AI/dify
1
0
Fork 0
mirror of https://git.mirrors.martin98.com/https://github.com/langgenius/dify.git synced 2025-08-12 02:39:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: add Novita AI image generation tool, implemented model search, text-to-image and create tile functionalities (#5308)

Browse Source 
Co-authored-by: crazywoola <100913391+crazywoola@users.noreply.github.com>
This commit is contained in:
Xiao Ley 2024-06-18 11:08:25 +08:00 committed by GitHub
parent 3828d4cd22
commit 132f5fb3de
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
10 changed files with 983 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

BIN
api/core/tools/provider/builtin/novitaai/_assets/icon.ico Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 9.1 KiB

30
api/core/tools/provider/builtin/novitaai/novitaai.py Normal file
View File

@ -0,0 +1,30 @@
from typing import Any
from core.tools.errors import ToolProviderCredentialValidationError
from core.tools.provider.builtin.novitaai.tools.novitaai_txt2img import NovitaAiTxt2ImgTool
from core.tools.provider.builtin_tool_provider import BuiltinToolProviderController
class NovitaAIProvider(BuiltinToolProviderController):
def _validate_credentials(self, credentials: dict[str, Any]) -> None:
try:
result = NovitaAiTxt2ImgTool().fork_tool_runtime(
runtime={
"credentials": credentials,
}
).invoke(
user_id='',
tool_parameters={
'model_name': 'cinenautXLATRUE_cinenautV10_392434.safetensors',
'prompt': 'a futuristic city with flying cars',
'negative_prompt': '',
'width': 128,
'height': 128,
'image_num': 1,
'guidance_scale': 7.5,
'seed': -1,
'steps': 1,
},
)
except Exception as e:
raise ToolProviderCredentialValidationError(str(e))

32
api/core/tools/provider/builtin/novitaai/novitaai.yaml Normal file
View File

@ -0,0 +1,32 @@
identity:
author: Xiao Ley
name: novitaai
label:
en_US: Novita AI
zh_Hans: Novita AI
pt_BR: Novita AI
description:
en_US: Innovative AI for Image Generation
zh_Hans: 用于图像生成的创新人工智能。
pt_BR: Innovative AI for Image Generation
icon: icon.ico
tags:
- image
- productivity
credentials_for_provider:
api_key:
type: secret-input
required: true
label:
en_US: API Key
zh_Hans: API 密钥
pt_BR: Chave API
placeholder:
en_US: Please enter your Novita AI API key
zh_Hans: 请输入你的 Novita AI API 密钥
pt_BR: Por favor, insira sua chave de API do Novita AI
help:
en_US: Get your Novita AI API key from Novita AI
zh_Hans: 从 Novita AI 获取您的 Novita AI API 密钥
pt_BR: Obtenha sua chave de API do Novita AI na Novita AI
url: https://novita.ai

51
api/core/tools/provider/builtin/novitaai/tools/novitaai_createtile.py Normal file
View File

@ -0,0 +1,51 @@
from base64 import b64decode
from copy import deepcopy
from typing import Any, Union
from novita_client import (
NovitaClient,
)
from core.tools.entities.tool_entities import ToolInvokeMessage
from core.tools.errors import ToolProviderCredentialValidationError
from core.tools.tool.builtin_tool import BuiltinTool
class NovitaAiCreateTileTool(BuiltinTool):
def _invoke(self,
user_id: str,
tool_parameters: dict[str, Any],
) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]:
"""
invoke tools
"""
if 'api_key' not in self.runtime.credentials or not self.runtime.credentials.get('api_key'):
raise ToolProviderCredentialValidationError("Novita AI API Key is required.")
api_key = self.runtime.credentials.get('api_key')
client = NovitaClient(api_key=api_key)
param = self._process_parameters(tool_parameters)
client_result = client.create_tile(**param)
results = []
results.append(
self.create_blob_message(blob=b64decode(client_result.image_file),
meta={'mime_type': f'image/{client_result.image_type}'},
save_as=self.VARIABLE_KEY.IMAGE.value)
)
return results
def _process_parameters(self, parameters: dict[str, Any]) -> dict[str, Any]:
"""
process parameters
"""
res_parameters = deepcopy(parameters)
# delete none and empty
keys_to_delete = [k for k, v in res_parameters.items() if v is None or v == '']
for k in keys_to_delete:
del res_parameters[k]
return res_parameters

