CivArchive
    Preview 30898667
    Preview 30898395
    Preview 30898394
    Preview 30898735
    Preview 30898737
    Preview 30898874
    Preview 30898736

    Styles are trained on synthetic data and do not copy the styles of artists!

    Photo 2 - is trained on real data (Dataset from Unsplash)

    [M] Merged LoRAs

    Recommended LoRA Strength (Weight): 0.8-1

    Preview generated without using ADetailer & Hires. fix

    Description

    Repeats: 10

    Epoch: 12

    Image Count: 20

    Network Rank (Dimension): 32

    Network Alpha: 16

    Batch size: 1

    Clip skip: 1

    Training steps: 2400

    Resolution: 1024x1024

    Train Resolution: 1024,1024

    Model: ponyDiffusionV6XL_v6

    LR Scheduler: cosine_with_restarts

    Optimizer: AdamW8bit

    Learning rate: 0,0005

    Text Encoder learning rate: 0,00005

    Unet learning rate: 0,0005

    Captioning: WD14 Captioning (SmilingWolf/wd-v1-4-moat-tagger-v2) + Edit

    Recommended LoRA Strength (Weight): 0.7-0.8

    Trigger Words: sketch, monochrome, greyscale,

    LoRA is trained on images created by me using ideogram.ai

    FAQ

    Comments (43)

    Sac_PoubelleSep 23, 2024· 13 reactions
    CivitAI

    Hi, is there a list of all the styles for this model? I've found that you've added so much 🤩 that the list between the title and the examples is really cumbersome to use. (^_^')
    EDIT: Thank you!

    obuyb404
    Author
    Sep 23, 2024· 2 reactions
    DitrytalesSep 24, 2024· 11 reactions
    CivitAI

    I like

    Mom0kiSep 25, 2024· 13 reactions
    CivitAI

    liked the model!

    MiwaKiwaSep 26, 2024· 14 reactions
    CivitAI

    it works well

    ouriken61343Oct 3, 2024· 12 reactions
    CivitAI

    bien

    CV_GOLFOct 10, 2024· 8 reactions
    CivitAI

    Good

    Murphy1239Oct 15, 2024· 11 reactions
    CivitAI

    I have a problem with this lora. It seems that this lora gives an extra belly button to my creations. Has someone else encountered this problem? And if yes, have how did you fix it?

    obuyb404
    Author
    Oct 15, 2024

    Can I have more details? What style is it?

    Murphy1239Oct 15, 2024

    Sorry. It's the Anime Summer Days 2 style. For some reason it gives my creations a second navel more often than not. Using it with 0.3 is somewhat viable, but the colors are just not the same when I use it with 0.5.

    Murphy1239Oct 15, 2024

    @prgfrg23 I should also mention that I use ADetailer and Hires.fix with R-ESRGAN 4x+ Anime6b

    obuyb404
    Author
    Oct 15, 2024

    @Murphy1239 I have never encountered this problem. I don't even know how to help, there's nothing in the dataset that could create problems. So I don't know why this is happening.

    Murphy1239Oct 15, 2024

    @prgfrg23 Thanks so much for your help anyway, mate. I'll try to somehow circumvent this problem, for whatever reason it occurs. I just noticed that my promps suddenly started to create this, for whatever reason. Thought someone here might have had similar problems. It is what it is. Don't sweat it.

    UnusNemoOct 16, 2024· 1 reaction

    @Murphy1239 I am going to take by your setting you use A1111? Try turning off (one at a time) ADeatailer and/or Hires.fix. You can also try different upscalers with Hires.fix. Basically use deductive reasoning eliminate possible problems then move forward. There are lora Detailers and your upscale can be done in a seperate utility if needed after the processing (such as Gimp or PhotoShop). Hope this helps.

    Murphy1239Oct 16, 2024· 2 reactions

    @UnusNemo Hey Nemo. I found it out to be a problem with the upscaler, actually. I went for the exact same method you suggested, by eliminating possible problems one at a time. When I changed the upscaler, it worked like a charm. :)

    UnusNemoOct 16, 2024

    @Murphy1239 I was actually leaning toward that as I have had the same issue with upscalers before. It is like specifying to high of a resolution for model and it duplicating things on you. Glad you found it and fixed it! Glad you came back and let us know so that if someone else has the problem they can get some advise from your comments. :)

    Waifus_SupremacyyDec 29, 2024· 1 reaction

    It might but i don't think that's the problem Sometimes you need to have your "Upscale By" set to 1.5 instead of 2 espeically if she's naked in a specific pose otherwise 2 belly buttons maybe even more

    Murphy1239Dec 30, 2024

    @DanielKusak I actually figured that one out as well. I have the upscale by value set to 1.65 and then in "Extra" tab upscale them by 1.22 and this gives me the resolution I'm after normally. But even with the upscaler change, it worked really good.

