Lacking models or Loras etc. that truly support 3D, Anime and Cartoons

I’ve been waiting for a new image model to be added to Venice’s selection for a few months that could consistently generate anime scenes when prompted to do so, and do so without the model becoming confused.

For example when prompting Flux, Pony and SD 3.5 to generate a image of a girl I’d detailed, along with a detailed background, with the image depicting the girl crouching in front of a stray cat, a small bowl of milk between them and the cat drinking from the bowl. I’d get the girl being warped, transformed into a cat hybrid, the image rarely ever being in anime, a part of the image being in anime, the milk defying gravity to reach the girl, the girl drinking from the bowl etc.

This is despite having negative prompts against such things, using various styles to see if the updates had made the styles functional (apparently they’re even less so now) and looking up the model via the link from prompting instructions and being up to date on the Venice blog or a myriad of other issues. This is something I noticed when moving along the SD upgrade path, the support for everything not realistic fell by the wayside more and more.

The Styles also don’t work well at all, maybe it’s because I use detailed prompts but I have had very little success in getting them to impact an image generation. Most of the time even selecting a styling for the topic styles just generates a normal real life looking image.

None of the models more than have token support for non-realistic styles, Flux and Pony were generally created to have a versatile model focused on realism, Stable Diffusion has had everything other then real looking images tuned out of it over time as far as I understand it.

There’s also no feature to allow the use of a selection of LoRA, LyCORIS etc which would be far better than using the current style functions, and I imagine, would also not be very resource intensive, though could be troublesome to set up on the API? I don’t know, there must be a reason that each model doesn’t have an at least small selection of popular LoRA etc.

For the most part real life looking images are fine, they tend to be more popular, but it’s annoying to have more incentive to go back to doing things locally again. With that in mind, having more image generative models really shouldn’t do much in terms of resources given they’re not any more demanding than the other generative models which aren’t that demanding in general.

Anyway, here’s some models that can add variety to Venice’s current selections, I’ll give a couple of options and whoever is implementing the new models can decide on which they feel best suits Venice’s services. This list is definitely not definitive but just my picks based on my subjective opinion and experiences, for example I selected the anime models avoiding the haloing yellow lighting often present in them, opting for models with less washed out colors. All models are NSFW capable to match Venice’s service and have been updated or released in the last year, they all also use Pony as a base and I could run them locally on a graphics card a few generations old, there are also no licensing requirements noted on the model page.

3d

NN - Pony 3D Mix

https://civitai.com/models/828393/nn-pony-3d-mix

If not than the next best I’ve found is

Solus Mix

https://civitai.com/models/738298/solus-mix

Anime

AutismMix SDXL

https://civitai.com/models/288584/autismmix-sdxl

If not than the next best I’ve found is

Prefect Pony XL

https://civitai.com/models/439889/prefect-pony-xl

Cartoon

Toonify

https://civitai.com/models/36281/toonify

Generally, I thought an anime model would be soon to come as I waited, it’s one of the most popular genre in image generation and why early Stable Diffusion launched with the feature. It later being removed due to concerns for how overwhelming the NSFW anime content had become and the threat it posed to women, they’d already gotten some bad press regarding the publics view of artistic theft etc and they didn’t want more controversy. With that in mind I think this is an important addition to Venice when it comes to anime.

When it comes to 3d, there have been some significant developments when it comes to 3d and the use of AI in video games. A lot of popular characters will be represented in this art style and video game development is becoming more accessible, especially so with the coding models even Venice offers. Allowing for developers to create imagines of fictional characters, especially to go along with the Character function Venice offers, only makes sense.

For Cartoons, well at this point why not? May as well cover the 3 main non-realistic art styles when we have 5 covering realistic alone.