80
api/core/tools/provider/builtin/novitaai/tools/novitaai_createtile.yaml Normal file
View File

@ -0,0 +1,80 @@
identity:
name: novitaai_createtile
author: Xiao Ley
label:
en_US: Novita AI Create Tile
zh_Hans: Novita AI 创建平铺图案
description:
human:
en_US: This feature produces images designed for seamless tiling, ideal for creating continuous patterns in fabrics, wallpapers, and various textures.
zh_Hans: 该功能生成设计用于无缝平铺的图像，非常适合用于制作连续图案的织物、壁纸和各种纹理。
llm: A tool for create images designed for seamless tiling, ideal for creating continuous patterns in fabrics, wallpapers, and various textures.
parameters:
- name: prompt
type: string
required: true
label:
en_US: prompt
zh_Hans: 提示
human_description:
en_US: Positive prompt word of the created tile, divided by `,`, Range [1, 512]. Only English input is allowed.
zh_Hans: 生成平铺图案的正向提示，用 `,` 分隔，范围 [1, 512]。仅允许输入英文。
llm_description: Image prompt of Novita AI, you should describe the image you want to generate as a list of words as possible as detailed, divided by `,`, Range [1, 512]. Only English input is allowed.
form: llm
- name: negative_prompt
type: string
required: false
label:
en_US: negative prompt
zh_Hans: 负向提示
human_description:
en_US: Negtive prompt word of the created tile, divided by `,`, Range [1, 512]. Only English input is allowed.
zh_Hans: 生成平铺图案的负向提示，用 `,` 分隔，范围 [1, 512]。仅允许输入英文。
llm_description: Image negative prompt of Novita AI, divided by `,`, Range [1, 512]. Only English input is allowed.
form: llm
- name: width
type: number
default: 256
min: 128
max: 1024
required: true
label:
en_US: width
zh_Hans: 宽
human_description:
en_US: Image width, Range [128, 1024].
zh_Hans: 图像宽度，范围 [128, 1024]
form: form
- name: height
type: number
default: 256
min: 128
max: 1024
required: true
label:
en_US: height
zh_Hans: 高
human_description:
en_US: Image height, Range [128, 1024].
zh_Hans: 图像高度，范围 [128, 1024]
form: form
- name: response_image_type
type: select
default: jpeg
required: false
label:
en_US: response image type
zh_Hans: 响应图像类型
human_description:
en_US: Response image type, png or jpeg
zh_Hans: 响应图像类型，png 或 jpeg
form: form
options:
- value: jpeg
label:
en_US: jpeg
zh_Hans: jpeg
- value: png
label:
en_US: png
zh_Hans: png

137
api/core/tools/provider/builtin/novitaai/tools/novitaai_modelquery.py Normal file
View File

