CivArchive
    Canvassolaris Artist Style [Pony|Noob|Illustrious] - Illustrious V2.6
    NSFW
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    A Lora to reproduce the style of Canvassolaris / kanvaSS0lariS / カンバスそらりす .

    Pixiv: https://www.pixiv.net/en/users/29249136

    Twitter: https://twitter.com/kanvaSS0lariS

    The style of Canvassolaris focuses on women and futanari with large bodies, thighs, breasts and comparatively small heads. It also features very large and often grotesque cocks, transformations and corruption.

    Please do not use this Lora to impersonate the artist.

    Should the Artist request it, I will delete this Lora.

    NoobAI EPS 1.1 | V2.6:

    Trained on https://civarchive.com/models/833294?modelVersionId=1116447

    Second NoobAI Epsilon-Pred version, works pretty well, style is stronger than the previous version.

    Same dataset and tagging as Pony version

    Illustrious | V2.6:

    Trained on https://civarchive.com/models/795765/illustrious-xl

    Experimental, just threw the dataset into training on this model, same tags as Pony.

    Let me know if it works well.

    V2.6:

    Improved background tagging - New Tags:

    "Casino", produces an overall orange background with green Tables.

    "Tiled wall" and "tiled floor" have been added to reduce random tiles showing up. Consider adding a "warning sign" to the wall if you want.

    "Tentacle background" to generate tentacle caves.

    "Armpit fetish" changed to just "Armpit".

    Improved tagging consistency of "Watermark" and "censored, censor bar, mosaic censor" as I had missed a handful of images in v2.5, hopefully leading to fewer random failures.

    V2.5:

    Trigger word "cnvsls" has been re-added.

    Some overall improvements to the dataset, pruned some overtagging. Removed any redundant tags like "gynomorph" as images were already tagged with the identical "intersex" tags. "cnvnun" tag has been added to the corrupted nun character.

    The overall style seems stronger than in V2.

    I am quite happy with the results for 2.5 and this will likely be the final version.

    See V2 for tagging.

    V2:

    The trigger word has been removed.

    I re-tagged the entire dataset of 500 images with e6 tags, it no longer produces random cock and balls, or other anomalies.

    The following characters were part of the training material:

    • "Sandy \(kemono friends\)",

    • "Shoebill \(kemono friends\)",

    • "African wild dog \(kemono friends\)",

    • "Dire wolf \(kemono friends\)"

    • "Nun" will produce a nun in the style of Canvassolaris, add "leg scar" and "mouth scar", for more accuracy.

    • "Hasshaku-sama", the Yokai.

    • "Selvaria bles"

    Anatomy:

    For the extra genitalia, use the tags "multi penis","2 penises","multi balls".

    "Huge glans" leads to a penis with a very large glans (surprise).

    "Armpit fetish" should lead to a focus on the unique way Canvassolaris draws armpits.

    "Ball tattoo" for a womb-tattoo-esque tattoo on the balls.

    V1:

    The trigger word for the style is "cnvsls".

    Trained on ~500 images with civitai autotagging, it will often generate censoring, futanari and multi-cocks without input, if you don't want that put "censored","futanari","multi penis" into the negative prompt.

    The Kemono Friends character Shoebill is featured in many works of Canvassolaris, and works without additional Loras.

    I recommend using Adetailer for faces.

    This is my first style Lora, made mostly for personal use, let me know what you think, I'd be happy to hear your feedback.

    Description

    Experimental, trained with the same dataset I used for Pony

    FAQ

    Comments (5)

    DrA_263Nov 25, 2024
    CivitAI

    "Style does not appear to come through quite as strong as in the Pony versions."

    Might need to run it with more steps tbh.

    In my poking and prodding with NAIXL so far, I've found that characters get trained in incredibly easily, but styles are extremely relucant, even moreso than with Pony.

    ContainmentBreach
    Author
    Nov 25, 2024· 1 reaction

    Thanks, if I end up using sticking with NAIXL, I'll train it again, with a ton more steps.

    The early adopter disadvantage is everyone still having to figure out how to best use a model.

    DrA_263Nov 25, 2024

    @ContainmentBreach

    For sure 😄

    imo it's proving better than either ILXL or Pony, but people are getting distracted by the v-pred models which seem to have progressively more issues the further they train it 🫠

    ContainmentBreach
    Author
    Dec 2, 2024· 1 reaction

    @DrA_263 Looks much better after training it for a ton more steps, thanks for the advice. ❤

    DrA_263Dec 2, 2024· 1 reaction

    @ContainmentBreach Looking great, nice work as always 😊

    LORA
    Illustrious

    Details

    Downloads
    1,089
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/25/2024
    Updated
    6/11/2026
    Deleted
    -
    Trigger Words:
    cnvsls

    Files

    Canvassolaris_Illustrious.safetensors