CivArchive
    Preview 44618170
    Preview 44618169
    Preview 44618151
    Preview 44618153
    Preview 44618154
    Preview 44618152
    Preview 44618157
    Preview 44618150
    Preview 44618156
    Preview 44618155
    Preview 44618159
    Preview 44618161
    Preview 44618165
    Preview 44618162
    Preview 44618163
    Preview 44618160
    Preview 44618164
    Preview 44618166
    Preview 44618168
    Preview 44764065

    Pro version of Babes_By_Stable_Yogi Pony is available on my Patreon

    Onsite generations are permanently available on this version:
    👉 Babes_By_Stable_Yogi: https://civarchive.com/models/174687?modelVersionId=2110984

    Babes By Stable Yogi V6.5

    V6.5 is a substantial step up from V6.0.

    Anatomy, Expressions and gestures, Environments, Pose stability

    A lot of these fixes came directly from feedback identified by our Patreon community — folks running the in-development checkpoints, noting what broke, and helping us prioritize what to fix for the next version.

    ## 🚀 Be part of the next version

    We run a free image generator at discord.gg/DZEenb5wGc that ships the next-in-development checkpoint inline. Explorers generate with it, report shortfalls, and we record that feedback into the build queue for the following version.

    If you've ever wanted to be part of how a checkpoint is actually made — hop in. Free to use, no install, no setup. Generate, find issues, tell us, and you'll see your fixes land in the next release.

    ## File lineup

    Six format variants ship for general use:

    (FP16) | Default for most users — best quality / speed balance |

    (BF16) | RTX 3000+ with BF16 support, slight speed edge |

    (FP8 Scaled) | ~8 GB VRAM cards — Forge & ComfyUI native, no extra nodes |

    (NF4) | Lowest VRAM (4-6 GB) — Forge native, ComfyUI with bnb_nf4 node |

    (Q6_K) | Highest quality GGUF — needs ComfyUI-GGUF or [my Forge GGUF extension for forge ui] (https://github.com/brandulateai/sd-forge-sdxl-gguf-brandulateai) |

    DMD2 — 4 steps, CFG 1.2, LCM sampler. Fastest option |

    ## Recommended settings

    Standard variants:

    - Sampler: Euler a or DPM++ 2M

    - Scheduler: Karras

    - Steps: 25–30

    - CFG: 5–7

    DMD2 variant only:

    - Sampler: LCM

    - Steps: 4 (yes, four)

    - CFG: 1.2

    Resolution: native SDXL — 1024×1024, 832×1216, 1216×832 work cleanly.

    ## Where else to find me

    Pro variants and feedback: [Patreon](https://patreon.com/brandulate)

    (https://github.com/brandulateai)

    Free generator with the next checkpoint waiting for your feedback at Discord Community: https://discord.gg/DZEenb5wGc

    Join me on my Patreon. for exclusive perks and early access to unique resources.

    To discuss custom LoRa's or models, feel free to connect on Discord.

    • 👍 Like this model to keep me motivated and inspired to create more!

    • 💬 Drop a comment and let me know what you'd love to see next.

    • 🌟 Review this model to help me improve and make even better creations.

    • 🔔 Hit that notification bell to stay updated with my latest models and updates!

    Important Usage Tips

    Description

    Adjustments to this version

    Overall cleaner and sharper image quality

    Slightly more intricate detailing throughout the image

    Enhanced accuracy in rendering hands and feet

    Improved control over lighting and darkness

    Better alignment with given prompts

    More nuanced and refined color representation

    Noticeable enhancements to skin texture

    Slightly improved facial structures from multiple angles

    Fixed the bug that caused same face in all images.

    Steps: 20+

    Sampler: Euler a, DPm2a, DPM++ SDE Karras, DPM++ 2M SDE Karras

    CFG scale: 4+

    Size: All SDXL Compatible Sizes

    Negative Embedding: Stable_Yogis_PDXL_Negatives-neg

    Positive Embedding: Stable_Yogis_PDXL_Positives

    I am happy to listen to your feedback, Share your thoughts, comments and relevant questions.

    Join me on my Patreon for exclusive perks and early access to unique resources.

    To discuss custom Lora's or models, feel free to connect on Discord.

    FAQ

    Comments (40)

    Stable_Yogi
    Author
    Dec 10, 2024· 72 reactions
    CivitAI

    Hey everyone! We’re thinking about switching our generation model from Pony V3 to Pony V4. What do you think? React to this comment to vote:

    Heart ❤ if you’re excited for Pony V4

    Like 👍if you’d rather keep Pony V3

    Feel free to leave a comment too, we’d love to hear your thoughts!

