This SDK is based on the official API documentation.
Join our discord server for help:
pip install novita-client
- fine tune example
- cleanup
- controlnet
- img2img
- img2video
- inpainting
- instantid
- merge-face
- model-search
- reimagine
- remove-background
- remove-text
- replace-background
- txt2img-with-hiresfix
- txt2img-with-lora
- txt2img-with-refiner
- txt2video
import os
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.cleanup(
image="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png",
mask="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
)
base64_to_image(res.image_file).save("./cleanup.png")
#!/usr/bin/env python
# -*- coding: UTF-8 -*-
import os
from novita_client import NovitaClient, Img2ImgV3Request, Img2ImgV3ControlNetUnit, ControlnetUnit, Samplers, Img2ImgV3Embedding
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.img2img_v3(
input_image="https://img.freepik.com/premium-photo/close-up-dogs-face-with-big-smile-generative-ai_900101-62851.jpg",
model_name="dreamshaper_8_93211.safetensors",
prompt="a cute dog",
sampler_name=Samplers.DPMPP_M_KARRAS,
width=512,
height=512,
steps=30,
controlnet_units=[
Img2ImgV3ControlNetUnit(
image_base64="https://img.freepik.com/premium-photo/close-up-dogs-face-with-big-smile-generative-ai_900101-62851.jpg",
model_name="control_v11f1p_sd15_depth",
strength=1.0
)
],
embeddings=[Img2ImgV3Embedding(model_name=_) for _ in [
"BadDream_53202",
]],
seed=-1,
)
base64_to_image(res.images_encoded[0]).save("./img2img-controlnet.png")
import pdb
import os
from novita_client import NovitaClient, Img2ImgV3ControlNetUnit, ControlNetPreprocessor, Img2ImgV3Embedding
from novita_client.utils import base64_to_image, input_image_to_pil
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.img2img_v3(
model_name="MeinaHentai_V5.safetensors",
steps=30,
height=512,
width=512,
input_image="https://img.freepik.com/premium-photo/close-up-dogs-face-with-big-smile-generative-ai_900101-62851.jpg",
prompt="1 cute dog",
strength=0.5,
guidance_scale=7,
embeddings=[Img2ImgV3Embedding(model_name=_) for _ in [
"bad-image-v2-39000",
"verybadimagenegative_v1.3_21434",
"BadDream_53202",
"badhandv4_16755",
"easynegative_8955.safetensors"]],
seed=-1,
sampler_name="DPM++ 2M Karras",
clip_skip=2,
# controlnet_units=[Img2ImgV3ControlNetUnit(
# model_name="control_v11f1p_sd15_depth",
# preprocessor="depth",
# image_base64="./20240309-003206.jpeg",
# strength=1.0
# )]
)
base64_to_image(res.images_encoded[0]).save("./img2img.png")
import os
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URNOVITA_API_URII', None))
res = client.img2video(
model_name="SVD-XT",
steps=30,
frames_num=25,
image="https://replicate.delivery/pbxt/JvLi9smWKKDfQpylBYosqQRfPKZPntuAziesp0VuPjidq61n/rocket.png",
enable_frame_interpolation=True
)
with open("test.mp4", "wb") as f:
f.write(res.video_bytes[0])
import os
import base64
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.inpainting(
model_name = "realisticVisionV40_v40VAE-inpainting_81543.safetensors",
image="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png",
mask="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png",
seed=1,
guidance_scale=15,
steps = 20,
image_num = 4,
prompt = "black rabbit",
negative_prompt = "white rabbit",
sampler_name = "Euler a",
inpainting_full_res = 1,
inpainting_full_res_padding = 32,
inpainting_mask_invert = 0,
initial_noise_multiplier = 1,
mask_blur = 1,
clip_skip = 1,
strength = 0.85,
)
with open("result/result_image/inpaintingsdk.jpeg", "wb") as image_file:
image_file.write(base64.b64decode(res.images_encoded[0]))```
### instantid
```python
import os
from novita_client import NovitaClient, InstantIDControlnetUnit
import base64
if __name__ == '__main__':
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.instant_id(
model_name="sdxlUnstableDiffusers_v8HEAVENSWRATH_133813.safetensors",
face_images=[
"https://raw.githubusercontent.com/InstantID/InstantID/main/examples/yann-lecun_resize.jpg",
],
