edenartlab/sdxl-lora-trainer:e88fc5fa – Run with an API on Replicate

Version

You're looking at a specific version of this model. Jump to the model overview.

edenartlab /sdxl-lora-trainer:e88fc5fa

Playground API

Input

name

string

Shift + Return to add a new line

Name of new LORA concept

Default: "unnamed"

lora_training_urls

string

Shift + Return to add a new line

Training images for new LORA concept (can be image urls or a .zip file of images)

concept_mode

string

Shift + Return to add a new line

'face' / 'style' / 'object' (default)

Default: "object"

seed

integer

Random seed for reproducible training. Leave empty to use a random seed

resolution

integer

Square pixel resolution which your images will be resized to for training recommended [768-1024]

Default: 960

train_batch_size

integer

Batch size (per device) for training

Default: 4

num_train_epochs

integer

Number of epochs to loop through your training dataset

Default: 10000

max_train_steps

integer

Number of individual training steps. Takes precedence over num_train_epochs

Default: 800

checkpointing_steps

integer

Number of steps between saving checkpoints. Set to very very high number to disable checkpointing, because you don't need one.

Default: 10000

is_lora

boolean

Whether to use LoRA training. If set to False, will use Full fine tuning

Default: true

prodigy_d_coef

number

Multiplier for internal learning rate of Prodigy optimizer

Default: 0.8

ti_lr

number

Learning rate for training textual inversion embeddings. Don't alter unless you know what you're doing.

Default: 0.001

ti_weight_decay

number

weight decay for textual inversion embeddings. Don't alter unless you know what you're doing.

Default: 0.0003

lora_weight_decay

number

weight decay for lora parameters. Don't alter unless you know what you're doing.

Default: 0.002

l1_penalty

number

Sparsity penalty for the LoRA matrices, increases merge-ability and maybe generalization

Default: 0.1

lora_param_scaler

number

Multiplier for the starting weights of the lora matrices

Default: 0.5

snr_gamma

number

see https://arxiv.org/pdf/2303.09556.pdf, set to None to disable snr training

Default: 5

lora_rank

integer

Rank of LoRA embeddings. For faces 5 is good, for complex concepts / styles you can try 8 or 12

Default: 12

caption_prefix

string

Shift + Return to add a new line

Prefix text prepended to automatic captioning. Must contain the 'TOK'. Example is 'a photo of TOK, '. If empty, chatgpt will take care of this automatically

Default: ""

left_right_flip_augmentation

boolean

Add left-right flipped version of each img to the training data, recommended for most cases. If you are learning a face, you prob want to disable this

Default: true

augment_imgs_up_to_n

integer

Apply data augmentation (no lr-flipping) until there are n training samples (0 disables augmentation completely)

Default: 20

mask_target_prompts

string

Shift + Return to add a new line

Prompt that describes most important part of the image, will be used for CLIP-segmentation. For example, if you are learning a person 'face' would be a good segmentation prompt

crop_based_on_salience

boolean

If you want to crop the image to `target_size` based on the important parts of the image, set this to True. If you want to crop the image based on face detection, set this to False

Default: true

use_face_detection_instead

boolean

If you want to use face detection instead of CLIPSeg for masking. For face applications, we recommend using this option.

Default: false

clipseg_temperature

number

How blurry you want the CLIPSeg mask to be. We recommend this value be something between `0.5` to `1.0`. If you want to have more sharp mask (but thus more errorful), you can decrease this value.

Default: 0.6

verbose

boolean

verbose output

Default: true

run_name

string

Shift + Return to add a new line

Subdirectory where all files will be saved

Default: "1701359580"

debug

boolean

for debugging locally only (dont activate this on replicate)

Default: false

hard_pivot

boolean

Use hard freeze for ti_lr. If set to False, will use soft transition of learning rates

Default: false

off_ratio_power

number

How strongly to correct the embedding std vs the avg-std (0=off, 0.05=weak, 0.1=standard)

Default: 0.1

Run this model in Node.js with one line of code:

npx create-replicate --model=edenartlab/sdxl-lora-trainer

or set up a project from scratch

Install Replicate’s Node.js client library:

npm install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import and set up the client:

import Replicate from "replicate";

const replicate = new Replicate({
  auth: process.env.REPLICATE_API_TOKEN,
});

