README.md
23.6 KB · 351 lines · markdown Raw
1 ---
2 language:
3 - en
4 base_model:
5 - krea/Krea-2-Raw
6 pipeline_tag: text-to-image
7 library_name: diffusers
8 license: other
9 license_name: krea-2-community-license
10 license_link: https://huggingface.co/krea/Krea-2-Turbo/blob/main/LICENSE.pdf
11 widget:
12 - text: A small, dark-colored cat is captured mid-stride, walking down the center
13 of a narrow, abandoned street. The street is paved and appears cracked and worn.
14 On either side of the street are tall, dilapidated buildings with visible brickwork
15 and windows. A street lamp stands on the right side. The entire image is rendered
16 in a monochromatic blue, with a distinct halftone dot pattern overlaying the scene,
17 giving it a retro or printed appearance. The focus is soft, and the lighting is
18 diffused, creating a hazy, atmospheric effect. The perspective is from ground
19 level, looking down the length of the street, which narrows into the distance.,
20 halftone texture
21 output:
22 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/00.jpg
23 - text: 'This is a digital illustration with a retro, pixelated aesthetic, depicting
24 a young boy and a dog in an abstract indoor setting. The boy, with short brown
25 hair and wearing a white t-shirt and dark shorts, is seated at a light-colored
26 table. He is leaning over the table, looking down at papers or books. His face
27 is rendered with a soft, somewhat blurry effect. To the right of the table, a
28 tan, long-haired dog with large ears is lying on a red surface, possibly a rug
29 or blanket. The dog''s head is resting on the surface, and its eyes are closed,
30 suggesting it is asleep. Its face is also rendered with a soft, somewhat distorted
31 effect. The background features reddish-pink walls and a hint of a window or doorway
32 on the left, with a hazy, abstract purple area. Scattered around the scene are
33 various pixelated elements: a blue floppy disk icon, a yellow star icon, a green
34 star icon, a blue thumbs-up icon, and a small black pixelated object near the
35 boy''s feet. The overall style is reminiscent of early computer graphics or video
36 games, with a limited color palette and a focus on simple, blocky shapes. The
37 composition is somewhat chaotic, with elements overlapping and a lack of clear
38 spatial depth. The lighting is artificial and creates strong shadows, particularly
39 on the table and the boy., low-poly 3D models'
40 output:
41 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/03.jpg
42 - text: A young woman with fair skin and blonde hair styled in curls sits on the floor
43 of a vast, empty opera hall. She is dressed in a fluffy, light pink tutu and a
44 white lace shawl draped over her shoulders. Her legs are extended forward, and
45 her hands rest on her lap. The opera hall features rows of empty, plush red velvet
46 seats arranged in a circular pattern around a central floor area. Ornate balconies
47 with decorative railings line the upper levels, and large, elaborate chandeliers
48 hang from the high ceiling, casting a soft glow. The image has a soft, romantic,
49 and slightly ethereal aesthetic, with a sense of grandeur and emptiness. The composition
50 is centered on the woman, with the vastness of the opera hall surrounding her.
51 The medium appears to be a painting, rendered with visible brushstrokes and a
52 soft focus. The camera angle is a medium shot, slightly elevated, looking down
53 towards the woman. The color palette is dominated by pastels, especially pinks,
54 whites, and creams, with touches of deep red from the seats and gold from the
55 chandeliers and ceiling. The lighting is soft and diffused, creating gentle shadows
56 and highlights., impressionist painting, visible brushstrokes
57 output:
58 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/05.jpg
59 - text: A black and white photograph captures a male rockstar performing on stage,
60 illuminated by a bright spotlight. He is holding a microphone to his open mouth,
61 with his head tilted back and his right arm raised in a dynamic pose. His figure
62 is rendered in high contrast, with sharp highlights and deep shadows. The background
63 depicts a concert setting with a drum set and various amplifiers and speakers.
64 The stage is bathed in intense light, creating a hazy, abstract effect. Silhouettes
65 of an audience are visible in the distance, some with their hands raised. The
66 overall style is gritty and energetic, with a dramatic composition focusing on
67 the performer., thermal imaging style
68 output:
69 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/06.jpg
70 - text: A black and white photograph captures a solitary man standing on a wooden
71 dock, facing away from the viewer and looking out at the vast expanse of the sea.