    UnusNemoOct 16, 2024· 17 reactions
    CivitAI

    You do not need to do anything fancy to produce Clip Skip: 2 in ComfyUI. All you need to do is add the 'Clip Set Last Layer' node in your work flow and tie it in between your Lora Loader and your negative-prompt and positive-prompt (Checkpoint if you do not use a lora loader) . Set it to -2. You are done. I read a lot of complicated tutorials that had me installing specialized nodes before I stumbled into this so obvious and easy solution.

    obuyb404
    Author
    Oct 16, 2024

    The CLIP Set Last Layer node can be removed from Workflow and connected directly, Load Checkpoint + Load LoRA + KSampler and it will work

    Image generation with CLIP Set Last Layer -2 and without this node are literally the same! This most likely means that the default model uses CLIP SKIP 2

    CLIP SKIP does not matter for SDXL, CLIP SKIP in A1111 Cannot be changed and CLIP SKIP is used by default, I never understood why? But most likely this is because you don't need to change the CLIP SKIP

    UnusNemoOct 16, 2024

    @prgfrg23  I have considered changing its location in the workflow to see if it made a difference though have not checked that out yet. You can play with the clip skip in ComfyUI. -1 one is the default. For SDXL it goes to like -36, half that for SD1.5. Though you can test it because ComfyUI will give a warning if you set it to low. Though Last Layer set to -2 is equivalent to Clip Skip: 2 in A1111. I went through a lot of nonesense and wasted time finding it out. It would have been nice to find this out before the many tutorials of using unnecessary complicated work around.

    Sometimes I play with the Last Layer to see the variances. It can be interesting at time to experiment with. The other pain was finding out how to HiRes-Fix in ComfyUI which is just about adding some upscale models and nodes. They can make a huge difference. When I have learnt enough to make it worth while I am going to write an article here about what I have found. Hopefully the next creator teaching themselves will not have to go through all the tangents I did. Though I am not complaining. A lot I learnt on those tangents is still worth while. Keep in mind I have only been doing this for like 4 weeks now.

    UnusNemoOct 16, 2024

    @prgfrg23 By the way I do not know how to cross post here on CivitAI or I would have. I posted the last two images I created to the Checkpoint I used. When I do some more I will post them here so you can get credit too. If you know how to cross post then let me know and I will do that moving forward.

    civicivicivNov 20, 2024· 20 reactions
    CivitAI

    good

    GreatHuntNov 21, 2024· 22 reactions
    CivitAI

    After having been using Ponyv6 for like 6months I'm convinced this is one of the best loras for ponyv6 (darkside that is). It offers superb base material for pony to work with. It enhances clothing, poses, environments. An all in all package that still to a large degree keeps the original style intact if you have other stuff going. It can of course be used together with things like g0thic etc. Its one downside is that it makes things dark but you can add lighter loras to compensate like expressive for example. Or lowlight on negative. Unless of course you want it dark then its a win-win. The advantage over g0thic, twilight and others that are popular for same reason are quite clear: much less impact on the general style. Making it ideal for mixing in at low values to increase prompt adherence.

    obuyb404
    Author
    Dec 9, 2024· 12 reactions
    CivitAI
    6417697Dec 11, 2024· 15 reactions
    CivitAI

    Very nice.

    wastedeep5868638Dec 12, 2024· 13 reactions
    CivitAI

    effing bad ass dude

    OnlyFamsDec 13, 2024· 19 reactions
    CivitAI

    Thank you so much! Do you have anything in the negative?

    obuyb404
    Author
    Dec 13, 2024

    For Pony, I use this Negative Prompt

    score_6_up, score_5_up, score_4_up, text, watermark, low-quality, signature, moiré pattern, downsampling, aliasing, distorted, blurry, glossy, blur, jpeg artifacts, compression artifacts, poorly drawn, low-resolution, bad, distortion, twisted, excessive, exaggerated pose, exaggerated limbs, grainy, symmetrical, duplicate, error, pattern, beginner, pixelated, fake, hyper, glitch, overexposed, high-contrast, bad-contrast

    TokalaWolfJan 7, 2025· 13 reactions
    CivitAI

    So wait, do I merge this with others?

    obuyb404
    Author
    Jan 7, 2025

    What do you mean?

