A hand-inked storybook look for Krea 2: heavy brush ink, dry-brush texture, hatching and ink spatter on bone-coloured paper. The picture stays near-monochrome, with one accent colour that you choose.
The trigger is grittyink. It works on plain prompts, the kind that come out photoreal without it. In testing, six plain subjects (a lighthouse, a fisherman, a coffee still life, a night market, a bear, a spaceship on a red planet) were photographs with no LoRA and brush ink with it. My other Krea 2 utility is Dutch Angle, for a tilted camera.
HOW TO USE
Load it in ComfyUI with
LoraLoaderModelOnlyat strength 0.8.Put
grittyinkat the start of the prompt.Then write a plain prompt: subject, scene, light. No style words are needed. The training captions had none, so the style sits on the trigger.
The trigger is the switch. Without it, at 0.8, the same prompts only drift a little: the lighthouse went slightly painterly, and the fisherman and the coffee stayed close to a photograph. With it, you get the full style.
Strength 0.6 to 1.0 all work. 0.6 is already fully styled. 1.0 lays on heavier ink and is not overcooked.
Example prompt at 0.8 (1152x896, seed 7404). It is the cover image, the dragon. The fire is the one named colour:
grittyink a huge black-scaled dragon coiled around the broken top of a ruined stone tower on a mountain peak, its tail wrapped twice round the tower, wings half spread, neck stretched out, breathing a long jet of cobalt blue fire across a low grey sky, crumbling battlements under its claws, seen from below on the mountainside
ONE ACCENT COLOUR
Name one colour on one object. It becomes the only strong colour in the picture. The showcase does it eight times: a crimson umbrella, a gold sash, a violet crystal, cobalt blue fire, green eyes, orange lava, a cyan eye, a rose pink peony. Small touches can still slip in, like red trim on a saddle or the peony's green leaves.
The whole subject turns one colour. A colour word on the subject overrides the accent. rusted painted the whole robot rust red-brown, and its cyan eye became a small second colour. Calling it dull grey all over fixed it. Keep the body in greys and black, and give the colour to one object.
No accent named. The scene keeps its natural colours. In testing, the boats, the lanterns and the red planet all kept theirs.
RECOMMENDED SETTINGS
Checkpoint: RedCraft 3.0 (Krea 2). The only one I tested.
8 steps, CFG 1, sampler er_sde, scheduler simple.
About 1 MP: 896x1152 portrait, 1152x896 landscape or 1024x1024 square. All three are in the showcase.
Strength 0.8. I also tested 0.6 and 1.0, and both work.
Other checkpoints and other UIs are untested. If you try it on plain Krea 2 Turbo or somewhere else, tell me how it went.
WHAT I TESTED
Before release I ran 6 plain prompts it was not trained on, on steps 1000, 1500 and 2000 at 0.8, next to the same prompts with no LoRA:
a lighthouse keeper's cottage on a cliff at dusk
a fisherman mending a net on a pier
a coffee and a croissant by a rainy window
a night market under red paper lanterns
a bear catching a salmon in a waterfall
a small spaceship landing on a red desert planet
Step 1000 already turned all 6 into brush ink, with dry-brush, hatching and spatter. Step 2000 carries the most dry-brush, on the bear's fur and the rocks. Hands and faces held. Then strength 0.6 and 1.0 on step 2000, and a pass at 0.8 with the trigger left out.
Then the showcase at 0.8: 8 scenes, all 8 kept, from 9 images. 7 passed on the first try. The miss was the rusted robot.
a woman with a crimson umbrella on an iron footbridge in the rain
a horse archer on a snowy steppe
a woman raising a staff in a stone circle in a storm
a dragon on a ruined tower
a black hound in a foggy lane
a volcano erupting at night
an old robot in a snowy birch forest
a peony in a cracked clay jug
LIMITS
Tested only on RedCraft 3.0, and only from 0.6 to 1.0.
The base model's pseudo-lettering on signs and boat hulls comes through.
A colour word on the subject overrides the one accent.
Not tested together with character LoRAs, other style LoRAs or Dutch Angle.
MAKER'S NOTE
Trained on Krea 2 Raw with ai-toolkit (arch krea2), rank 32, batch 1, lr 2e-4, 1024 px, 2000 steps, about 56 epochs on 36 images. The file is the final step, 2000.
The dataset is synthetic: 36 images rendered on Krea 2 with no LoRA from long style prompts, 12 at each of the three sizes. People, animals, landscapes, buildings, rooms, still lifes and a few creatures. The captions carry only the trigger, the subject and the accent colour, with no style words, so the style lands on the trigger.
My previous style LoRA, at lr 1e-4, was still undertrained at step 2000. This one ran at 2e-4 and was already fully styled at step 1000. The earlier one also came through with or without its trigger, and this one switches cleanly. The higher learning rate may be why. I have not proven it.
Free to use. Credit is appreciated. The licence is on this page.
If this worked for you, a thumbs-up here helps the next person find it.

