3i
    @KalLif-k3i 3 месяца назад +1

    Hahahaha nerdy rodent brings us AI stuff in a very British way!!!

  • @deividaszubLT
    @deividaszubLT 3 месяца назад +2

    Could you make HunYuan installation tutorial?

    • @NerdyRodent
      @NerdyRodent  3 месяца назад +1

      There’s always last week’s video 😉 HunYuan DiT - Open Source & Better Than Stable Diffusion 3?
      ruclips.net/video/oDK0-KesWQo/видео.html

  • @Avenger222
    @Avenger222 3 месяца назад +3

    They're both really quite good out of the box too! I really hope the community ralies behind these models.

  • @gurilagardnr2688
    @gurilagardnr2688 3 месяца назад +2

    When comparing apples to apples, sd3 appears to have really fallen on it's face here, at least for me. Hunyuan took the gold overall. This was a fun and interactive experience. Nice work.

  • @luman1109
    @luman1109 3 месяца назад +4

    it's pronounced /hʊnˈjuːæn/ or Hun-you-wen

  • @richgates
    @richgates 3 месяца назад +4

    In my very unscientific and subjective test, SD3 only won with prompt #13 and it was virtually a tie with HunYuan DiT.

    • @NerdyRodent
      @NerdyRodent  3 месяца назад +5

      Good to know, thanks!

    • @richgates
      @richgates 3 месяца назад +2

      @@NerdyRodent Thank you for your work on this! I'm in the middle of doing a threeway test on 30 prompts. I modified your workflow slightly to also do the SDXL refinement on PixArt Sigma like your other workflow. I'll be posting the results either on Twitter or Reddit.

  • @vVinchi
    @vVinchi 3 месяца назад +2

    That was too far Rodent! Or was it☺

  • @jimdelsol1941
    @jimdelsol1941 3 месяца назад +1

    Damn pixart gives some beautiful images.

    • @NerdyRodent
      @NerdyRodent  3 месяца назад +1

      This isn’t even it’s final form…

  • @juanjesusligero391
    @juanjesusligero391 3 месяца назад +1

    Absolutely splendid video, as always! :D
    If I didn't know your channel I wouldn't even know these model exist, thank you ever so much for your work! ^^

  • @WatchNoah
    @WatchNoah 3 месяца назад +2

    Do you have a discord server?

  • @KlimovArtem1
    @KlimovArtem1 2 месяца назад

    The empty prompt test is a pretty cool idea - immediately shows the average samples it was trained on. When they look too similar to each other and too far from what you’re going to create with it - it usually means the model is bad for your use case.

  • @JustMaier
    @JustMaier 3 месяца назад +1

    Thanks for taking the time to prepare this comparison. It’s really good to understand where these other foundations stand in terms of performance

  • @kira7x2
    @kira7x2 3 месяца назад +1

    Exactly the video i was searching, thanks

  • @USBEN.
    @USBEN. 3 месяца назад

    This is weird that you cannot even feature the model results in a video.

  • @ryshabh11
    @ryshabh11 2 месяца назад +1

    Thanks

  • @terbospeed
    @terbospeed 3 месяца назад +1

    Wow top level snark :)

  • @carterknudsen525
    @carterknudsen525 3 месяца назад +1

    That really is a lovely Cathulhu

  • @bolon667
    @bolon667 3 месяца назад +4

    Sadly, without comparison with Sd3-medium

    • @WatchNoah
      @WatchNoah 3 месяца назад +3

      Would've cost him 20$

    • @bolon667
      @bolon667 3 месяца назад

      @@WatchNoah He not using this model commercially, so it's free.

    • @Cara.314
      @Cara.314 3 месяца назад +10

      @@bolon667 if he makes any money off this video, he is using it commercially.

    • @xpecto7951
      @xpecto7951 3 месяца назад +1

      @@Cara.314 just release a new short video of sd3 pics and demonetize it

    • @Avenger222
      @Avenger222 3 месяца назад +2

      Bonus meme, if he stops paying them, he would have to delete his video if he used a deriviative model of SD3.
      Why? Outputs only created by the core model (aka SD3) are immune, but anything outside that must be destroyed. I think this is their "we _really_ don't want people undoing our censorship" clause.

  • @LouisGedo
    @LouisGedo 3 месяца назад

    👋

  • @RamonGuthrie
    @RamonGuthrie 3 месяца назад

    Why did you flip the sides?

  • @MarcSpctr
    @MarcSpctr 3 месяца назад +2

    You can for sure show the SD3 without the liscence, it is not meant for you.
    People are acting like this is something so terrible of a licence even in use cases it is not meant to be for.
    It is meant for, people making money "DIRECTLY" from SD3 models.
    Your job is indirect.
    By that logic a creator who teaches stuff like Unity or Unreal Engine on youtube should also pay them them as well, right ? Cause they are making money ????
    Absolutely not, but if they make a game with Unreal or Unity, then you need to pay them for "USING" their product.
    People like you are deliberately making the situation from bad to worse.
    Cascade had similar licence.
    Did people pay for it before making videos about it ?

    • @juanjesusligero391
      @juanjesusligero391 3 месяца назад +2

      I'm not a lawyer, so I don't know for sure, however, to be safe, I think it's best to be cautious unless you're 100% certain. Also, asking Stability AI for clarification on this part of the license could be helpful for everyone

    • @justinw1219
      @justinw1219 3 месяца назад

      It wad billed as open source

    • @WatchNoah
      @WatchNoah 3 месяца назад +4

      I asked in the official sai discord server and they confirmed that you can't use sd3 in a monetarized RUclips video (or with patreon) since you are making money using sd3. In that case you would need to purchase the 20$/month license.

    • @zeevdrifter2707
      @zeevdrifter2707 3 месяца назад +1

      Trusting a corporation to not abuse vague licenses is like trusting a dog with a pack of raw steaks
      Quit speaking on issues you have zero knowledge in.