@ -0,0 +1,137 @@
import json
from copy import deepcopy
from typing import Any, Union
from pandas import DataFrame
from yarl import URL
from core.helper import ssrf_proxy
from core.tools.entities.tool_entities import ToolInvokeMessage
from core.tools.errors import ToolProviderCredentialValidationError
from core.tools.tool.builtin_tool import BuiltinTool
class NovitaAiModelQueryTool(BuiltinTool):
_model_query_endpoint = 'https://api.novita.ai/v3/model'
def _invoke(self,
user_id: str,
tool_parameters: dict[str, Any],
) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]:
"""
invoke tools
"""
if 'api_key' not in self.runtime.credentials or not self.runtime.credentials.get('api_key'):
raise ToolProviderCredentialValidationError("Novita AI API Key is required.")
api_key = self.runtime.credentials.get('api_key')
headers = {
'Content-Type': 'application/json',
'Authorization': "Bearer " + api_key
}
params = self._process_parameters(tool_parameters)
result_type = params.get('result_type')
del params['result_type']
models_data = self._query_models(
models_data=[],
headers=headers,
params=params,
recursive=False if result_type == 'first sd_name' or result_type == 'first name sd_name pair' else True
)
result_str = ''
if result_type == 'first sd_name':
result_str = models_data[0]['sd_name_in_api']
elif result_type == 'first name sd_name pair':
result_str = json.dumps({'name': models_data[0]['name'], 'sd_name': models_data[0]['sd_name_in_api']})
elif result_type == 'sd_name array':
sd_name_array = [model['sd_name_in_api'] for model in models_data]
result_str = json.dumps(sd_name_array)
elif result_type == 'name array':
name_array = [model['name'] for model in models_data]
result_str = json.dumps(name_array)
elif result_type == 'name sd_name pair array':
name_sd_name_pair_array = [{'name': model['name'], 'sd_name': model['sd_name_in_api']} for model in models_data]
result_str = json.dumps(name_sd_name_pair_array)
elif result_type == 'whole info array':
result_str = json.dumps(models_data)
else:
raise NotImplementedError
return self.create_text_message(result_str)
def _query_models(self, models_data: list, headers: dict[str, Any],
params: dict[str, Any], pagination_cursor: str = '', recursive: bool = True) -> list:
"""
query models
"""
inside_params = deepcopy(params)
if pagination_cursor != '':
inside_params['pagination.cursor'] = pagination_cursor
response = ssrf_proxy.get(
url=str(URL(self._model_query_endpoint)),
headers=headers,
params=params,
timeout=(10, 60)
)
res_data = response.json()
models_data.extend(res_data['models'])
res_data_len = len(res_data['models'])
if res_data_len == 0 or res_data_len < int(params['pagination.limit']) or recursive is False:
# deduplicate
df = DataFrame.from_dict(models_data)
df_unique = df.drop_duplicates(subset=['id'])
models_data = df_unique.to_dict('records')
return models_data
return self._query_models(
models_data=models_data,
headers=headers,
params=inside_params,
pagination_cursor=res_data['pagination']['next_cursor']
)
def _process_parameters(self, parameters: dict[str, Any]) -> dict[str, Any]:
"""
process parameters
"""
process_parameters = deepcopy(parameters)
res_parameters = {}
# delete none or empty
keys_to_delete = [k for k, v in process_parameters.items() if v is None or v == '']
for k in keys_to_delete:
del process_parameters[k]
if 'query' in process_parameters and process_parameters.get('query') != 'unspecified':
res_parameters['filter.query'] = process_parameters['query']
if 'visibility' in process_parameters and process_parameters.get('visibility') != 'unspecified':
res_parameters['filter.visibility'] = process_parameters['visibility']
if 'source' in process_parameters and process_parameters.get('source') != 'unspecified':
res_parameters['filter.source'] = process_parameters['source']
if 'type' in process_parameters and process_parameters.get('type') != 'unspecified':
res_parameters['filter.types'] = process_parameters['type']
if 'is_sdxl' in process_parameters:
if process_parameters['is_sdxl'] == 'true':
res_parameters['filter.is_sdxl'] = True
elif process_parameters['is_sdxl'] == 'false':
res_parameters['filter.is_sdxl'] = False
res_parameters['result_type'] = process_parameters.get('result_type', 'first sd_name')
res_parameters['pagination.limit'] = 1 \
if res_parameters.get('result_type') == 'first sd_name' \
or res_parameters.get('result_type') == 'first name sd_name pair'\
else 100
return res_parameters

174
api/core/tools/provider/builtin/novitaai/tools/novitaai_modelquery.yaml Normal file
View File

