LoRA for subject training

You can train a LoRA model to generate images featuring a subject such as yourself.

Create subject training task

POST https://api.novita.ai/v3/training/subject

Use this API to start a subject training task.

This is an asynchronous API; only the task_id is returned initially. Utilize this task_id to query the Task Result API at Get subject training result API in order to retrieve the results of the image generation.

Request header parameters

AuthorizationstringRequired
In Bearer {{Key}} format.
Content-TypestringRequired
Enum: application/json

Request Body parameters

namestringRequired
Task name of this model training.
base_modelstringRequired
Base models used for training.
Enum: stable-diffusion-xl-base-1.0, dreamshaperXL09Alpha_alpha2Xl10_91562, protovisionXLHighFidelity3D_release0630Bakedvae_154359, v1-5-pruned-emaonly, epicrealism_naturalSin_121250, chilloutmix_NiPrunedFp32Fix, abyssorangemix3AOM3_aom3a3_10864, dreamshaper_8_93211, WFChild_v1.0, majichenmixrealistic_v10, realisticVisionV51_v51VAE_94301, sdxlUnstableDiffusers_v11_216694, realisticVisionV40_v40VAE_81510, epicrealismXL_v10_247189
widthintegerRequired
Training images width, Minimum 1
heightintegerRequired
Training images height, Minimum 1
image_dataset_items[object]Required
ImageUrl and image caption
Show properties
expert_settingobject
Show properties
components[object]Required
Common parameters configured for training.
Show properties

Responses

task_idstring

curl --location 'https://api.novita.ai/v3/training/subject' \ --header 'Accept: application/json' \ --header 'Authorization: Bearer {{key}}' \ --header 'Content-Type: application/json' \ --data '{ 'name': 'test_1208_01', 'base_model': 'v1-5-pruned-emaonly', 'width': 512, 'height': 512, 'image_dataset_items': [ { 'assets_id': 26, 'caption': '' } ], 'expert_setting': { 'train_batch_size': 2, 'learning_rate': 0.0001, 'max_train_steps': 1, 'seed': 2023, 'lr_scheduler': 'constant', 'lr_warmup_steps': null, 'instance_prompt': 'Xstyle, of a young woman, profile shot, from side,sitting, looking at viewer, smiling, head tilt, eyes open,long black hair, glowing skin,light smile,cinematic lighting,dark environment', 'class_prompt': 'person', 'with_prior_preservation': true, 'prior_loss_weight': null, 'train_text_encoder': false, 'lora_r': null, 'lora_alpha': null, 'lora_text_encoder_r': null, 'lora_text_encoder_alpha': null }, 'components': [ { 'name': 'face_crop_region', 'args': [ { 'name': 'ratio', 'value': '1.0' } ] }, { 'name': 'resize', 'args': [ { 'name': 'width', 'value': '512' }, { 'name': 'height', 'value': '512' } ] }, { 'name': 'face_restore', 'args': [ { 'name': 'method', 'value': 'gfpgan_1.4' }, { 'name': 'upscale', 'value': '1.0' } ] } ] }'

{ 'task_id': 'fe84ac1b-5a01-487c-a56d-2c6c93b61f49' }

Get subject training result

GET https://api.novita.ai/v3/training/subject

Use this API to get the subject training result, including the model.

Request header parameters

AuthorizationstringRequired
In Bearer {{Key}} format.

Request Body parameters

task_idstringRequired

Responses

task_idstring
The task id of training.
task_statusstring
task training status
Enum: UNKNOWN, QUEUING, TRAINING, SUCCESS, CANCELED, FAILED
model_typestring
Model trained type.
Enum: lora
models[object]
models info
Show properties
extraobject
extra info
Show properties

Example

In this document we will explain step by step how to use our API for LoRA model training.

Generally, model training involves following steps.

Upload the images for model training.
Set training parameters and start the training.
Get the training results and generate images with the trained model.

1. Upload images for training

Currently we only supports uploading images in png / jpeg / webp format.
Each task supports uploading up to 50 images. In order to make the final effect good, the images uploaded should meet some basic conditions, such as: "portrait in the center", "no watermark", "clear picture", etc.

1.1 Get image upload URL

This interface returns the URL for single image to upload and can be called multiple times to upload images for training.

curl --location --request POST 'https://api.novita.ai/v3/assets/training_dataset' \
--header 'Authorization: Bearer {{key}}' \
--header 'Content-Type: application/json' \
--data-raw '{
    "file_extension": "png"
}'

Response:

{
    "assets_id": "34558688e2f42a0137ca2d5274a8cf43",
    "upload_url": "https://faas-training-dataset.s3.ap-southeast-1.amazonaws.com/test/******",
    "method": "PUT",
    "headers": {
        "Host": {
            "values": [
                "faas-training-dataset.s3.ap-southeast-1.amazonaws.com"
            ]
        }
    }
}

assets_id: The unique identifier of the image, which will be used in the training task.
upload_url: The URL for image upload.
method: The HTTP method for image upload.