    It's not that I couldn't guess, but I'd like to know for sure.

    TokalaWolfJan 7, 2025

    sorry just a moment of dumb there, I I thought I had to merge it somehow with checkpoint, im new lmao

    obuyb404
    Author
    Jan 7, 2025· 1 reaction

    @vilefox Okay, sometimes it's okay to be dumb.

    obuyb404
    Author
    Jan 10, 2025· 16 reactions
    CivitAI
    konsaiFeb 18, 2025· 12 reactions
    CivitAI

    I am new to all these yours AI... how to see whole list of styles and tags for them? THX in advance

    obuyb404
    Author
    Feb 18, 2025· 2 reactions

    Title/Link

    Trigger Words

    -

    Smooth

    Smooth 2

    Smooth Night

    dark theme, low light,

    Anime

    Anime 2

    Cold Night

    dark theme, low light,

    Smooth Anime [M]

    Smooth Anime 2 [M]

    Smooth Anime Night [M]

    dark theme, low light,

    Photo

    realistic,

    Photo 2

    raw, photo, realistic,

    Summer Days

    Summer Days 2

    Summer Night

    dark theme, low light,

    Gothic Art

    traditional media, realistic,

    Gothic Art 2

    gothic art, oil painting, traditional media, realistic,

    Concept Art

    concept art, realistic,

    Concept Art 2

    concept art, realistic,

    Concept Art Twilight [M]

    concept art, realistic,

    Concept Art Eclipse [M]

    concept art, realistic,

    Concept Art Brush

    concept art, digital art, realistic,

    Concept Art Ultimatum [M]

    concept art, digital art, realistic,

    Concept Art DarkSide [M]

    concept art, realistic, dark theme, low light,

    Oil Painting

    traditional media, realistic,

    Oil Painting Sharp

    oil painting, traditional media, realistic,

    Oil Gothic [M]

    traditional media, realistic,

    Cold Oil [M]

    oil painting, traditional media, realistic,

    Cold Oil Gothic [M]

    oil painting, traditional media, realistic,

    Impressionism Oil

    oil painting, traditional media, realistic,

    Palette Knife

    oil painting, traditional media, realistic,

    Eldritch Oil [M]

    oil painting, traditional media, realistic,

    Digital Art

    digital art, realistic,

    Line Art

    monochrome, greyscale,

    Line Art Mono

    flat color,

    lineart, monochrome, greyscale,

    Faux Oil Painting

    painterly, faux traditional media, realistic,

    Faux Oil Painting Soft

    painterly, faux traditional media, realistic,

    Watercolor Painting

    watercolor \(medium\), traditional media, realistic,

    Watercolor Anime

    watercolor \(medium\), traditional media,

    Dream Painting

    painterly, traditional media, realistic,

    Sketch Illustration

    sketch,

    monochrome, greyscale,

    Rainbow

    rainbow style, vibrant colors, vivid colors,

    Ink

    konsaiFeb 19, 2025

    @obuyb404 thank you very much!!!!

    redprince2025Feb 23, 2025· 18 reactions
    CivitAI

    this is great

    2780820Feb 26, 2025· 18 reactions
    CivitAI

    Can you give any tips on how to make style loras without baking in extra stuff (like characters, clothing, etc.)? Thanks.

    obuyb404
    Author
    Feb 28, 2025· 4 reactions

    If we take the Pony model as an example, then this model is flexible enough not to take too much of something from the dataset.

    I think the model will still take something from the images that LoRA trained on.

    The more diverse the images are, and the more images there are, the less likely it is that the model will take something extra.

    I have styles that are trained on only 4-8 images, so flexible styles don't always need a lot of images. (Especially if it's a Pony/Illustrious model)

    DhomMar 6, 2025· 24 reactions
    CivitAI

    top demais

    potecoMar 20, 2025· 22 reactions
    CivitAI

    Amazing! It works very well!

    DescargerFMMay 12, 2025

    👍🏽