@ -0,0 +1,174 @@
identity:
name: novitaai_modelquery
author: Xiao Ley
label:
en_US: Novita AI Model Query
zh_Hans: Novita AI 模型查询
description:
human:
en_US: Retrieve information on both public and private models. It allows users to access details such as model specifications, status, and usage guidelines, ensuring comprehensive insight into the available modeling resources.
zh_Hans: 检索公开和私有模型信息。它允许用户访问模型规范、状态和使用指南等详细信息，确保了解可用的建模资源。
llm: A tool for retrieve information on both public and private Novita AI models.
parameters:
- name: query
type: string
required: false
label:
en_US: query
zh_Hans: 查询
human_description:
en_US: Seaching the content of sd_name, name, tags.
zh_Hans: 搜索 sd_name、name、tags 中的内容
form: form
- name: result_type
type: select
default: "first sd_name"
required: true
label:
en_US: result format
zh_Hans: 结果格式
human_description:
en_US: The format of result
zh_Hans: 请求结果的格式
form: form
options:
- value: "first sd_name"
label:
en_US: "first sd_name"
zh_Hans: "第一个 sd_name"
- value: "first name sd_name pair"
label:
en_US: "first name and sd_name pair: {name, sd_name}"
zh_Hans: "第一个 name sd_name 组合：{name, sd_name}"
- value: "sd_name array"
label:
en_US: "sd_name array: [sd_name]"
zh_Hans: "sd_name 数组：[sd_name]"
- value: "name array"
label:
en_US: "name array: [name]"
zh_Hans: "name 数组：[name]"
- value: "name sd_name pair array"
label:
en_US: "name and sd_name pair array: [{name, sd_name}]"
zh_Hans: "name sd_name 组合数组：[{name, sd_name}]"
- value: "whole info array"
label:
en_US: whole info array
zh_Hans: 完整信息数组
- name: visibility
type: select
default: unspecified
required: false
label:
en_US: visibility
zh_Hans: 可见性
human_description:
en_US: Whether the model is public or private
zh_Hans: 模型是否公开或私有
form: form
options:
- value: unspecified
label:
en_US: Unspecified
zh_Hans: 未指定
- value: public
label:
en_US: Public
zh_Hans: 公开
- value: private
label:
en_US: Private
zh_Hans: 私有
- name: source
type: select
default: unspecified
required: false
label:
en_US: source
zh_Hans: 来源
human_description:
en_US: Source of the model
zh_Hans: 模型来源
form: form
options:
- value: unspecified
label:
en_US: Unspecified
zh_Hans: 未指定
- value: civitai
label:
en_US: Civitai
zh_Hans: Civitai
- value: training
label:
en_US: Training
zh_Hans: 训练
- value: uploading
label:
en_US: Uploading
zh_Hans: 上传
- name: type
type: select
default: unspecified
required: false
label:
en_US: type
zh_Hans: 类型
human_description:
en_US: Specifies the type of models to include in the query.
zh_Hans: 指定要查询的模型类型
form: form
options:
- value: unspecified
label:
en_US: Unspecified
zh_Hans: 未指定
- value: checkpoint
label:
en_US: Checkpoint
zh_Hans: Checkpoint
- value: lora
label:
en_US: LoRA
zh_Hans: LoRA
- value: vae
label:
en_US: VAE
zh_Hans: VAE
- value: controlnet
label:
en_US: ControlNet
zh_Hans: ControlNet
- value: upscaler
label:
en_US: Upscaler
zh_Hans: Upscaler
- value: textualinversion
label:
en_US: Textual inversion
zh_Hans: Textual Inversion
- name: is_sdxl
type: select
default: unspecified
required: false
label:
en_US: is sdxl
zh_Hans: 是否是 SDXL
human_description:
en_US: Whether sdxl model or not. Setting this parameter to `true` includes only sdxl models in the query results, which are typically large-scale, high-performance models designed for extensive data processing tasks. Conversely, setting it to `false` excludes these models from the results. If left unspecified, the filter will not discriminate based on the sdxl classification, including all model types in the search results.
zh_Hans: 是否是 SDXL 模型。设置此参数为 `是`，只查询 SDXL 模型，并包含大规模，高性能的模型。相反，设置为 `否`，将排除这些模型。如果未指定，将不会根据 SDXL 分类进行区分，包括查询结果中的所有模型类型。
form: form
options:
- value: unspecified
label:
en_US: Unspecified
zh_Hans: 未指定
- value: "true"
label:
en_US: "True"
zh_Hans: 是
- value: "false"
label:
en_US: "False"
zh_Hans: 否

137
api/core/tools/provider/builtin/novitaai/tools/novitaai_txt2img.py Normal file
View File