Run edenartlab/sdxl-lora-trainer using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const output = await replicate.run(
  "edenartlab/sdxl-lora-trainer:e88fc5fa7857bcc5d9aa3c76de8f59ef0ef4aa8f2206f20ccee9e6795a4cad2e",
  {
    input: {
      name: "unnamed",
      debug: false,
      ti_lr: 0.001,
      is_lora: true,
      verbose: true,
      run_name: "1701359580",
      lora_rank: 12,
      snr_gamma: 5,
      hard_pivot: false,
      l1_penalty: 0.1,
      resolution: 960,
      concept_mode: "object",
      caption_prefix: "",
      prodigy_d_coef: 0.8,
      max_train_steps: 800,
      off_ratio_power: 0.1,
      ti_weight_decay: 0.0003,
      num_train_epochs: 10000,
      train_batch_size: 4,
      lora_param_scaler: 0.5,
      lora_weight_decay: 0.002,
      checkpointing_steps: 10000,
      clipseg_temperature: 0.6,
      augment_imgs_up_to_n: 20,
      crop_based_on_salience: true,
      use_face_detection_instead: false,
      left_right_flip_augmentation: true
    }
  }
);
console.log(output);

To learn more, take a look at the guide on getting started with Node.js.

Install Replicate’s Python client library:

pip install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import the client:

import replicate

Run edenartlab/sdxl-lora-trainer using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

output = replicate.run(
    "edenartlab/sdxl-lora-trainer:e88fc5fa7857bcc5d9aa3c76de8f59ef0ef4aa8f2206f20ccee9e6795a4cad2e",
    input={
        "name": "unnamed",
        "debug": False,
        "ti_lr": 0.001,
        "is_lora": True,
        "verbose": True,
        "run_name": "1701359580",
        "lora_rank": 12,
        "snr_gamma": 5,
        "hard_pivot": False,
        "l1_penalty": 0.1,
        "resolution": 960,
        "concept_mode": "object",
        "caption_prefix": "",
        "prodigy_d_coef": 0.8,
        "max_train_steps": 800,
        "off_ratio_power": 0.1,
        "ti_weight_decay": 0.0003,
        "num_train_epochs": 10000,
        "train_batch_size": 4,
        "lora_param_scaler": 0.5,
        "lora_weight_decay": 0.002,
        "checkpointing_steps": 10000,
        "clipseg_temperature": 0.6,
        "augment_imgs_up_to_n": 20,
        "crop_based_on_salience": True,
        "use_face_detection_instead": False,
        "left_right_flip_augmentation": True
    }
)

# The edenartlab/sdxl-lora-trainer model can stream output as it's running.
# The predict method returns an iterator, and you can iterate over that output.
for item in output:
    # https://replicate.com/edenartlab/sdxl-lora-trainer/api#output-schema
    print(item)

To learn more, take a look at the guide on getting started with Python.

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Run edenartlab/sdxl-lora-trainer using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d $'{
    "version": "e88fc5fa7857bcc5d9aa3c76de8f59ef0ef4aa8f2206f20ccee9e6795a4cad2e",
    "input": {
      "name": "unnamed",
      "debug": false,
      "ti_lr": 0.001,
      "is_lora": true,
      "verbose": true,
      "run_name": "1701359580",
      "lora_rank": 12,
      "snr_gamma": 5,
      "hard_pivot": false,
      "l1_penalty": 0.1,
      "resolution": 960,
      "concept_mode": "object",
      "caption_prefix": "",
      "prodigy_d_coef": 0.8,
      "max_train_steps": 800,
      "off_ratio_power": 0.1,
      "ti_weight_decay": 0.0003,
      "num_train_epochs": 10000,
      "train_batch_size": 4,
      "lora_param_scaler": 0.5,
      "lora_weight_decay": 0.002,
      "checkpointing_steps": 10000,
      "clipseg_temperature": 0.6,
      "augment_imgs_up_to_n": 20,
      "crop_based_on_salience": true,
      "use_face_detection_instead": false,
      "left_right_flip_augmentation": true
    }
  }' \
  https://api.replicate.com/v1/predictions

To learn more, take a look at Replicate’s HTTP API reference docs.

Output

No output yet! Press "Submit" to start a prediction.