prompt="Flat illustration, a Chinese a man, ancient style, wearing a red cloth, smile face, white skin, clean background, fireworks blooming, red lanterns",
negative_prompt="(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
id_strength=0.8,
adapter_strength=0.8,
steps=20,
seed=42,
width=1024,
height=1024,
controlnets=[
InstantIDControlnetUnit(
model_name='controlnet-openpose-sdxl-1.0',
strength=0.4,
preprocessor='openpose',
),
InstantIDControlnetUnit(
model_name='controlnet-canny-sdxl-1.0',
strength=0.3,
preprocessor='canny',
),
],
response_image_type='jpeg',
enterprise_plan=False,
)
print('res:', res)
if hasattr(res, 'images_encoded'):
with open(f"instantid.png", "wb") as f:
f.write(base64.b64decode(res.images_encoded[0]))
import os
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.merge_face(
image="https://toppng.com/uploads/preview/cut-out-people-png-personas-en-formato-11563277290kozkuzsos5.png",
face_image="https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQDy7sXtuvCNUQoQZvTbLRbX6qK9_kP3PlQfg&s",
enterprise_plan=False,
)
base64_to_image(res.image_file).save("./merge_face.png")
#!/usr/bin/env python
# -*- coding: UTF-8 -*-
from novita_client import NovitaClient, ModelType
# get your api key refer to https://docs.novita.ai/get-started/
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
# filter by model type
print("lora count", len(client.models().filter_by_type(ModelType.LORA)))
print("checkpoint count", len(client.models().filter_by_type(ModelType.CHECKPOINT)))
print("textinversion count", len(
client.models().filter_by_type(ModelType.TEXT_INVERSION)))
print("vae count", len(client.models().filter_by_type(ModelType.VAE)))
print("controlnet count", len(client.models().filter_by_type(ModelType.CONTROLNET)))
# filter by civitai tags
client.models().filter_by_civi_tags('anime')
# filter by nsfw
client.models().filter_by_nsfw(False) # or True
# sort by civitai download
client.models().sort_by_civitai_download()
# chain filters
client.models().\
filter_by_type(ModelType.CHECKPOINT).\
filter_by_nsfw(False).\
filter_by_civitai_tags('anime')
import os
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.reimagine(
image="/home/anyisalin/develop/novita-client-python/examples/doodle-generated.png"
)
base64_to_image(res.image_file).save("./reimagine.png")
import os
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.remove_background(
image="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png",
)
base64_to_image(res.image_file).save("./remove_background.png")
import os
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.remove_text(
image="https://images.uiiiuiii.com/wp-content/uploads/2023/07/i-banner-20230714-1.jpg"
)
base64_to_image(res.image_file).save("./remove_text.png")
import os
from novita_client import NovitaClient
from novita_client.utils import base64_to_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.replace_background(
image="./telegram-cloud-photo-size-2-5408823814353177899-y.jpg",
prompt="in living room, Christmas tree",
)
base64_to_image(res.image_file).save("./replace_background.png")
import os
from novita_client import NovitaClient, Samplers, Txt2ImgV3HiresFix
from novita_client.utils import base64_to_image
from PIL import Image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.txt2img_v3(
model_name='dreamshaper_8_93211.safetensors',
prompt="a cute girl",
width=384,
height=512,
image_num=1,
guidance_scale=7.5,
seed=12345,
sampler_name=Samplers.EULER_A,
hires_fix=Txt2ImgV3HiresFix(
# upscaler='Latent'
target_width=768,
target_height=1024,
strength=0.5
)
)
base64_to_image(res.images_encoded[0]).save("./txt2img_with_hiresfix.png")
#!/usr/bin/env python
# -*- coding: UTF-8 -*-
import os
from novita_client import NovitaClient, Txt2ImgV3LoRA, Samplers, ProgressResponseStatusCode, ModelType, add_lora_to_prompt, save_image
from novita_client.utils import base64_to_image, input_image_to_pil
from PIL import Image
def make_image_grid(images, rows: int, cols: int, resize: int = None):
"""
Prepares a single grid of images. Useful for visualization purposes.