@ -0,0 +1,137 @@
from base64 import b64decode
from copy import deepcopy
from typing import Any, Union
from novita_client import (
NovitaClient,
Txt2ImgV3Embedding,
Txt2ImgV3HiresFix,
Txt2ImgV3LoRA,
Txt2ImgV3Refiner,
V3TaskImage,
)
from core.tools.entities.tool_entities import ToolInvokeMessage
from core.tools.errors import ToolProviderCredentialValidationError
from core.tools.tool.builtin_tool import BuiltinTool
class NovitaAiTxt2ImgTool(BuiltinTool):
def _invoke(self,
user_id: str,
tool_parameters: dict[str, Any],
) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]:
"""
invoke tools
"""
if 'api_key' not in self.runtime.credentials or not self.runtime.credentials.get('api_key'):
raise ToolProviderCredentialValidationError("Novita AI API Key is required.")
api_key = self.runtime.credentials.get('api_key')
client = NovitaClient(api_key=api_key)
param = self._process_parameters(tool_parameters)
client_result = client.txt2img_v3(**param)
results = []
for image_encoded, image in zip(client_result.images_encoded, client_result.images):
if self._is_hit_nsfw_detection(image, 0.8):
results = self.create_text_message(text='NSFW detected!')
break
results.append(
self.create_blob_message(blob=b64decode(image_encoded),
meta={'mime_type': f'image/{image.image_type}'},
save_as=self.VARIABLE_KEY.IMAGE.value)
)
return results
def _process_parameters(self, parameters: dict[str, Any]) -> dict[str, Any]:
"""
process parameters
"""
res_parameters = deepcopy(parameters)
# delete none and empty
keys_to_delete = [k for k, v in res_parameters.items() if v is None or v == '']
for k in keys_to_delete:
del res_parameters[k]
if 'clip_skip' in res_parameters and res_parameters.get('clip_skip') == 0:
del res_parameters['clip_skip']
if 'refiner_switch_at' in res_parameters and res_parameters.get('refiner_switch_at') == 0:
del res_parameters['refiner_switch_at']
if 'enabled_enterprise_plan' in res_parameters:
res_parameters['enterprise_plan'] = {'enabled': res_parameters['enabled_enterprise_plan']}
del res_parameters['enabled_enterprise_plan']
if 'nsfw_detection_level' in res_parameters:
res_parameters['nsfw_detection_level'] = int(res_parameters['nsfw_detection_level'])
# process loras
if 'loras' in res_parameters:
loras_ori_list = res_parameters.get('loras').strip().split(';')
locals_list = []
for lora_str in loras_ori_list:
lora_info = lora_str.strip().split(',')
lora = Txt2ImgV3LoRA(
model_name=lora_info[0].strip(),
strength=float(lora_info[1]),
)
locals_list.append(lora)
res_parameters['loras'] = locals_list
# process embeddings
if 'embeddings' in res_parameters:
embeddings_ori_list = res_parameters.get('embeddings').strip().split(';')
locals_list = []
for embedding_str in embeddings_ori_list:
embedding = Txt2ImgV3Embedding(
model_name=embedding_str.strip()
)
locals_list.append(embedding)
res_parameters['embeddings'] = locals_list
# process hires_fix
if 'hires_fix' in res_parameters:
hires_fix_ori = res_parameters.get('hires_fix')
hires_fix_info = hires_fix_ori.strip().split(',')
if 'upscaler' in hires_fix_info:
hires_fix = Txt2ImgV3HiresFix(
target_width=int(hires_fix_info[0]),
target_height=int(hires_fix_info[1]),
strength=float(hires_fix_info[2]),
upscaler=hires_fix_info[3].strip()
)
else:
hires_fix = Txt2ImgV3HiresFix(
target_width=int(hires_fix_info[0]),
target_height=int(hires_fix_info[1]),
strength=float(hires_fix_info[2])
)
res_parameters['hires_fix'] = hires_fix
if 'refiner_switch_at' in res_parameters:
refiner = Txt2ImgV3Refiner(
switch_at=float(res_parameters.get('refiner_switch_at'))
)
del res_parameters['refiner_switch_at']
res_parameters['refiner'] = refiner
return res_parameters
def _is_hit_nsfw_detection(self, image: V3TaskImage, confidence_threshold: float) -> bool:
"""
is hit nsfw
"""
if image.nsfw_detection_result is None:
return False
if image.nsfw_detection_result.valid and image.nsfw_detection_result.confidence >= confidence_threshold:
return True
return False

341
api/core/tools/provider/builtin/novitaai/tools/novitaai_txt2img.yaml Normal file
View File

