📄 Krea 2 Origami World turns a whole scene into a folded-paper world: the subject, the ground, the water, the props and the room are made of creased, layered paper, shot like a real miniature diorama.
✨ Trigger word: orgmworld (put it at the start of the prompt)
🧩 Base model: Krea 2 (tested on Krea 2 Turbo; trained on Krea 2 Raw)
⚙️ Settings that worked for me (Turbo)
• LoRA strength 1.0
• 10 steps, CFG 1, sampler euler, scheduler simple
• Size tested: 1088 x 1920 (portrait)
• Strength 0.6 was too weak in my Raw tests (I did not test 0.6 on Turbo). 1.3 to 1.6 gave more paper in clothes and rooms (tested on only 4 prompts, with a slightly longer-trained checkpoint).
🖼️ Example prompts
• orgmworld, a black and white penguin with orange feet standing on a pale blue ice floe, white and blue icebergs against a lavender sky, three-quarter view
• orgmworld, a red locomotive with a black chimney and a blue passenger carriage on gray tracks, close elevated three-quarter view against a pale mint background
• orgmworld, a quiet city street at dusk with parked cars and tall buildings
• orgmworld, a bright home kitchen with a wooden table, a bowl of fruit, copper pots on the wall and a window with morning light
• orgmworld, a snow-covered mountain range reflected in a calm lake with pine trees along the shore at sunrise
Short plain prompts like these already work; you do not need a long style paragraph.
📊 What my tests showed (honest numbers)
• 10 fixed prompts, same seeds, Turbo: base 0/10 look like folded paper, with this LoRA at 1.0: 7/10 (4 animal/object scenes + street + kitchen + mountain lake all read as paper).
• The 3 people prompts did not turn into paper with plain prompts (see limits).
• The same LoRA scored 6/10 on the Raw model, so Turbo is the recommended base.
⚠️ Known limits
• People and faces stay photographic. With a plain prompt the clothes, props and room become paper but the face and hands stay photo-like. If you want a paper figure, write it in the prompt ("the whole scene including the woman, her hair, her skin and her clothes is folded paper too") - the base model can do that too, so this LoRA is mainly for scenes, animals, objects and places.
• Faces can look a bit older with this LoRA than without it.
• A blurred window or sky backdrop can stay photographic.
• Small training set, one look: it is a style, not a guarantee for every subject.
🛠️ About the training
• 24 synthetic images (animals, objects, places and a few adult women) made with Krea 2 by me, each with a written caption; no scraped, third-party or real-person images.
• LoRA rank 32 on Krea 2 Raw, 720 training steps (7.5 epochs), trained locally with musubi-tuner.
• 5 images were held out of training, and 10 fixed test prompts (4 taken from the held-out set) with the same seeds were used to judge the result and pick this checkpoint.
📜 Licence and notice
Krea 2 is licensed under the Krea 2 Community License Agreement. For more information, visit https://krea.ai/krea-2-licensing.
This LoRA is a modified version (a Derivative) of the Krea 2 model. It is not an official Krea product and is not endorsed by Krea. Use of this LoRA and of the images it makes is subject to that Agreement and to the Krea Acceptable Use Policy (https://krea.ai/krea-2-use-policy). Commercial use is only allowed under the Agreement's company-wide revenue limit of US$1,000,000 per year (section 2.3).
🙌 If you make something with it, post it with the resource linked. Tell me what works and what does not.