About CRM Crisis Anima V1.0*

*Or Crisis Anima for short, CRM-CA even shorter

CRM-CA aims to produce a painterly, semi-real and creamy look with a hint of mythic. It uses natural language prompts and can be very good at faithfully reproducing details.

Proudly introducing the predecessor models:

The time has come, although bittersweet. I knew this day would come, when I'd look at SDXL and say "Well, why keep trying to use this old thing?" Anima is better.

"BUT surely you can't be turning your back on all that work you've done in SDXL! Right?!"

Nearly every image came out better than SDXL, so there.

And don't call me Shirley.

The Source Models:

Base Model:

Animosity by Lonecatone23 - a vibrant model from a creator I know and trust, the #1 Tool Creator and someone I can rely on when I need help with my workflows. I used his workflow for my entire checkpoint gallery! Lonecat also knows how to make stellar checkpoints and I've been taking notes!

Added Model:

Real Dream Anime Beta v1 by sinatra- This is a great looking model that had the permissions and the look I needed - no outlines, approaching photorealism. Thank you sinatra for providing this model. Sincerely.

After that, I didn't have a Bright Style anime to put negatively against the merge but I found some others, namely Typhatena Style (-.23) and anima style: s1w00p.(-.33) To push it over the edge and ensure no outlines, I added Rigid3D style at +.35. All of these were compatible with CRM's non-restrictive usage rights which is part of why I used them.

People, fellow Civitans, hear me out on this. When you make a merge, wisely select and apply a style LoRA negatively. This is the way to get rid of watermarks that would otherwise be inherent in the model.

A word about Prompting in Anima

You should use natural language prompts. This model was designed to create a painterly style. I find that concluding the positive prompt with the following makes a noticeable difference:

Photorealistic and highly detailed.

In the negative, I also find that using "fake" has some result, and I also like to use "drawing" and for the most part my negative prompt looks like this:

worst quality, drawing, low quality, score_1, score_2, score_3, bad anatomy

About Crisis V2.0

I just want to start by saying, I never thought Crisis would become such a popular model, and once I had made it, I struggled to recapture or even guess just what it was that made it so popular. The thing is, I felt it needed a little more structure. And if I got it wrong, I don't know, and I need your feedback.

This checkpoint gives the same painterly style, with the same approach using negative Bright Style: in all testing I found it mandatory and this is the core of this model's design. I also found that this negative application helped to dispel stubborn watermarks, that came from the origin checkpoints (which one, or both, I didn't check) but watermarks may appear if you use LoRAs that are watermark contaminated. Also, in testing, I discovered that if I increased the number of steps for generation, the style would change from realistic anime to anime, and so to stabilize it, I gave it a touch of anime.

This checkpoint is a merge using the following super-popular and awesome Checkpoints:
YiH: VANTABLACK by LatteLeopard - The Yiff in Hell checkpoint is legendary, for fans of Anthro and those interested in NSFW, this checkpoint has you covered

SweetHoneyv2 by Mystifying - I love this checkpoint and really want to thank Mystifying for it. It gives a non-outlined anime look that is really capable and definitely up on my list.

A note about this gallery: All images were made without upscale or other postprocessing EXCEPT the Bee Girl image. Obviously some tiny details don't usually come out well without upscaling but this was avoided in order to give you a good idea of what a straight generation would look like.

Crisis V2 is very capable of switching styles without even using LoRAs.
These are a few tags you can use that will alter the style, particularly of the face:

anthro, toon style, retro style, anime source, realistic

For example, you may set your quality presets like so:

masterpiece, anime source, best quality, amazing quality, 4k, very aesthetic, high resolution, ultra-detailed, absurdres, newest,

And insert toon style, or retro style, or realistic, in place of anime source. I like to put it right there second, right after masterpiece. I always first want to tell the model it's job is to create a masterpiece.

This checkpoint is very good at understanding natural language so feel free to change it up and use that either instead of or alongside booru tags.

If you are trying to get a complex generation to come out right and you want to avoid spending compute on background and keep things focused on the character, here are some tags you can use, but I would avoid using all of them at the same time.


simple background, dark background, white background, abstract background

About DairyLand

This model is based upon Crisis and merged with some other models and gently incorporated another LoRA to go with the negative Bright Style application. Creamy Ranch Mix creates a non-outlined and creamy look. Crisis has a versatile range from anime to toon style while DairyLand is firmly within anime semi-real territory with enhanced detail, shading and contrast.

I have been absolutely amazed at how many people have been using CRM-Crisis and felt that it was time to release a sequel. Those who prefer the more toony style can continue to use CRM-Crisis: I intend to keep it up as long as it makes sense to. DairlyLand offers a less stylized mode that remains anthro-capable, can generate mature females, muscular males, and other subjects, it can effortlessly conjure up steamy scenes and gorgeous landscapes.

Dairyland creates a painterly, shaded semi-real style: no outlines unless a style LoRA is used. Some anthro subjects will still look toony but otherwise it tends to anime semi-realism faces.

The gallery for this model was generated with Euler a - Simple, using a CFG of 5-6.4. I recommend a CFG of under 5 unless you are using Skimmed CFG and Dynamic Thresholding. A CFG of 6 or more will create rather vivid and perhaps overly colorful renders, and the image gets further degraded by over contrast if CFG is further increased.

Generation Guide:

Euler A and Simple works fine. This model does not have as much flexibility with various sampler/scheduler combos as, say, WaiSnackMix. Feel free to experiment.

Dairyland works well with a variety of quality presets, which can be found in my workflows, but in general:

Positive:

masterpiece, anime source, best quality, amazing quality, 4k, very aesthetic, high resolution, ultra-detailed, absurdres, newest,

Negative:

lowres, bad quality, worst quality, bad anatomy, ugly, poorly drawn, blurry,  artist name, signature, watermark,

This model tends to create clean renders although sometimes in the details of the face (whiskers/freckles) a slight discoloration may emerge. For this reason, and in general, use of an upscaler is strongly recommended.

About Crisis

This is an experimental model that I hope punches above its weight. It is a daring model, which creates painterly images drawing upon the following checkpoints:
Yiff in Hell - Tended - the base model
Honey Sweet Mix Illustrious - the added model

Bright Anime style: -.31 strength

I stumbled upon this blend by accident to be honest, I had noticed how well the LoRA worked on another model and started testing it, and it looked good and stable. So, I kept tweaking until it made good with the fingers a few times. I don't know how the LoRA will affect it.

This model creates a creamy, painterly anime style that is versatile and stable to a degree. Instability may come from its Yiff roots, so if you want anthro, or not, make sure the prompt is clear. The model can also randomly change certain clothes colors, assigning the opposite color.

I called this one Crisis because it has that baked in LoRA. It can cause certain things to be opposite, sometimes -- such as clothes color.

If you are using an upscaler, you do not need a fancy one, but you should incorporate a color match node, particularly if you are upscaling with this model.

Quality Tags Guide

Positive Prompt (slightly adds detail/refinement/shading quality, purely optional)

masterpiece, amazing quality, best quality, ultra-detailed,

For greater realism, fleshy lips, significant reduction in size of anime eyes:

masterpiece, realistic, amazing quality, best quality, ultra-detailed,

Replace with the following for a slight adjustment to toon style

toon style,

Other tags that affect generation, albeit slightly:

anime style, retro style, kawaii style

Tags that will profoundly alter the style, if used all together:

vibrant hues, dynamic pose, blue palette, realistic, abstract background,

Search AI Models & Addons

very few results