72 He is wearing a full-length coat and a flat-rimmed hat. The man is silhouetted
73 against the bright sky and water. The dock occupies the lower portion of the frame,
74 with the sea stretching to the horizon. The image is composed with a low camera
75 angle, emphasizing the scale of the sea and the isolation of the figure. The lighting
76 creates strong contrasts, with the man and dock appearing as dark silhouettes
77 against the lighter background., black and white photography
78 output:
79 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/07.jpg
80 - text: An anime illustration depicts a young boy and girl walking through a lush
81 forest. The boy, on the left, wears a white short-sleeved shirt, a dark tie, and
82 a blue cap. He has short brown hair and looks to his right with a curious expression.
83 The girl, on the right, wears a white dress with a blue collar and cuffs, and
84 her brown hair is tied back. She carries a woven basket over her right shoulder
85 and also looks to her right with an inquisitive gaze. The forest background is
86 filled with green foliage and trees, with sunlight filtering through the leaves.
87 Large rocks are scattered in the foreground, with a small brown bird perched on
88 a rock to the left. To the right, a small brown monkey is visible climbing a tree.
89 Red and yellow flowers add pops of color to the scene. The overall style is characteristic
90 of traditional hand-drawn animation, with soft lighting and a natural color palette.,
91 whimsical woodland creatures
92 output:
93 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/13.jpg
94 - text: vintage analog collage, central irregularly shaped snowy mountain range with
95 a section featuring distinct wavy edges, structured within a 12x16 grid of square
96 tiles, composition fragments the subject by alternating tiles with solid azure
97 blue background squares, thin white grid lines, grainy paper texture, retro aesthetic
98 of mid-century print, vibrant cyan and warm neutral tones, experimental layout,
99 tactile quality, high-contrast graphic composition
100 output:
101 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/18.jpg
102 - text: A minimalist flat-color illustration of a person wading through expansive
103 shallow ocean waves beneath a pale peach sky. The dark-skinned figure, wearing
104 an orange swim cap, light blue top, and bright green shorts, steps carefully through
105 knee-deep water. The ocean is rendered in muted mint green with delicate, thin
106 black linework detailing the continuous ripples and gentle whitecaps. Soft pinkish-peach
107 reflections echo the sky on the water's surface. A dark, jagged rock rests in
108 the lower left foreground near a pale grey shoreline. The horizon features a solid
109 purplish-blue landmass and a stylized, layered yellow and blue cloud. The high-angle
110 wide perspective emphasizes the vast negative space of the water, utilizing a
111 clean ligne claire drawing aesthetic with a subtle paper texture.
112 output:
113 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/20.jpg
114 - text: A tiny figure and a small white dog sit side-by-side in the deep green shadow
115 of a massive tree on a sloping grassy hill. The enormous tree canopy dominates
116 the upper composition, textured with thousands of stippled, light blue and yellow
117 dabs representing leaves. A sharp diagonal line divides the vibrant, sunlit yellow-green
118 grass in the foreground from the dark shade sheltering the pair. The stylized,
119 painterly landscape features flattened perspective, visible brushstrokes, and
120 intense color contrast.
121 output:
122 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/21.jpg
123 - text: A close-up portrait of a young East Asian woman with straight black hair,
124 loose strands sweeping across her fair skin, and an intense gaze. She wears a
125 light grey collared shirt with a black tie. A vibrant bouquet of pink and orange
126 lilies with lush green leaves sits in the blurred right foreground. The background
127 is a solid, striking crimson red. Soft, directional studio lighting highlights
128 her facial features, creating a high-contrast composition with a shallow depth
129 of field.
130 output:
131 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/22.jpg
132 - text: A tiny, russet-brown harvest mouse clings to a slender diagonal branch amid
133 vibrant green lobed leaves and small round buds. The mouse has soft textured fur,
134 glossy black eyes, a pink nose, fine whiskers, and delicate pink paws firmly gripping
135 the wood. In this macro photograph, an extremely shallow depth of field sharply
136 focuses on the animal's face. The deep green background dissolves into a smooth,
137 creamy bokeh, illuminated by soft, diffused natural lighting that highlights the
138 intricate details of the fur and foliage.