    BigSad11Dec 11, 2024· 2 reactions

    I came here to comment. This seems like a good play. First I'd like to say, I'm a big fan of your models Yogi and love these models. With that said, v4 is VERY blurry and hazy for me. No matter how I prompt it, I can't get the model to look good 😥. Any advice would be great but for now, v3 has much more clarity for me. Let me know what you think may be causing this or if it was just simply the way it was trained. Thanks.

    Small edit, I've got some high quality images but I have to run 50-60 steps and it's not consistent. I'm a 30 step kind of man with my GPU if I can help it. (2080ti)

    Stable_Yogi
    Author
    Dec 11, 2024· 3 reactions

    @BigSad11 Thanks for the Love and feedback. Please share some image links with me , you can via discord. or civit chat.

    BigSad11Dec 11, 2024· 1 reaction

    @Stable_Yogi Will do when I get some time today

    4783700Dec 12, 2024· 1 reaction

    Take my vote with a grain of salt as I do not use Civit generator and generally prefer your Realism model over Babes. But at least in Forge Babes 4 looks like plastic. I tried quite a few samplers, schedulers, diffusion modes, with and without your embeddings and cannot get it to produce anything close to Babes 3, not even talking about Realism 3.

    Stable_Yogi
    Author
    Dec 12, 2024· 1 reaction

    @iheartlumina Noted.

    BigSad11Dec 12, 2024· 1 reaction

    @iheartlumina I shared some photos with him Iheart with pretty much the same results. Like you, I do local image generation. It might be better with Civitai's hardware. I'm sure he'll fix it in the next one. It would be cool to see an updated v4 though though if possible 😅. With all that said, v3 and realistic 3 are still my 2 favorites above all others so no complaining here. Just trying to help. Maybe our hardware generates differently which would be good for the Dev to know.

    Stable_Yogi
    Author
    Dec 12, 2024· 5 reactions

    @BigSad11  We are listening to your feedback, and working on the fixes.

    Stable_Yogi
    Author
    Dec 15, 2024· 2 reactions

    We have updated all 5 models on 15-12-2024
    Babes Pony V4 Fp16
    Babes Pony V4 Fp32
    Babes Pony V4 DMD2
    Babes Pony V4 Lightning
    Babes Pony V4 Hyper.
    if you already have downloaded, then please re-download.
    Thanks, Dont forget to share your feedback here.

    Danielkb3Dec 18, 2024· 2 reactions

    V4 32 Fix is absolutely CRUSHING IT in the Draw Things app. I use it to generate locally on iPhone and iPad. The couple I did on CivitAI with the online version turned out great also!

    BigSad11Dec 16, 2024· 21 reactions
    CivitAI

    Wow! The clarity on the Fixed version is 10x better! I don't know what you did Stable_Yogi & team but it came out great! Redownload anyone who hasn't and to anyone wondering if this model is worth the buzz, it is! Buy it!

    My personal favorite of the 4 versions (if you're deciding) is the FP32 version if you can handle the data.

    Stable_Yogi
    Author
    Dec 16, 2024· 2 reactions

    Thank you for the Feedback and testing,
    I really appreciate your time and recommending it. 🙏

    4783700Dec 16, 2024· 1 reaction

    Agreed. Huge improvement. Ancestral samplers seem to work better: Euler A (a bit soft) around 4.5 CFG or DPM++ 2S a or DPM2 a around 3 CFG. I have only tried VAE, not FP32, Hyper or Lightning. Amazing that most tokens do not move away from racial / ethnic features as common with realistic Pony checkpoints.
    My only pet peeve is the puckered anus by default. I use the token to force certain poses but not the particular look. It would be nice if future versions left that look to a more specific token.

    Stable_Yogi
    Author
    Dec 16, 2024· 1 reaction

    @iheartlumina share the tokens and the images, will see to it.

    4783700Dec 16, 2024· 1 reaction

    @Stable_Yogi Thank you for being willing to look into it. Check out this post
    https://civitai.com/posts/10388778
    First image is the default sitting position with spread legs. By adding "anus" token I force a position that is more open toward viewer (second image) But compare it to the third image (same tokens, same seed, Realism 3). See how in the last image the body part in question is ... more delicate, I guess.