@ -0,0 +1,341 @@
identity:
name: novitaai_txt2img
author: Xiao Ley
label:
en_US: Novita AI Text to Image
zh_Hans: Novita AI 文字转图像
description:
human:
en_US: Generate images from text prompts using Stable Diffusion models
zh_Hans: 通过 Stable Diffusion 模型根据文字提示生成图像
llm: A tool for generate images from English text prompts.
parameters:
- name: model_name
type: string
required: true
label:
en_US: model name
zh_Hans: 模块名字
human_description:
en_US: Specify the name of the model checkpoint. You can use the "Novita AI Model Query" tool to query the corresponding "sd_name" value (type select "Checkpoint").
zh_Hans: 指定 Model Checkpoint 名称。可通过“Novita AI 模型请求”工具查询对应的“sd_name”值（类型选择“Checkpoint”）。
form: form
- name: prompt
type: string
required: true
label:
en_US: prompt
zh_Hans: 提示
human_description:
en_US: Text input required to guide the image generation, divided by `,`, Range [1, 1024]. Only English input is allowed.
zh_Hans: 生成图像的正向提示，用 `,` 分隔，范围 [1, 1024]。仅允许输入英文。
llm_description: Image prompt of Novita AI, you should describe the image you want to generate as a list of words as possible as detailed, divided by `,`, Range [1, 1024]. Only English input is allowed.
form: llm
- name: negative_prompt
type: string
required: false
label:
en_US: negative prompt
zh_Hans: 负向提示
human_description:
en_US: Text input that will not guide the image generation, divided by `,`, Range [1, 1024]. Only English input is allowed.
zh_Hans: 生成图像的负向提示，用 `,` 分隔，范围 [1, 1024]。仅允许输入英文。
llm_description: Image negative prompt of Novita AI, divided by `,`, Range [1, 1024]. Only English input is allowed.
form: llm
- name: width
type: number
default: 512
min: 128
max: 2048
required: true
label:
en_US: width
zh_Hans: 宽
human_description:
en_US: Image width, Range [128, 2048].
zh_Hans: 图像宽度，范围 [128, 2048]
form: form
- name: height
type: number
default: 512
min: 128
max: 2048
required: true
label:
en_US: height
zh_Hans: 高
human_description:
en_US: Image height, Range [128, 2048].
zh_Hans: 图像高度，范围 [128, 2048]
form: form
- name: image_num
type: number
default: 1
min: 1
max: 8
required: true
label:
en_US: image num
zh_Hans: 图片数
human_description:
en_US: Image num, Range [1, 8].
zh_Hans: 图片数，范围 [1, 8]
form: form
- name: steps
type: number
default: 20
min: 1
max: 100
required: true
label:
en_US: steps
zh_Hans: 步数
human_description:
en_US: The number of denoising steps. More steps usually can produce higher quality images, but take more time to generate, Range [1, 100].
zh_Hans: 生成步数。更多步数可能会产生更好的图像，但生成时间更长，范围 [1, 100]
form: form
- name: seed
type: number
default: -1
required: true
label:
en_US: seed
zh_Hans: 种子
human_description:
en_US: A seed is a number from which Stable Diffusion generates noise, which, makes generation deterministic. Using the same seed and set of parameters will produce identical image each time, minimum -1.
zh_Hans: 种子是 Stable Diffusion 生成噪声的数字，它使生成具有确定性。使用相同的种子和参数设置将生成每次生成相同的图像，最小值 -1。
form: form
- name: clip_skip
type: number
min: 1
max: 12
required: false
label:
en_US: clip skip
zh_Hans: 层跳过数
human_description:
en_US: This parameter indicates the number of layers to stop from the bottom during optimization, so clip_skip on 2 would mean, that in SD1.x model where the CLIP has 12 layers, you would stop at 10th layer, Range [1, 12], get reference at https://novita.ai/get-started/Misc.html#what-s-clip-skip.
zh_Hans: 此参数表示优化过程中从底部停止的层数，因此 clip_skip 的值为 2，表示在 SD1.x 模型中，CLIP 有 12 层，你将停止在 10 层，范围 [1, 12]，参考 https://novita.ai/get-started/Misc.html#what-s-clip-skip。
