Sign In

Krea 2 Inkveil Style

9

Recently loaded

Download

1 variant available

fp32 SafeTensor

inkveil.safetensors

Full precision, largest file • 447.57 MB

Verified:

Type
LoRA
Stats

37

9

Reviews
Published

Nov 19, 2026

Base Model
Training
Steps: 1,250Epochs: 17
Usage Tips
Strength: 1
Hash
AutoV2CDE7C5C34D
Trigger Words
An image in the inkveil style.

Krea 2 is licensed under the Krea 2 Community License Agreement. For more information, visit https://krea.ai/krea-2-licensing.

01_Krea2-2x-face_00067.png

A made-up style LoRA for Krea 2 that blends three games: the sumi-e ink brushwork of Ōkami, the soft watercolour washes of Gris, and the clean outlines and flat comic portraits of Hades. The result is ink-like lines of varying weight, wash colours with a paper or cloud texture, simple readable shapes and strong graphic compositions. It works on things none of the games had: fishing towns, libraries, steam trains, blacksmiths, owls, old men, and it can draw a game-style HUD when asked.

Trigger word

An image in the inkveil style.

Start every prompt with this sentence, typed into the prompt. The training captions all open this way, then describe only the content. Without the sentence the style does not apply: a prompt without it gave a photo in both test seeds.

  • Krea 2 Raw with the Krea 2 Turbo LoRA at 0.6: 8 steps, CFG 1, sampler er_sde, scheduler simple, no negative prompt

  • LoRA strength 1.0. In a strength test on one scene (0.6, 0.8, 1.0) the look was already visible at 0.6.

  • Sizes: wide such as 1600×896 for scenes, tall such as 960×1280 for portraits and single subjects.

  • Upscaling: the examples are decoded 2× with the spacepxl 2× VAE (with a face detailer that only runs when a face is found). Backgrounds stay soft because the style is soft; detail upscalers sharpen that away.

  • Write plain sentences, not tags. Krea 2 reads prompts like captions.

Example prompt

An image in the inkveil style. A rainy street in an old town at night, with glowing paper lanterns, a tea shop, a cart under a canopy and people holding umbrellas.

Opening sentence

  • An image in the inkveil style. followed by your scene

  • HUD: ask for it. The example uses A top-down view of a dark dungeon room with a glowing blue pool and a small hero fighting skeletons, with small health icons along the left edge and a boss health bar at the top. The training captions named HUD pieces the same way.

With a character LoRA

Not a strong point. Realistic or soft-anime character LoRAs bring their own look:

  • A soft anime character LoRA at 0.5–1.0 with this LoRA at 0.8–1.2: the image stayed in the character LoRA's soft anime look, no ink style.

  • Toshinden Ronron at 0.5–0.8 with this LoRA at 1.0–1.2: the character stays recognisable at 0.65–0.75 (glasses, braid, tie, skirt), and the style shows in the setting (rain, lanterns, wood), less on the figure. At Ronron 0.5 or this LoRA at 1.2 the glasses and outfit drift.

  • Tip: put a scene from the style's own world (bridge, lanterns, water) in the prompt.

Prompt tips

  • Always give a subject. A queen. fills in as a Hades-style portrait with a crown, jewels and blue skin.

  • Leaks seen in testing: red hair tufts on a stone golem (from Ōkami's wolf mane); a dark starry backdrop behind a bare-chested portrait; a plain light-grey backdrop behind full-body characters (from the Hades portraits). No wolf, no unprompted HUD and no tiny caped figure appeared in the scenes that did not ask for them.

  • Text on signs and in the HUD comes out as made-up writing.

  • photograph-like in the prompt gives an illustrated face in one seed and a photograph in another. Leave it out if you want the style.

  • Bright scenes stay bright (a sunny beach kept its blue sea and sand).

  • Interiors come out warm and dark with heavy outlines, not top-down. Top-down views only appear when asked.

Content note

A style LoRA: it changes the look, not the subject. Trained on frames from game recordings (Ōkami, Gris, Hades) and official Hades portrait art; frames with dialogue boxes, menus and loading screens were left out. A few of the portraits are bust shots with bare shoulders and chest; no fully nude images were used in training.

License

This LoRA modifies Krea 2 and is distributed under the Krea 2 Community License Agreement. By downloading it you agree to that license (see the attached NOTICE and license files). Krea 2 is licensed under the Krea 2 Community License Agreement. For more information, visit https://krea.ai/krea-2-licensing. This is not an official Krea product and is not endorsed by Krea.

Ōkami, Gris and Hades belong to their respective rights holders. This is an unofficial fan work.

If you like this LoRA, you can support me on Ko-fi.