    Stable_Yogi
    Author
    Dec 16, 2024· 1 reaction

    @iheartlumina got it, Noted, will fix in next update. Thanks.

    xDVxDec 20, 2024· 1 reaction

    Which checkpoint is better v4 vae fix or the v4 fp32 vae fix? like whats better for on site gen?

    Stable_Yogi
    Author
    Dec 20, 2024· 3 reactions

    @xDVx On Site Gen only Available on V4_VAE_FIX. (FP32 Is better if you can run it locally)

    xDVxDec 20, 2024· 1 reaction

    @Stable_Yogi I like the checkpoints, just wish i could run local lol. Would be neat if there was a way to have checkpoints cycle week to week.

    Danielkb3Dec 18, 2024· 23 reactions
    CivitAI

    @Stable_Yogi this is the best model I’ve tried since I first started generating images. VERY well done!

    Stable_Yogi
    Author
    Dec 18, 2024· 1 reaction

    🙏❤

    NoneMoreBlackDec 20, 2024· 19 reactions
    CivitAI

    This is my go-to model, and the Pony V4 version is the best yet! I usually generate somewhat hyper-realistic images, but this model provides excellent details and good control.

    Stable_Yogi
    Author
    Dec 20, 2024· 1 reaction

    Thanks for the feedback !

    _quietJan 5, 2025· 18 reactions
    CivitAI

    amazing, thank you, still trying to figure out how to use it with consistent characters. any one have any tips?

    silverbot01Feb 26, 2025· 1 reaction

    I know if i use too many loras or too many redundant tags i might end up with overly similar faces. While I don't like this, i enjoy some variety, you could mess with this to see where it takes you.

    The kinds of loras i find do it are ones with low training data. Usually something you don't want.

    The real answer is probably taking a resulting image you like and using tools to replicate the face in additional images (ive seen some guides, but havent tried it).

    Stable_Yogi
    Author
    Feb 26, 2025· 3 reactions

    Try making a lora and then you can have unlimited images of that character.

    kaiboshMar 5, 2025· 10 reactions
    CivitAI

    My results for v4 are just slightly different than v3 - enough that I recc. that people keep the old one around for a while. Excellent work, though - it keeps getting better.

    Stable_Yogi
    Author
    Mar 5, 2025· 1 reaction

    Thanks for the feedback.

    ruimcosme18Mar 10, 2025· 1 reaction

    i cant generate nothing can you point me some basic workflow please

    Stable_Yogi
    Author
    Mar 10, 2025· 1 reaction

    @ruimcosme18 
    try this workflow and use the DMD2 checkpoint with this.
    https://civitai.com/models/1215102/sdxldmd2workflowbystableyogi

    ruimcosme18Mar 10, 2025· 1 reaction

    thankssss

    littlefluffyballMar 25, 2025· 9 reactions
    CivitAI

    it is ok, and makes good anatomy, especially if you prompted "anu$" it will draw gigantic wrinkled a$$ hole.
    but this style is too much, it is too different, and too stylized to be realistic, same time it is not artistic to considered as stylized.

    menyan1381Apr 12, 2025· 11 reactions
    CivitAI

    Edit: Easy to direct, ipadapter works if you keep from the optional clip vision and insightvision input free. Being pony based, this model is easy to manipulate, especially face and pose. the realistic aspect is the main selling point.

    Sorry for my mistake with below comment. It was the shadow of my inexperienced self. I will leave it as reminder of my incompetence. Thank you.

    too strong on the face, hard to workwith ipadapter faceid, other than that works really well with pure generation. Really well packed and follow prompts quite well

    Stable_Yogi
    Author
    Apr 12, 2025· 1 reaction

    Thanks for the feedback !

    sparksflyai317Apr 15, 2025· 14 reactions
    CivitAI

    Really like the results of this model. The exaggeration is subtle(ish), and very appealing.

    It handles prompts for older subjects well.

    Stable_Yogi
    Author
    Apr 15, 2025· 2 reactions

    Thanks for the feedback 🙏

    summer_loveMay 2, 2025· 13 reactions
    CivitAI

    The quality and details are amazing!

    Stable_Yogi
    Author
    May 2, 2025· 2 reactions

    Thanks for the support and love much appreciated 🙏

    FutopiaMay 5, 2025· 11 reactions
    CivitAI

    One of my all time favs! Thank you!

    Stable_Yogi
    Author
    May 5, 2025· 2 reactions

    Thanks for the love and support ! 🙏