1.2 Upload images

After obtaining the upload_url at step Get image upload URL, please refer to the following document to complete the image upload: https://docs.aws.amazon.com/zh_cn/AmazonS3/latest/userguide/PresignedUrlUploadObject.html.open in new window

Put images:

curl -X PUT -T "{{filepath}}" "{{upload_url}}"

or

curl --location --request PUT '{{upload_url}}' \
--header 'Content-Type: image/png' \
--data '{{filepath}}'

2. Start training task and configure parameters

In this step, we will begin the model training process, which is expected to take approximately 10 minutes, depending on the actual server's availability.

There are four types of parameters for model traning: Model info parameters, dataset parameters, components parameters,expert parameters, you can set them according to our tables below.

Here are some tips to train a good model:

At least 10 photos of faces that meet the requirements.
For parameters instance_prompt, we suggests using "a close photo of ohwx <man|woman>"
For parameters base_model, value v1-5-pruned-emaonly has better generalization ability and can be used in combination with various Base models, such as dreamshaper 2.5D, value epic-realism has a strong sense of reality.

Type	Parameters	Description
Model info parameters	name	Name of your training model
Model info parameters	base_model	base_model type
Model info parameters	width	Target image width
Model info parameters	height	Target image height
dataset parameters	image_dataset_items	Array: consist of `imageUrl` and image `caption`
dataset parameters	- image_dataset_items.assets_id	images assets_id, which can be found in step `Get image upload URL`
components parameters	components	Array: consist of `name` and `args`, this is a common parameters configured for training.
components parameters	- components.name	Type of components, Enum: `face_crop_region`, `resize`, `face_restore`
components parameters	- components.args	Detail values of components.name
expert parameters	expert_setting	expert parameters.
expert parameters	- instance_prompt	Captions for all the training images, here is a guidance of how to make a effective prompt : Click Hereopen in new window
expert parameters	- batch_size	batch size of training.
expert parameters	- max_train_steps	Max train steps, 500 is enought for lora model training.
expert parameters	- ......	More expert parameters can be access at api reference.

Here is a example of how to start training:

curl --location --request POST 'https://api.novita.ai/v3/training/subject' \
--header 'Accept: <Accept>' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer {{key}}' \
--data-raw '{
    "name": "test_subject_01",
    "base_model": "v1-5-pruned-emaonly",
    "width": 512,
    "height": 512,
    "image_dataset_items": [
        {
            "assets_id": "34558688e2f42a0137ca2d5274a8cf43"
        },
        {
            "assets_id": "1231231243f42a0137ca2d5274a8cf43"
        }
    ],
    "expert_setting": {
        "instance_prompt": "Xsubject, of a young woman, profile shot, from side,sitting, looking at viewer, smiling, head tilt, eyes open,long black hair, glowing skin,light smile,cinematic lighting,dark environment",
        "class_prompt": "person"
    },
    "components": [
        {
            "name": "face_crop_region",
            "args": [
                {
                    "name": "ratio",
                    "value": "1"
                }
            ]
        },
        {
            "name": "resize",
            "args": [
                {"name": "width", "value": "512"},
                {"name": "height", "value": "512"}
            ]
        },
        {
            "name": "face_restore",
            "args": [
                {"name": "method", "value": "gfpgan_1.4"},
                {"name": "upscale", "value": "1.0"}
            ]
        }
    ]
}'

Response：

{
    "task_id": "d660cdd0-ab9b-4a55-8b78-4bc851051fb0"
}

The task_id is the unique identifier of the training task, which can be used to query the training status and results.

3. Get training status

3.1 Get model training and deployment status

In this step, we will obtain the progress of model training and the status of model deployment after training

curl --location --request GET 'https://api.novita.ai/v3/training/subject?task_id=d660cdd0-ab9b-4a55-8b78-4bc851051fb0' \
--header 'Authorization: Bearer {{Key}}'

Response:

{
    "task_id": "d660cdd0-ab9b-4a55-8b78-4bc851051fb0",
    "task_status": "SUCCESS",
    "model_type": "",
    "models": [
        {
            "model_name": "model_1698904832_F2BB461625.safetensors",
            "model_status": "DEPLOYING"
        }
    ]
}

task_status: The status of the training task, Enum: UNKNOWN, QUEUING, TRAINING, SUCCESS, CANCELED, FAILED.
model_status: The status of the model, Enum: DEPLOYING, SERVING.
model_name: The name of the model, which can be used to generate images in next step.

When the task_status is SUCCESS, the model_status is SERVING we can starting to use the lora model.

3.2 Start using the trained model

After model deployed successfully, we can download the model files or generate images directly.

3.2.1 Use the generated models to create images

In order to use the trained lora models, We need to add model_name into the request of endpoint /v3/async/txt2img or /v3/async/img2img. Currently trained lora model can not be used in /v3 endpoint.

Below is a example of how to generate images with trained model:

Please set the Content-Type header to application/json in your HTTP request to indicate that you are sending JSON data. Currently, only JSON format is supported.

