Model Training

• Use at least 10-12 high quality images.

• Ensure size of images is 512x512 px for training on SD-1.5 and 1024x1024 px for SDXL.

• Try to use a base model that is trained for the kind of images you have trained with. For example, here since the images are of a human, Realistic Vision is being used.

• Experiment with the Guidance Scale and LoRA weights (start from 0.5)

Train Model

ENDPOINT:

https://api.imagepipeline.io/train_model/v1/

AUTHORIZATION: API-Key

JSON BODY:

{
  "model_name": "Nisha",
  "model_type": "lora",
  "base_model": "sd1.5",
  "max_train_steps": 150,
  "validation_prompt": "Nisha woman ",
  "images": ["https://i.ibb.co/pdzjRdH/1.png", "https://i.ibb.co/tKhJv1n/2.png", "https://i.ibb.co/Z1qfc9q/3.png", "https://i.ibb.co/XDnyLKd/4.png", "https://i.ibb.co/zP9DtPj/5.png", "https://i.ibb.co/zP9DtPj/5.png", "https://i.ibb.co/Q8WDksv/7.png"],
  "trigger_words": "nisha",
  "description": "trained to make images of nisha",
  "tags": "lora, SD1.5",
  "nsfw": false,
  "visibility": "private"
}

Public Model Status

ENDPOINT:

https://api.imagepipeline.io/train_model/status/{{task_id}}

AUTHORIZATION: API-Key

Body needs to be set to none

Sync model

Once the training is complete, you will need to sync the model to use it.

Sync Models

ENDPOINT:

https://api.imagepipeline.io/sync_model/v1

AUTHORIZATION: API-Key

JSON BODY:

{
  "model_id": "Model ID"
}

Generate images

You can use the generation API endpoints to pass all the parameters to generate the image by passing the model_id in the lora_models parameter.

Generate Image

ENDPOINT:

https://api.imagepipeline.io/sd/text2image/v1/

AUTHORIZATION: API-Key

JSON BODY:

{
  "model_id":  "b1d00402-b0eb-4d7d-9fbf-f27fcd49c7c0",
  "prompt": "Nisha woman close up in sparkling princess gown in a forest",
  "negative_prompt": "error, cropped, worst quality, low quality, duplicate, blurry, (unfinished objects), stock image, artifacts, blurry face, ((deformed eyes)), ((deformed face)), bad proportion, (deformed iris), too many fingers",
  "num_inference_steps": 20,
  "refiner": false,
  "samples": 4,
  "guidance_scale": 5,
  "width": 512,
  "height": 512,
  "lora_models": ["8157a4c5-065b-4183-81ce-6b5238e073d6"],
  "lora_weights": ["0.7"]
}

JSON Parameters

Parameter	Type	Allowed Values
model_name	str
model_type	str	"lora"
base_model	str	"sd1.5", "sdxl"
max_train_steps	int	(0, 2000]
validation_prompt	str
images	str, array
trigger_words	str	Trigger words
description	str
tags	str	Optional comma-separated tags
nsfw	boolean	Whether model is not safe for work.
visibility	str	"private", "public"