Inference

model_id: String, required, see how to get the id of a model

prompt: String, **required

version**: String, optional Model version, default to the latest version

width: Int, optional default = 512, must be increment of 8 valid range is 128 to 896 commonly used values: 512, 640, 768

height: Int, optional default = 768, must be increment of 8 valid range is 128 to 896 commonly used values: 512, 640, 768

negative_prompt: String, optional use_default_neg: String, optional Append the default negative prompt or not pass in “true” or “false”, default = “true”

steps: Int, optional Number of inference steps default = 30 valid range is 1 to 50

scale: Float, optional Guidance scale default = 7.5 or model's default scale if one is set valid range is 1 to 20

num_images: Int, optional, default = 4 seed: Int, optional, default = -1 scheduler: String, optional, default = ”DPMSolverMultistep” or model’s default scheduler if one is set, see options

lora: String, optional id of the LoRA model. You can query /models to get the full list of LoRA lora_scale: Float, optional, default = 0.75 | | Extra input for img2img | init_image_file: a file object, the base image, required for img2img

image_strength: Float, optional How much to transform the base image, default = 0.75

controlnet: String, optional ControlNet to use, legit values are canny, depth and openpose Note when controlnet is set, image_strength will have no effect

see one example of how to make img2img request | | Output | Success: { error_code: 0, images=[ ‘image url’, ‘image url’, …] , credit_cost: 2.2, inf_id: 'xxxxxxxxxxxxxx' }

Failure: { error_code: 1, message: “This is an error message” } |

Model ID

You can call /models to get the complete model list, including image models, LoRAs and video models. The id of each model can be found in the returned json.

You can also go to sinkin.ai, enter a model page and the last part of the url is the model id. E.g. for the model at https://sinkin.ai/m/vlDnKP6, the model id is vlDnKP6.

The ids of some commonly used models:

Scheduler Options

An example of how to make an img2img request in python

    params = {
        'access_token': 'xxxxxxxxxxxxxxxxxxxxxxxxxx',
        'model_id': 'xxxxx',
        'prompt': 'an angry orc looking at camera smiling',
        'num_images': 1,
        'scale': 7,
        'steps': 30,
        'width': 512,
        'height': 768,
        # 'image_strength': 0.75,
        # 'controlnet': 'openpose'
    }

    files = {'init_image_file': open('path-to-image-file', 'rb')}
    
    r = requests.post('<https://sinkin.ai/api/inference>', files=files, data=params)

    print(r.text)

Models

Get all available image models, LoRAs and video models

error_code: 0, 
models: [ {'id': 'XXXX', 'title': 'XXXX', 'cover_img': 'xxxxxxxx', 'link': 'xxxxxxx'}, ... ] ,
loras: [ {'id': 'XXXX', 'title': 'XXXX', 'cover_img': 'xxxxxxxx', 'link': 'xxxxxxx'}, ... ] ,
video_models: [ {'id': ..., 'title': ..., ... }]

}`

Failure: { error_code: 1, message: “This is an error message” } |

Upscale