139 output:
140 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/23.jpg
141 - text: A dynamic digital painting of a joyful girl in a sailor uniform stretching
142 her arms high against a solid vibrant blue background. She has short dark windblown
143 hair, amber eyes, and a bright smile. She wears a white shirt, striped blue collar,
144 flowing red neckerchief, and a billowing blue pleated skirt. Expressive thick
145 brushstrokes and bold shading emphasize energetic motion.
146 output:
147 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/24.jpg
148 - text: stylized digital painting of a dark convertible on a winding coastal cliff
149 road, high-angle perspective, blocky painterly brushstrokes, golden hour sunlight
150 hitting rocky orange terrain and green vegetation, flock of white abstract birds
151 flying in foreground, blinding bright sun reflection on vast ocean, vibrant warm
152 color palette, sharp graphic shadows
153 output:
154 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/25.jpg
155 - text: A stylized jungle illustration densely packed with oversized flora and surreal
156 characters, rendered with smooth geometric shapes and granular stippled shading.
157 Two pale figures with flowing, star-speckled black hair navigate the lush environment
158 in blue garments. On the left, a figure grasps a vine as a white, long-beaked
159 bird perches on their outstretched hand. On the right, the second figure reclines
160 beside a sleek, pinkish-orange fox. The dense surroundings feature sweeping green
161 stalks and colossal blooms in brilliant golden yellow, coral pink, and deep red.
162 A second white bird emerges from the lower foliage. The vibrant composition forms
163 a seamless tapestry, utilizing rich colors and volumetric grain to create a dreamlike,
164 textured depth.
165 output:
166 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/27.jpg
167 - text: A surreal retro-futuristic space scene features liquid chrome forming an abstract
168 face merging with a glowing planetary horizon. The foreground is dominated by
169 swirling, highly reflective metallic fluid that distorts into a stylized, melting
170 facial profile with deep shadows and bright silver highlights. This undulating
171 chrome form rests against the curved, atmospheric edge of a massive planet bathed
172 in a soft electric blue and purple glow. Above the primary planet, a smaller eclipsed
173 celestial sphere sits in the upper center, crowned by a sharp, cross-shaped starburst
174 flare. Two additional radiant flares burst from the left and right edges of the
175 horizon. Set against a deep black starfield, the artwork employs a vintage 1980s
176 airbrush aesthetic with smooth gradients, ethereal lighting, and high-contrast
177 metallic rendering.
178 output:
179 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/28.jpg
180 - text: An extreme close-up portrait featuring pale, freckled skin and a single blue
181 eye wrapped in reflective metallic gold ribbons. Thin gold strips crisscross diagonally
182 over the cheek and forehead, casting sharp, hard shadows onto the face. Strands
183 of copper hair frame the top edge while the left ear softly blurs out of focus.
184 Harsh, direct lighting highlights intricate skin pores and bright golden reflections,
185 isolating the brightly lit features against a pitch-black background in a bold,
186 high-contrast macro editorial style.
187 output:
188 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/29.jpg
189 - text: Stylized digital painting of a menacing jester figure rendered with bold,
190 expressive brushstrokes and a vibrant, almost psychedelic color palette against
191 a pitch-black background. Dynamic low-angle perspective forces a dramatic, imposing
192 composition as the character leans forward, one leg raised high. The jester wears
193 a classic multi-pointed hat with bells, a ruffled collar, puffed sleeves, harlequin-patterned
194 shorts in muted gold and dark brown, and striped tights in alternating shades
195 of purple, blue, and chartreuse. A heavily textured, flowing cape billows outward
196 to the left, decorated with abstract, fluid patterns of saturated purples, greens,
197 and iridescent hues resembling oil slicks or marbled paper. The figure's face
198 is completely obscured, appearing as a smooth, faceless, pale mauve mask with
199 a single, glowing bright white point of light in the center. In its right hand,
200 clad in a grey-blue gauntlet, the jester grips a massive, ornate sword with a
201 wide, glowing, ethereal white blade, its crossguard intricately sculpted. Lighting
202 is dramatic and theatrical, casting strong shadows and highlighting the painterly
203 texture, giving the artwork a dark fantasy, surreal aesthetic reminiscent of concept
204 art.
