Saya no Uta版本Saya1e4-000010 (ID: 72145)

As the title says, Saya no Uta style model that includes all main characters except for Tsukuba You - perhaps most notably it can reproduce the meaty backgrounds from the visual novel. Should get the characters right with eye/hair color and length alone, but refer to grid and bottom of the description if encountering issues. As with all my loras, grid images not cherrypicked and definitely not the best it can generate.

Should work with other loras without much issue, might have to adjust weight though, as seen on the Meiya one - too big for Civit, so link to full resolution here https://files.catbox.moe/tdieli.png Better examples of how the model can be used for the meaty backgrounds can be found on my pixiv under the same nickname.

Regarding the fuckton of versions:

• 1e-4 ones are the newest and probably best ones, I picked 08 as the main one, but it's not necessarily better than 06 or 10.

• 7e-5 ones are decently new and an okay'ish alternative. Noiseoffset has its uses (better at light and darkness than regular) and seems surprisingly good on the grids.

• The older ones have a slightly different style to them which might be preferable, but they may be less responsive to prompts and smear eyes more - I believe "warm04" turned out best out of them, and "restarts5000" might be closest to the novel, but also overbaked and hard to use.

If you're interested in the meaty backgrounds, there's a chance the old ones do it better/in a different way (I have not tested this enough on the new versions). The one you want to pay special attention to is the "noiseoffset" one because it's trained to be able to handle light and darkness better, thus should be better suited towards horror. Included a new set of grids for that one on its version page.

Meatscape prompts out of a random file

• "scenery, no humans, red theme, guro, horror (theme), outdoors, indoors, stairs, door, monster, hallway, blood, horror \(theme\), surreal"

play with weights in order to obtain the desired effect - for example the weight and inclusion of "monster" might cause monsters/skulls/faces to appear somewhere or deform bodies, though "guro" might also be affecting the last one. Adjusting the text encoder separately will have an effect on this aswell - if you lower it, pretty sure the AI will move closer to its original understanding of these prompts so there's a lot of room for experiment.

Full character prompts for their in-game outfits (you do not have to use all of them, this is simply how they were tagged for reference)

• "Saya, 1girl, long hair, ribbon, bare shoulders, very long hair, green eyes, collarbone, black hair, white dress, flat chest, hair flaps, sundress, small breasts, black ribbon, sleeveless dress"

• "Fuminori, solo, bangs, shirt, black hair, long sleeves, 1boy, jacket, white shirt, black eyes, pocket, open jacket, buttons, unbuttoned, short hair, breast pocket, green jacket, undershirt, collared shirt"

• "Kouji, undershirt, long sleeves, 1boy, jacket, grey hair, collared shirt, pants, open jacket, black pants, hands in pockets, collarbone, brown eyes, short hair"

• "Ryouko, 1girl, long hair, long sleeves, ponytail, earrings, green hair, glasses, pants, turtleneck, jeans, leather, open coat, trench coat, sweater, black eyes, parted bangs, zipper, sidelocks"

描述:

训练词语: same as others