"""
assert len(images) == rows * cols
if resize is not None:
images = [img.resize((resize, resize)) for img in images]
w, h = images[0].size
grid = Image.new("RGB", size=(cols * w, rows * h))
for i, img in enumerate(images):
grid.paste(img, box=(i % cols * w, i // cols * h))
return grid
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res1 = client.txt2img_v3(
prompt="a photo of handsome man, close up",
image_num=1,
guidance_scale=7.0,
sampler_name=Samplers.DPMPP_M_KARRAS,
model_name="dreamshaper_8_93211.safetensors",
height=512,
width=512,
seed=1024,
)
res2 = client.txt2img_v3(
prompt="a photo of handsome man, close up",
image_num=1,
guidance_scale=7.0,
sampler_name=Samplers.DPMPP_M_KARRAS,
model_name="dreamshaper_8_93211.safetensors",
height=512,
width=512,
seed=1024,
loras=[
Txt2ImgV3LoRA(
model_name="add_detail_44319",
strength=0.9,
)
]
)
make_image_grid([base64_to_image(res1.images_encoded[0]), base64_to_image(res2.images_encoded[0])], 1, 2, 512).save("./txt2img-lora-compare.png")
import os
from novita_client import NovitaClient, Txt2ImgV3Refiner, Samplers
from novita_client.utils import base64_to_image
from PIL import Image
def make_image_grid(images, rows: int, cols: int, resize: int = None):
"""
Prepares a single grid of images. Useful for visualization purposes.
"""
assert len(images) == rows * cols
if resize is not None:
images = [img.resize((resize, resize)) for img in images]
w, h = images[0].size
grid = Image.new("RGB", size=(cols * w, rows * h))
for i, img in enumerate(images):
grid.paste(img, box=(i % cols * w, i // cols * h))
return grid
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
r1 = client.txt2img_v3(
model_name='sd_xl_base_1.0.safetensors',
prompt='a astronaut riding a bike on the moon',
width=1024,
height=1024,
image_num=1,
guidance_scale=7.5,
sampler_name=Samplers.EULER_A,
)
r2 = client.txt2img_v3(
model_name='sd_xl_base_1.0.safetensors',
prompt='a astronaut riding a bike on the moon',
width=1024,
height=1024,
image_num=1,
guidance_scale=7.5,
sampler_name=Samplers.EULER_A,
refiner=Txt2ImgV3Refiner(
switch_at=0.7
)
)
r3 = client.txt2img_v3(
model_name='sd_xl_base_1.0.safetensors',
prompt='a astronaut riding a bike on the moon',
width=1024,
height=1024,
image_num=1,
guidance_scale=7.5,
sampler_name=Samplers.EULER_A,
refiner=Txt2ImgV3Refiner(
switch_at=0.5
)
)
make_image_grid([base64_to_image(r1.images_encoded[0]), base64_to_image(r2.images_encoded[0]), base64_to_image(r3.images_encoded[0])], 1, 3, 1024).save("./txt2img-refiner-compare.png")
import os
from novita_client import NovitaClient
from novita_client.utils import save_image
client = NovitaClient(os.getenv('NOVITA_API_KEY'), os.getenv('NOVITA_API_URI', None))
res = client.txt2video(
model_name = "dreamshaper_8_93211.safetensors",
prompts = [{
"prompt": "A girl, baby, portrait, 5 years old",
"frames": 16,},
{
"prompt": "A girl, child, portrait, 10 years old",
"frames": 16,
}
],
steps = 20,
guidance_scale = 10,
height = 512,
width = 768,
clip_skip = 4,
negative_prompt = "a rainy day",
response_video_type = "mp4",
)
save_image(res.video_bytes[0], 'output.mp4')