findix/sd-scripts | Run with an API on Replicate

findix / sd-scripts

Training LoRA by sd-scripts

Cold

Public
20 runs
A100 (80GB)
GitHub

Iterate in playground

Run with an API

Playground API Examples README Versions

Input

pretrained_model_name_or_path

string

Shift + Return to add a new line

base model name or path | 底模名称或路径

Default: "CompVis/stable-diffusion-v1-4"

train_data_zip

*file

train dataset zip file | 训练数据集zip压缩包

network_weights

file

pretrained weights for LoRA network | 若需要从已有的 LoRA 模型上继续训练，请上传文件

training_comment

string

Shift + Return to add a new line

this LoRA model credit from replicate-sd-scriptsthis LoRA model credit from replicate-sd-scripts

training_comment | 训练介绍，可以写作者名或者使用触发关键词

Default: "this LoRA model credit from replicate-sd-scripts"

output_name

string

Shift + Return to add a new line

output model name | 模型保存名称

save_model_as

string

model save ext | 模型保存格式 ckpt, pt, safetensors

Default: "safetensors"

resolution

string

Shift + Return to add a new line

image resolution must be 'size' or 'width,height'. 图片分辨率，正方形边长或宽,高。支持非正方形，但必须是 64 倍数

Default: "512"

batch_size

integer

(minimum: 1)

batch size 一次性训练图片批处理数量，根据显卡质量对应调高

Default: 1

max_train_epoches

integer

(minimum: 1)

max train epoches | 最大训练 epoch

Default: 10

save_every_n_epochs

integer

(minimum: 1)

save every n epochs | 每 N 个 epoch 保存一次

Default: 2

network_dim

integer

(minimum: 1)

network dim | 常用 4~128，不是越大越好

Default: 32

network_alpha

integer

(minimum: 1)

network alpha | 常用与 network_dim 相同的值或者采用较小的值，如 network_dim的一半防止下溢。默认值为 1，使用较小的 alpha 需要提升学习率

Default: 32

train_unet_only

boolean

train U-Net only | 仅训练 U-Net，开启这个会牺牲效果大幅减少显存使用。6G显存可以开启

Default: false

train_text_encoder_only

boolean

train Text Encoder only | 仅训练文本编码器

Default: false

seed

integer

(minimum: 1)

reproducable seed | 设置跑测试用的种子，输入一个prompt和这个种子大概率得到训练图。可以用来试触发关键词

Default: 1337

noise_offset

number

(minimum: 0, maximum: 1)

noise offset | 在训练中添加噪声偏移来改良生成非常暗或者非常亮的图像，如果启用，推荐参数为 0.1

Default: 0

keep_tokens

integer

(minimum: 0)

keep heading N tokens when shuffling caption tokens | 在随机打乱 tokens 时，保留前 N 个不变

Default: 0

learning_rate

number

(minimum: 0)

Learning rate | 学习率

Default: 0.00006

unet_lr

number

(minimum: 0)

UNet learning rate | UNet 学习率

Default: 0.00006

text_encoder_lr

number

(minimum: 0)

Text Encoder learning rate | Text Encoder 学习率

Default: 0.000007

lr_scheduler

string

"linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup" | PyTorch自带6种动态学习率函数 constant，常量不变, constant_with_warmup 线性增加后保持常量不变, linear 线性增加线性减少, polynomial 线性增加后平滑衰减, cosine 余弦波曲线, cosine_with_restarts 余弦波硬重启，瞬间最大值。推荐默认cosine_with_restarts或者polynomial，配合输出多个epoch结果更玄学

Default: "cosine_with_restarts"

lr_warmup_steps

integer

(minimum: 0)

warmup steps | 仅在 lr_scheduler 为 constant_with_warmup 时需要填写这个值

Default: 0

lr_scheduler_num_cycles

integer

(minimum: 1)

cosine_with_restarts restart cycles | 余弦退火重启次数，仅在 lr_scheduler 为 cosine_with_restarts 时起效

Default: 1

min_bucket_reso

integer

(minimum: 1)

arb min resolution | arb 最小分辨率

Default: 256

max_bucket_reso

integer

(minimum: 1)

arb max resolution | arb 最大分辨率

Default: 1024

persistent_data_loader_workers

boolean

makes workers persistent, further reduces/eliminates the lag in between epochs. however it may increase memory usage | 跑的更快，吃内存。大概能提速2.5倍，容易爆内存，保留加载训练集的worker，减少每个 epoch 之间的停顿