205 output:
206 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/30.jpg
207 - text: A surreal black-and-white ink illustration of three interlocking, heavily
208 wrinkled elderly faces merging into a landscape. The top face covers one eye,
209 crowned by dense leaves, a live bird, and a skeletal bird. It flows into a profile
210 face and a third face featuring a solid black eye and a hand on its cheek. The
211 bottom neck plunges into a cross-section of earth, morphing into swirling subterranean
212 roots, buried bones, and abstract organic forms. Above ground, weathered wooden
213 cabins and tall grass flank the facial monolith. Meticulous stippling and cross-hatching
214 define the high-contrast, intricate vertical composition.
215 output:
216 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/32.jpg
217 - text: 1990s vintage anime style cel animation, densely packed crowd of teenagers
218 in summer uniforms, central boy with short black hair raising a clenched right
219 fist, squinting one eye with a determined expression, wearing a white short-sleeve
220 shirt and solid green necktie, surrounding students looking in various directions,
221 girls in white sailor blouses with green striped collars and neckerchiefs, light
222 blue skirts and trousers, tightly framed medium shot, flat shading, soft muted
223 retro.
224 output:
225 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/33.jpg
226 - text: extreme close-up of a woman's face partially obscured by tousled dark brown
227 hair, soft parted lips, smooth skin on lower cheek and jawline, stray hair strands
228 falling loosely across the nose, deep moody shadows enveloping the left frame,
229 cinematic warm lighting, delicate highlights on the mouth, muted earthy color
230 palette, sepia-toned warmth, intimate portrait photography, macro lens, shallow
231 depth of field, distinct film grain texture, vintage atmospheric aesthetic
232 output:
233 url: https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/35.jpg
234 ---
235
236 # Krea 2 Text-to-Image Model
237
238 ![Krea 2 sample outputs](https://raw.githubusercontent.com/krea-ai/krea-2/main/assets/hf_samples/turbo/header.jpg)
239
240 ## Inference with the official codebase
241
242 1. Setup the official Krea 2 [codebase](https://github.com/krea-ai/krea-2)
243 2. Download `turbo.safetensors` in this repo
244 3. `export OSS_TURBO=<path-to-turbo.safetensors>`
245 Run inference:
246 ```bash
247 uv run inference.py "a fox walking in the snow" \
248 --checkpoint oss_turbo --steps 8 --cfg 0.0 --mu 1.15 --width 2048 --height 2048
249 ```
250
251 ## Inference with diffusers
252
253 Install diffusers from source (for `Krea2Pipeline`):
254
255 ```bash
256 pip install git+https://github.com/huggingface/diffusers.git
257 ```
258
259 ```python
260 import torch
261 from diffusers import Krea2Pipeline
262
263 pipe = Krea2Pipeline.from_pretrained("krea/Krea-2-Turbo", torch_dtype=torch.bfloat16).to("cuda")
264 image = pipe("a fox in the snow", num_inference_steps=8, guidance_scale=0.0).images[0]
265 image.save("krea2.png")
266 ```
267
268 ## Inference with SGLang
269
270 Install SGLang from source (https://github.com/sgl-project/sglang)
271
272 From the CLI:
273 ```bash
274 sglang generate --model-path krea/Krea-2-Turbo \
275 --prompt "a red fox sitting in fresh snow, golden hour, photorealistic" \
276 --num-inference-steps 8 --height 1024 --width 1024 --save-output
277 ```
278
279 See the full SGLang Krea 2 Cookbook [here](https://docs.sglang.io/cookbook/diffusion/Krea/Krea-2)
280
281 ## Model Overview
282
283 - **Model Name:** Krea 2
284 - **Version:** v1.0
285 - **Release Date:** June 22, 2026
286 - **Model Type:** Text-to-image diffusion model
287 - **Architecture:** Diffusion Transformer with 12 billion parameters
288 - **License:** Krea 2 Community License
289 - **Release Format:** Open-weight release and Krea-hosted product integrations
290 - **Model Developer:** Krea.ai, Inc.
291
292 ## Model Family and Release Checkpoints
293
294 This model card covers the Krea 2 model family, including the following release checkpoints:
295
296 - **Krea 2 Raw:** Base release checkpoint, prior to additional post-training and fine-tuning.
