Grandline doesnโt do polite. Itโs a style LoRA built for images that look like they belong on a premium cover โ bold, deliberate and impossible to ignore.
Point it at a warrior, a monster, a portrait, a relic, whatever's in your head โ and it comes back looking like someone with real craft sat down and illustrated it, not like it rolled off a default pipeline.
If you want to support me, post your first picture directly on this model page by clicking on "Add Post" below. And if you like this model, don't forget to ๐

๐จ The Look
Confident ink linework โ clean, deliberate contours instead of soft photographic edges
Rich, saturated color grading โ deep blacks, punchy warm/cool contrasts, no washed-out midtones
Cinematic, dramatic lighting โ rim light, glow, and shadow doing narrative work
Expressive glow & particle accents โ embers, mist, magical light, floating debris
Optional graphic paint-splatter texture โ for compositions that want extra bite
๐ผ๏ธ Where It Shines
๐ Mythic creatures & bestiary-style monster design
๐ Tarot-inspired and occult iconography
โ๏ธ Battle-scene key art and elemental magic
๐ญ Surreal, symbolic close-up portraiture
๐ก๏ธ Ornate relics, weapons & wearable artifact design
๐ง Antihero and villain character studies
๐ฎ Stylized crossover takes on characters and mascots
If it needs to feel illustrated, not photographed โ this is home turf.
๐ Trigger Word & Weight
Trigger word: in the style of mobe-st4ill
Recommended weight: 0.7โ1.0 โ sweet spot around 0.85
Works best with full descriptive sentences (subject, lighting, mood) rather than tag lists โ Krea 2 rewards specificity over keyword stacking.
๐ Example Prompts
prompt1:
A close-up shot of a Pikachu (pokemon), he wears an elegant white suit as a yakuza member and a gold necklace stylish sunglasses, and has a small teardrop shaped black tattoo on his cheek and an intricately detailed tattoo on his neck, he smoking a cigarette, it is a gangster and has a though facial expression., stylized illustration in the style of mobe-st4illprompt2:
An intensely focused Viking woman warrior with curly hair hurls a burning meteorite directly toward the viewer. The glowing sphere dominates the foreground as it leaves her outstretched hand, trailing smoke, sparks, and fragments of molten stone. Behind her lies a snowy battlefield filled with army banners, abandoned swords, and shields half-buried in the snow. Firelight clashes with the cold blue environment, creating a forceful sense of depth, motion, and impact., stylized illustration in the style of mobe-st4illprompt3:
A surreal assassin takes the form of a floating fractal eyeball. Layered geometric iris structures fold inward around a watchful central pupil, while blade-like ornamental shapes suggest stealth and danger. Suspended water droplets orbit the creature and catch faint colored light against an empty dark background. The compact composition feels strange, precise, and threatening., stylized illustration in the style of mobe-st4illprompt4:
A realistic anthropomorphic duck wizard stands ready for battle in a lush enchanted forest. The duck has sharp, determined eyes and wears a textured green hooded cloak marked by many adventures. It grips an ornate staff confidently at its side. Bright flowers and mushrooms grow among mossy roots, while dappled sunlight filters through dense leaves and casts soft shadows. Detailed feathers, worn fabric, carved wood, and vibrant woodland life balance charm, bravery, and anticipation., stylized illustration in the style of mobe-st4illprompt5:
A close, low-angle monochrome portrait of a shaved-head man with intricate scalp tattoos and a faint mustache. He wears a dark textured garment with subtle patterns. His hands, decorated with a patterned bracelet and substantial rings, are clasped together in a delicate, almost intimate gesture. Stark dramatic light casts restrained shadows across his face, hands, jewelry, and clothing. The dark textured atmosphere conveys quiet intensity, stoic contemplation, and restrained emotion., stylized illustration in the style of mobe-st4illprompt6:
Illustration, a mystical winged figure emerges from the center of a turbulent mass of charcoal-black clouds, its enormous tattered wings spread from edge to edge. A vivid red halo floats behind the hidden face and casts a narrow warm glow across the wing edges. Layered black washes, pale empty space, and bold streaks create tactile depth while keeping the nearly monochrome silhouette readable. Clouds coil around the lower body as if the figure is forming from the storm, creating solemn apocalyptic grandeur and an ancient, otherworldly presence., stylized illustration in the style of mobe-st4ill๐ท๏ธ Suggested Tags
stylized illustration ยท concept art ยท dark fantasy ยท ink and color ยท painterly
โค๏ธ Enjoyed Grandline?
If it earned a spot in your workflow, drop a review and share your renders โ nothing helps a style LoRA more than seeing what people actually build with it. Follow for updates and future versions.
Description
FAQ
Comments (9)
I really like this LoRA style, hats off.
One question regarding your prompt1 exemple tho. To my understanding Krea 2 rather like extremely accurate sentances when it comes to refering to a movie, an anime, a cartoon, etc... Also it is supposed to require capital letters to names for real people (for a comics illustrator for exemple), characters, licences, etc... So a wording such as "pikachu (pokemon)" shouldn't be optimal, right? The correct way to have the base Krea 2 model taking this sentance into account should be "with a Pikachu from the anime Pokรฉmon (1997)", if you want to get the style of the tv serie from 1997 isn't it?
Hey, thanks a lot! ๐ค๐ป๐ค Really happy the style's landing for people.
And yeah, good catch, Krea 2 does seem to "think" more in full sentences than in tags, so something like "a Pikachu from the anime Pokรฉmon (1997)" is a smart way to phrase it. ๐
Not 100% sure about the capitalization thing being an actual rule though, couldn't find Krea saying that anywhere official in the documentations. Might help, might just be a good habit, honestly not sure! But full sentences over tag-soup either way ๐
Great style! Thank you so much for sharing! ๐
๐ค๐ป๐คโค๏ธ
Now that's fantastic work. Well done, thank you!
๐ค๐ป๐คโค๏ธ
Mind sharing some training metadata basics? I'm trying some krea training myself. Curious about your LR, LR schedule, batch size, dataset size, step count, etc. Metadata viewer doesn't see anything
Sure โ I used the CivitAI default settings and only adjusted the epoch and step counts. These were the exact settings:
{
"lr": 0.0001,
"steps": 2600,
"engine": "ai-toolkit",
"epochs": 20,
"batchSize": 1,
"ecosystem": "krea2",
"keepTokens": 0,
"networkDim": 32,
"resolution": 1024,
"lrScheduler": "constant",
"minSnrGamma": null,
"noiseOffset": null,
"networkAlpha": 32,
"optimizerType": "adamw8bit",
"shuffleTokens": false,
"textEncoderLr": null,
"flipAugmentation": false,
"trainTextEncoder": false
}
The dataset contained 50 images. The training was configured for 20 epochs and 2,600 steps, but after comparing the checkpoints, I selected the checkpoint from epoch 9 as the final version.
Thanks! That's useful. Based on my attempts, it seems pretty wild that the high learning rate on constant didn't completely overbake things. Maybe I'll experiment with some bigger LRs
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.



