Default: true

clip_skip

integer

(minimum: 0)

clip skip | 玄学一般用 2

Default: 2

optimizer_type

string

优化器，"adaFactor","AdamW","AdamW8bit","Lion","SGDNesterov","SGDNesterov8bit","DAdaptation", 推荐新优化器Lion。推荐学习率unetlr=lr=6e-5,tenclr=7e-6

Default: "Lion"

network_module

string

Network module

Default: "networks.lora"

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

npx create-replicate --model=findix/sd-scripts

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";
import fs from "node:fs";

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

Run findix/sd-scripts using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const output = await replicate.run(
  "findix/sd-scripts:43889e982cd38203192f5b6f008b147aa1d8eee795121cb4c11a034d88449bf5",
  {
    input: {
      seed: 1337,
      unet_lr: 0.00006,
      clip_skip: 2,
      batch_size: 1,
      resolution: "512",
      keep_tokens: 0,
      network_dim: 32,
      lr_scheduler: "cosine_with_restarts",
      noise_offset: 0,
      learning_rate: 0.00006,
      network_alpha: 32,
      save_model_as: "safetensors",
      network_module: "networks.lora",
      optimizer_type: "Lion",
      lr_warmup_steps: 0,
      max_bucket_reso: 1024,
      min_bucket_reso: 256,
      text_encoder_lr: 0.000007,
      train_unet_only: false,
      training_comment: "this LoRA model credit from replicate-sd-scripts",
      max_train_epoches: 10,
      save_every_n_epochs: 2,
      lr_scheduler_num_cycles: 1,
      train_text_encoder_only: false,
      pretrained_model_name_or_path: "CompVis/stable-diffusion-v1-4",
      persistent_data_loader_workers: true
    }
  }
);

// To access the file URL:
console.log(output.url()); //=> "http://example.com"

// To write the file to disk:
fs.writeFile("my-image.png", 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 findix/sd-scripts using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

output = replicate.run(
    "findix/sd-scripts:43889e982cd38203192f5b6f008b147aa1d8eee795121cb4c11a034d88449bf5",
    input={
        "seed": 1337,
        "unet_lr": 0.00006,
        "clip_skip": 2,
        "batch_size": 1,
        "resolution": "512",
        "keep_tokens": 0,
        "network_dim": 32,
        "lr_scheduler": "cosine_with_restarts",
        "noise_offset": 0,
        "learning_rate": 0.00006,
        "network_alpha": 32,
        "save_model_as": "safetensors",
        "network_module": "networks.lora",
        "optimizer_type": "Lion",
        "lr_warmup_steps": 0,
        "max_bucket_reso": 1024,
        "min_bucket_reso": 256,
        "text_encoder_lr": 0.000007,
        "train_unet_only": False,
        "training_comment": "this LoRA model credit from replicate-sd-scripts",
        "max_train_epoches": 10,
        "save_every_n_epochs": 2,
        "lr_scheduler_num_cycles": 1,
        "train_text_encoder_only": False,
        "pretrained_model_name_or_path": "CompVis/stable-diffusion-v1-4",
        "persistent_data_loader_workers": True
    }
)
print(output)

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 findix/sd-scripts 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": "findix/sd-scripts:43889e982cd38203192f5b6f008b147aa1d8eee795121cb4c11a034d88449bf5",
    "input": {
      "seed": 1337,
      "unet_lr": 0.00006,
      "clip_skip": 2,
      "batch_size": 1,
      "resolution": "512",
      "keep_tokens": 0,
      "network_dim": 32,
      "lr_scheduler": "cosine_with_restarts",
      "noise_offset": 0,
      "learning_rate": 0.00006,
      "network_alpha": 32,
      "save_model_as": "safetensors",
      "network_module": "networks.lora",
      "optimizer_type": "Lion",
      "lr_warmup_steps": 0,
      "max_bucket_reso": 1024,
      "min_bucket_reso": 256,
      "text_encoder_lr": 0.000007,
      "train_unet_only": false,
      "training_comment": "this LoRA model credit from replicate-sd-scripts",
      "max_train_epoches": 10,
      "save_every_n_epochs": 2,
      "lr_scheduler_num_cycles": 1,
      "train_text_encoder_only": false,
      "pretrained_model_name_or_path": "CompVis/stable-diffusion-v1-4",
      "persistent_data_loader_workers": 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.

Run time and cost

This model runs on Nvidia A100 (80GB) GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

…

Model description

…

Intended use

…

Ethical considerations

…

Caveats and recommendations

…