Ashlin from Alien: MONDAY (anime, horror) by Paul "OtaKing" Johnson and ProducerMaki

All images have a prompt you can look at and choose what you are interested in!

CHECK OUT THE RECOMMENDED CFG FOR ILLUSTRIOUS!

General:

Trigger words: ot0m, as0, de1

ot0m for Ashlin in general

Recommend words:

ot0m, 1girl, green eyes, brown sidelocks hair

as0 for default outfit

Recommend words:

ot0m, as0, 1girl, belt, pouch, green crop top, fingerless gloves, grey pants, knee pads, (scar on shoulder:0.5), green eyes, brown sidelocks hair

clarification: on different models, the weight of the word scar on shoulder may vary, this recommendation is tested only on this model. On pony the scar is more stable, so you can remove this word if it is not important for you.

KEEP IN MIND IF YOU USE ADETAILER YOU SHOULD ADD A POSITIVE PROMPT THERE WITHOUT THE SCAR, OTHERWISE IT MAY START APPEARING ON FACE.

de1 for bodysuit outfit

Recommend words:

ot0m, de1, 1girl, space helmet, white pilot suit, brown hair, green eyes

Also may use like trigger/useful words:

retro artstyle for the original character design

you can also use especially on pony: 1990s \(style\), 1980s \(style\)

dark this word was used in scenes where Ashlin's design took on blue hues in both costumes, even if the scene itself wasn't particularly dark. So it might help if you're trying to take that particular character design and have the same unexpected effect, but I haven't noticed much of a change, either

xenomorph for the original xenomorph design from the short film. Caption on 32 dim versions almost only!

indoors/outdoors these words was used to emphasize what the image depicted, the ship inside or the space outside. So this will help if you want a more specific background or scenery. Caption on 32 dim versions only!

General possible problems/Non-obvious points:

Without specifying the color of clothing and hair they can sometimes take on a bluish tint, hair can pass through the space helmet and the glass itself can be barely distinguishable, a scar on the left shoulder, can and will disappear and appear on the right shoulder or both at once.

32 dim versions has a more persistent and recognizable character design, as well as here contains the style, that is, all the environment, ship, space, ship captain, dead cargo man, xenomorph. Is it possible to actually generate these? Tough question, it's worth trying to put all Ashlin trigger words in negative prompt, it might help a bit.

The differences between pony xl and Illustrious/NAI XL:

The pony is more stable and detailed, has so far more different models and other lora, as well as all the other pluses of this model. It can be used in case of original design, nsfw.

The NAI has more variation in default style, better finger work, so better suited for sfw.

*Maybe it's changed over time and is less relevant.

Illustrious/NAI XL

Recommended weight: 0.7-1 (0.9)

Recommend cfg scale: 2.5-4 (3)

at least for this noobai xl model.

v1.9 (16 dim) NAI-XL

Possible problems/Non-obvious points:

This version is the best option for noobai similar models, because there is no style dataset and the character is much more flexible to adapt to the style, as well as because of the lack of unnecessary data, the details of the character are better preserved. The model also seems to have less bleeding as well.

Also this version has not only a different dim, but also a different way of training.

v1.7 (32 dim) NAI-XL

Possible problems/Non-obvious points:

This version problems, more bleeding between character costumes, here more often one design can overlap with another in some places. The background if it is not labeled in any way will reference the station background, also in special cases it affects the labeled background, but rarely, rather if you try to generate a sci fi background it will reference the background from a short movie. The style here is more specific and unified, so you may need to reduce the weight of the lora. Otherwise, the adherence to the style prompt will be lower except for the retro artstyle which is original.

Trained on NoobAI-XL Epsilon v1

Pony XL

Recommended weight: 0.7-1.1 (0.9) for more original style 1

v1.6 (16 dim) PDXL

Possible problems/Non-obvious points:

More flexible due to smaller dim so adjust to styles more strongly, weighs less. The version is rather just different than 32 dim I wouldn't call it better but more traditional, there is also a lack of style here but the training method was about the same.

v1.5 (32 dim) PDXL

Possible problems/Non-obvious points:

The original style is more recognizable, at least on snowpony, which I recommend for generation. A little more coherent and stable in the scar means. By default, you can use this version.

Trained on Pony diffusion v6

Recommended to use adetailer, hiresfix.

And so finally I released this lora, time to publish abandoned projects. Why was the project abandoned? I decided to experiment, what if you combine, style, two (three?) characters, 2 character costumes in one lora? That's right, it would be bad. The dataset was put together back in the first week of the short's release, but it came out the way it did. Still, does that mean I released an unfinished project? Almost no. Some might have noticed that the versions are unusual and have omissions, all true, in total it took over 20 versions for this lora. So yes this version can be tried to refine more, but it is stable and works well. And if it has problems, they are unsolvable within sdxl models. I tried to convey all the details of the character and clothing, in some versions I compromised, but I did what I could, for example the scar has fine details, I always considered it an important feature of the character and paid special attention to it. But in the end I left only the good versions of the scar in the dataset, because of which as I think quantitatively there is less of it, but qualitatively it is more stable.

Should I have used lycoris for my ideas? Perhaps, I still have very little familiarity with it and dora. But more likely it's just too much for lora, or I'm still not good enough at it. I'll release another similar experiment soon, a bit more successful but a bit raw.

Search AI Models & Addons

very few results