Request:

curl --location 'https://api.novita.ai/v3/async/txt2img' \
--header 'Authorization: Bearer {{key}};' \
--header 'Content-Type;' \
--data '{
  "extra": {
    "response_image_type": "jpeg"
  },
  "request": {
    "model_name": "realisticVisionV51_v51VAE_94301.safetensors",
    "prompt": "a young woman",
    "negative_prompt": "bottle, bad face",
    "sd_vae": "",
    "loras": [
      {
        "model_name": "model_1698904832_F2BB461625.safetensors",
        "strength": 0.7
      }
    ],
    "embeddings": [
      {
        "model_name": ""
      }
    ],
    "hires_fix": {
      "target_width": 1024,
      "target_height": 768,
      "strength": 0.5
    },
    "refiner": {
      "switch_at": null
    },
    "width": 512,
    "height": 384,
    "image_num": 2,
    "steps": 20,
    "seed": 123,
    "clip_skip": 1,
    "guidance_scale": 7.5,
    "sampler_name": "Euler a"
  }
}'

Response:

{
    "code": 0,
    "msg": "",
    "data": {
        "task_id": "bec2bcfe-47c6-4536-af34-f26cfe6fd457"
    }
}

Use task_id to get images

HTTP status codes in the 2xx range indicate that the request has been successfully accepted, while status codes in the 5xx range indicate internal server errors.

You can get images url in imgs of response.

Request:

curl --location 'https://api.novita.ai/v3/async/task-result?task_id=bec2bcfe-47c6-4536-af34-f26cfe6fd457' \
--header 'Authorization: Bearer {{key}}'

Response:

{
  "task": {
    "task_id": "bec2bcfe-47c6-4536-af34-f26cfe6fd457",
    "status": "TASK_STATUS_SUCCEED",
    "reason": ""
  },
  "images": [
    {
      "image_url": "https://faas-output-image.s3.ap-southeast-1.amazonaws.com/dev/replace_object_a910c8f7-76ce-40bd-b805-f00f3ddd7dc1_0.png?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIASVPYCN6LRCW3SOUV%2F20231019%2Fap-southeast-1%2Fs3%2Faws4_request&X-Amz-Date=20231019T084537Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&x-id=GetObject&X-Amz-Signature=b9ad40a5cb3aecf89602c15fe72d28be5d8a33e0bfe3656ce968295fde1aab31",
      "image_url_ttl": 3600,
      "image_type": "png"
    }
  ],
  "videos": [
    {
      "video_url": "https://faas-output-image.s3.ap-southeast-1.amazonaws.com/dev/replace_object_a910c8f7-76ce-40bd-b805-f00f3ddd7dc1_0.png?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIASVPYCN6LRCW3SOUV%2F20231019%2Fap-southeast-1%2Fs3%2Faws4_request&X-Amz-Date=20231019T084537Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&x-id=GetObject&X-Amz-Signature=b9ad40a5cb3aecf89602c15fe72d28be5d8a33e0bfe3656ce968295fde1aab31",
      "video_url_ttl": "3600",
      "video_type": "png"
    }
  ]
}

3.3 List training tasks

In this step, we can obtain all the info of trained models.

curl --location --request GET 'https://api.novita.ai/v3/training?pagination.limit=10&pagination.cursor=c_0' \
--header 'Authorization: Bearer {{key}}'

Response:

{
    "tasks": [
        {
            "task_name": "test_01",
            "task_id": "a0c4cc90-0296-4972-a1d8-e6e227daf094",
            "task_type": "subject",
            "task_status": "SUCCESS",
            "created_at": 1699325415,
            "models": [
                {
                    "model_name": "model_1699325939_E83A88DAC5.safetensors",
                    "model_status": "SERVING"
                }
            ]
        },
        {
            "task_name": "test_02",
            "task_id": "51e9bf41-8f7a-464d-b5ad-2fa217a1ec93",
            "task_type": "subject",
            "task_status": "SUCCESS",
            "created_at": 1699267268,
            "models": [
                {
                    "model_name": "model_1699267603_27F0D9C81C.safetensors",
                    "model_status": "SERVING"
                }
            ]
        },
        {
            "task_name": "test_03",
            "task_id": "7bd205ab-63e9-452b-9a66-39c597000eaa",
            "task_type": "subject",
            "task_status": "FAILED",
            "created_at": 1699264338,
            "models": []
        }
    ],
    "pagination": {
        "next_cursor": "c_10"
    }
}

task_name : The name of the training task.
task_id : The unique identifier of the training task, which can be used to query the training status and results.
task_type : The type of the training task.
task_status: The status of the training task, Enum: UNKNOWN, QUEUING, TRAINING, SUCCESS, CANCELED, FAILED.
created_at: The time when the training task was created.
model: The trained model.
model_name: The sd name of the model.
model_status: The status of the model, Enum: DEPLOYING, SERVING.

LoRA for subject training

LoRA for subject training

Create subject training task

Get subject training result

Example

1. Upload images for training

1.1 Get image upload URL

1.2 Upload images

2. Start training task and configure parameters

3. Get training status

3.1 Get model training and deployment status

3.2 Start using the trained model

3.2.1 Use the generated models to create images

3.3 List training tasks

4. Training playground

4.1 Input novita.ai API Key, images and select training type

4.2 Switch to the inferencing tab and add more detail

Review the training results