form: form
- name: guidance_scale
type: number
default: "7.5"
min: 1.0
max: 30.0
required: true
label:
en_US: guidance scale
zh_Hans: 提示词遵守程度
human_description:
en_US: This setting says how close the Stable Diffusion will listen to your prompt, higer guidance forces the model to better follow the prompt, but result in lower quality output.Range [1, 30].
zh_Hans: 此设置表明 Stable Diffusion 如何听从您的提示，较高的 guidance_scale 会强制模型更好跟随提示，但结果会更低质量输出。范围 [1.0, 30.0]。
form: form
- name: sampler_name
type: select
required: true
label:
en_US: sampler name
zh_Hans: 采样器名称
human_description:
en_US: This parameter determines the denoising algorithm employed during the sampling phase of Stable Diffusion. Get reference at https://novita.ai/get-started/Misc.htmll#what-is-samplers.
zh_Hans: 此参数决定了在稳定扩散采样阶段使用的去噪算法。参考 https://novita.ai/get-started/Misc.htmll#what-is-samplers。
form: form
options:
- value: "Euler a"
label:
en_US: Euler a
zh_Hans: Euler a
- value: "Euler"
label:
en_US: Euler
zh_Hans: Euler
- value: "LMS"
label:
en_US: LMS
zh_Hans: LMS
- value: "Heun"
label:
en_US: Heun
zh_Hans: Heun
- value: "DPM2"
label:
en_US: DPM2
zh_Hans: DPM2
- value: "DPM2 a"
label:
en_US: DPM2 a
zh_Hans: DPM2 a
- value: "DPM++ 2S a"
label:
en_US: DPM++ 2S a
zh_Hans: DPM++ 2S a
- value: "DPM++ 2M"
label:
en_US: DPM++ 2M
zh_Hans: DPM++ 2M
- value: "DPM++ SDE"
label:
en_US: DPM++ SDE
zh_Hans: DPM++ SDE
- value: "DPM fast"
label:
en_US: DPM fast
zh_Hans: DPM fast
- value: "DPM adaptive"
label:
en_US: DPM adaptive
zh_Hans: DPM adaptive
- value: "LMS Karras"
label:
en_US: LMS Karras
zh_Hans: LMS Karras
- value: "DPM2 Karras"
label:
en_US: DPM2 Karras
zh_Hans: DPM2 Karras
- value: "DPM2 a Karras"
label:
en_US: DPM2 a Karras
zh_Hans: DPM2 a Karras
- value: "DPM++ 2S a Karras"
label:
en_US: DPM++ 2S a Karras
zh_Hans: DPM++ 2S a Karras
- value: "DPM++ 2M Karras"
label:
en_US: DPM++ 2M Karras
zh_Hans: DPM++ 2M Karras
- value: "DPM++ SDE Karras"
label:
en_US: DPM++ SDE Karras
zh_Hans: DPM++ SDE Karras
- value: "DDIM"
label:
en_US: DDIM
zh_Hans: DDIM
- value: "PLMS"
label:
en_US: PLMS
zh_Hans: PLMS
- value: "UniPC"
label:
en_US: UniPC
zh_Hans: UniPC
- name: sd_vae
type: string
required: false
label:
en_US: sd vae
zh_Hans: sd vae
human_description:
en_US: VAE(Variational Autoencoder), get reference at https://novita.ai/get-started/Misc.html#what-s-variational-autoencoders-vae. You can use the "Novita AI Model Query" tool to query the corresponding "sd_name" value (type select "VAE").
zh_Hans: VAE（变分自编码器），参考 https://novita.ai/get-started/Misc.html#what-s-variational-autoencoders-vae。可通过“Novita AI 模型请求”工具查询对应的“sd_name”值（类型选择“VAE”）。
form: form
- name: loras
type: string
required: false
label:
en_US: loRAs
zh_Hans: loRAs
human_description:
en_US: LoRA models. Currenlty supports up to 5 LoRAs. You can use the "Novita AI Model Query" tool to query the corresponding "sd_name" value (type select "LoRA"). Input template is "<sd_name>,<strength [0-1.0]>;<sd_name>,<strength [0-1.0]>;...". Such as"Film Grain style_331903,0.5;DoggystylePOV_9600,0.5"
zh_Hans: LoRA 模型。目前仅支持 5 个 LoRA。可通过“Novita AI 模型请求”工具查询对应的“sd_name”值（类型选择“LoRA”）。输入模板：“<sd_name>,<strength [0-1.0]>;<sd_name>,<strength [0-1.0]>;...”，例如：“Film Grain style_331903,0.5;DoggystylePOV_9600,0.5”
form: form
- name: embeddings
type: string
required: false
label:
en_US: text embeddings
zh_Hans: 文本嵌入
human_description:
en_US: Textual Inversion is a training method for personalizing models by learning new text embeddings from a few example images, currenlty supports up to 5 embeddings. You can use the "Novita AI Model Query" tool to query the corresponding "sd_name" value (type select "Text Inversion"). Input template is "<sd_name>;<sd_name>;...". Such as "EasyNegativeV2_75525;AS-YoungerV2"
zh_Hans: 文本反转是一种通过从一些示例图像中学习新的文本嵌入来个性化模型的训练方法，目前仅支持 5 个嵌入。可通过“Novita AI 模型请求”工具查询对应的“sd_name”值（类型选择“Text Inversion”）。输入模板：“<sd_name>;<sd_name>;...”，例如：“EasyNegativeV2_75525;AS-YoungerV2”
form: form
- name: hires_fix
type: string
required: false
label:
en_US: hires fix
zh_Hans: 高分辨率修复
human_description:
en_US: Use high resolution image fix. Input template is "<target_width [128, 4096]>,<target_height [128, 4096]>,<strength [0, 1.0]>,<upscaler (optional, selec type, `RealESRGAN_x4plus_anime_6B`, `RealESRNet_x4plus` or `Latent`)>". Such as "1024,1024,0.8", "1024,1024,0.8,RealESRGAN_x4plus_anime_6B"
zh_Hans: 使用高分辨率修复。输入模板 “<target_width [128, 4096]>,<target_height [128, 4096]>,<strength [0, 1.0]>,<upscaler (可选, 选项类型, `RealESRGAN_x4plus_anime_6B`, `RealESRNet_x4plus` 或 `Latent`)>”。例如 “1024,1024,0.8”、“1024,1024,0.8,RealESRGAN_x4plus_anime_6B”
form: form
- name: refiner_switch_at
type: number
min: 0.0
max: 1.0
required: false
label:
en_US: refiner switch at
zh_Hans: 重采样参与时刻
human_description:
en_US: This parameter in the context of a refiner allows you to set the extent to which the refiner alters the output of a model. When set to 0, the refiner has no effect; at 1, it's fully active. Intermediate values like 0.5 provide a balanced effect, where the refiner is moderately engaged, enhancing or adjusting the output without dominating the original model's characteristics. This setting is particularly useful for fine-tuning the output to achieve a desired balance between refinement and the original generative features, Range [0, 1.0]. Is not all models support refiners!
zh_Hans: 此参数允许您设置重采样更改模型输出的程度。当设置为0时，重采样不起作用；1时，它处于完全活动状态。像0.5这样的中间值提供了一种平衡效果，其中重采样适度参与，增强或调整输出，而不会主导原始模型的特性。此设置对于微调输出特别有用，范围 [0, 1.0]。不是所有模型都支持重采样！
form: form
- name: response_image_type
type: select
default: jpeg
required: false
label:
en_US: response image type
zh_Hans: 响应图像类型
human_description:
en_US: Response image type, png or jpeg
zh_Hans: 响应图像类型，png 或 jpeg
form: form
options:
- value: jpeg
label:
en_US: jpeg
zh_Hans: jpeg
- value: png
label:
en_US: png
zh_Hans: png
- name: enabled_enterprise_plan
type: boolean
default: false
required: false
label:
en_US: enterprise plan enabled
zh_Hans: 企业版计划启用
human_description:
en_US: Enable enterprise plan
zh_Hans: 启用企业版计划
form: form
- name: enable_nsfw_detection
type: boolean
default: false
required: false
label:
en_US: enable nsfw detection
zh_Hans: 启用 NSFW 检测
human_description:
en_US: Enable nsfw detection
zh_Hans: 启用 NSFW 检测
form: form
- name: nsfw_detection_level
type: select
default: "2"
required: false
label:
en_US: nsfw detection level
zh_Hans: NSFW 检测级别
human_description:
en_US: Nsfw detection level, from low to high
zh_Hans: NSFW 检测级别，越高越严格
form: form
options:
- value: "0"
label:
en_US: low
zh_Hans: 低
- value: "1"
label:
en_US: middle
zh_Hans: 中
- value: "2"
label:
en_US: high
zh_Hans: 高

1
api/requirements.txt
View File

@ -88,5 +88,6 @@ google-cloud-aiplatform==1.49.0
vanna[postgres,mysql,clickhouse,duckdb]==0.5.5
tencentcloud-sdk-python-hunyuan~=3.0.1158
chromadb~=0.5.0
novita_client~=0.5.6
tenacity~=8.3.0
cos-python-sdk-v5==1.9.30