297 - **Krea 2 Turbo:** Post-trained release checkpoint with additional fine-tuning and distillation.
298
299 ## Capabilities and Intended use
300
301 Krea 2 is a text-to-image diffusion model that generates images from natural-language text descriptions. The model is designed to support creative, commercial, developer, and research use cases, including image generation, concepting, design exploration, visual production workflows, and integration into applications and creative tools.
302
303 ## Out-of-Scope Uses
304
305 This model is not intended or designed for uses that violate applicable law or regulations, infringe or misappropriate third-party rights, generate or facilitate unlawful or harmful content (including CSAM, NCII, harassment or defamation), or support fully automated decision-making that adversely affects legal rights of individuals. This summary is non-exhaustive. Use of Krea 2 is subject to the Krea 2 Community License Agreement and must comply with the Acceptable Use Policy. In the event of any conflict, the Krea Acceptable Use Policy and Krea 2 Community License control.
306
307 ## Training Data
308
309 This model was developed using a combination of publicly available data, data licensed from third-party providers, and synthetic data generated through proprietary methods. The training data includes images and their associated captions or text descriptions.
310
311 Prior to training, data was filtered to remove certain categories of harmful content and reduce low-quality, duplicative, or irrelevant data. Krea also used curated and synthetic training data selected to improve prompt following, visual quality, and alignment with intended use cases.
312
313 ## Safety Measures
314
315 We implemented safety measures across the full model development lifecycle. We applied targeted fine-tuning techniques to reduce the model's susceptibility to generating harmful content in response to both direct and adversarial prompts, and we conducted multiple rounds of internal and external safety evaluation before release.
316
317 For Krea's hosted products incorporating Krea 2, we deploy input and output classifiers using a combination of proprietary and third-party detection tools to flag or block policy-violating prompts and generated images.
318
319 Because this is also an open-weights release, Krea does not control downstream deployment of the model. Under the Krea 2 Community License, deployers are required to implement content filtering measures or equivalent review processes to prevent the generation or distribution of unlawful or policy-violating content appropriate to their use case. Deployers who fail to implement required safeguards are in breach of the license. See the license for details.
320
321 We conducted multiple rounds of internal and external safety evaluations before release, including adversarial testing designed to assess the model's resilience to attempts to elicit harmful or policy-violating outputs. Testing covered sexually explicit content, non-consensual intimate imagery, child-safety risks, and other high-risk content categories. Based on these evaluations, the release checkpoints demonstrated high resilience against violative inputs across the tested risk categories.
322
323 Krea maintains reporting channels for harmful, illegal, or policy-violating outputs at safety@krea.ai. Reports involving potential CSAM are escalated to NCMEC as required by law. Krea reserves the right to update model weights or revoke access in response to identified misuse patterns.
324
325 ## Risks and Limitations
326
327 Krea 2 is a new technology and there are risks associated with its use. Testing conducted to date has not covered, nor could it cover, all possible scenarios. The model's potential outputs cannot be predicted in advance and may, in some instances, produce inaccurate, objectionable, or otherwise undesirable outputs.
328
329 This model is not intended to provide factual information. The model may fail to generate output that matches the prompt, and prompt following may be influenced by prompt style, specificity, language, and phrasing.
330
331 Before deploying any application using this model, developers should perform safety testing and tuning tailored to their specific application and must implement safeguards required by the Krea 2 Community License.
332
333 ## License and Outputs
334
335 Krea does not claim copyright or other intellectual property rights over content generated by users of this model. Users are solely responsible for their outputs and any subsequent use of those outputs. As with other generative tools, the nature of a user's inputs influences the outputs produced, and prompts may produce images that implicate third-party rights. Users are solely responsible for assessing and addressing those risks. See the Krea 2 Community License for more information.
336
337 ## Reporting
338
339 To report harmful, illegal, or policy-violating outputs generated by this model:
340
341 - **Email:** safety@krea.ai
342
343 Reports involving potential CSAM will be escalated to the National Center for Missing & Exploited Children (NCMEC) as required by law.
344
345 ## Contact
346
347 - **General inquiries:** support@krea.ai
348 - **Safety and abuse:** safety@krea.ai
349 - **Model card version:** 1.0
350 - **Last updated:** June 22, 2026
351