Not loaded
Download
1 variant available
bf16 SafeTensor
krea2_mhDreamy(Krea2梦幻风格)-AURORAL-1.1_1500步.safetensors
BF16, good balance • 218 MB
Verified: 10 days ago
0
1
Created on Civitai
mhDreamy is a dreamy cinematic style LoRA for Krea 2. It gives images that soft-focus film-still feeling: warm golden practical lights blooming through haze, gentle highlight roll-off, deep-but-clean shadows, rim-lit silhouettes, a touch of film grain and atmosphere. Think "a frame from a dreamy arthouse film" rather than a flat digital illustration.
It does not force a subject. Point it at a person, an interior, a street, an animal, or a landscape and it re-lights the scene instead of repainting it.
How to use
TriggermhDreamy, dreamy golden-lit ethereal illustration Weight (Turbo / quantized)2.0 — int8/nf4/fp8 checkpoints swallow part of the delta Weight (raw bf16)1.0 – 1.5 Suggested negativestext, watermark, signboard, letters, signatureCaption style that works best — content first, style anchor last:
<what you want to see>. mhDreamy, dreamy golden-lit ethereal illustration
Example prompts
Fisherman
An elderly fisherman with a white beard mends a torn net on the deck of a small wooden boat while a lantern burns beside his knee. Thick fog drifts over the dark water behind him in a wide shot. mhDreamy, dreamy golden-lit ethereal illustration
Night street
A solitary hooded figure stands on a wet, reflective street under blooming cherry trees at night, lit by soft streetlights and car headlights. mhDreamy, dreamy golden-lit ethereal illustration
Bar interior
A woman in a black feather gown sits alone at a dimly lit bar, bottles lined up behind her under warm lantern light. mhDreamy, dreamy golden-lit ethereal illustration
Where it shines
- Fog, rain, night streets, candlelit interiors, golden-hour exteriors — anywhere light has something to travel through.
- Portraits and character shots where you want cinematic separation between subject and background.
- Painterly-but-grounded illustration; it keeps detail while softening the edges of the world.
Honest notes
- Anime drifts. On clean cel-shaded anime it pushes toward thick-paint / semi-realistic rendering. For anime, keep the weight under 0.6 — or skip the LoRA and just use the prompt anchor.
- Trigger can leak into the picture. If the prompt contains
mhDreamyand the scene has a natural text surface (shop signs, posters), the word may get painted in. The negative prompt above fixes it. - Hands meeting props (nets, hilts, instruments, glasses) can occasionally melt — normal for a style LoRA of this size.
- Style is tonality-specific: subjects far from its palette (bright modern daytime cityscapes) move very little.
File: bf16 safetensors, 218 MB. Base model: Krea 2. Trained with ai-toolkit — 2 rounds (2000 steps @ 1e-4, then 1500 steps @ 5e-5), rank 32 / alpha 16, 1024px.
Comments and example images are welcome — if you make something good with it, post it.