71 lines
1.9 KiB
TypeScript
71 lines
1.9 KiB
TypeScript
// http://localhost:3001/api/recipes/image/generate
|
||
|
||
import { meat, staple, vegetable } from '~/data/food'
|
||
|
||
// const sdBaseUrl = 'http://30.30.168.63:7860/'
|
||
const sdBaseUrl = 'https://85db1802ae46e57aab.gradio.live/'
|
||
|
||
const payload = {
|
||
// denoising_strength: 0,
|
||
prompt: 'food', // 提示词
|
||
negative_prompt: '', // 反向提示词
|
||
seed: -1, // 种子，随机数
|
||
batch_size: 2, // 每次张数
|
||
n_iter: 1, // 生成批次
|
||
steps: 50, // 生成步数
|
||
cfg_scale: 7, // 关键词相关性
|
||
width: 512, // 宽度
|
||
height: 512, // 高度
|
||
restore_faces: false, // 脸部修复
|
||
tiling: false, // 可平埔
|
||
// override_settings: {
|
||
// sd_model_checkpoint: 'wlop-any.ckpt [7331f3bc87]',
|
||
// }, // 一般用于修改本次的生成图片的stable diffusion 模型，用法需保持一致
|
||
// script_args: [
|
||
// 0,
|
||
// true,
|
||
// true,
|
||
// 'LoRA',
|
||
// 'dingzhenlora_v1(fa7c1732cc95)',
|
||
// 1,
|
||
// 1,
|
||
// ], // 一般用于lora模型或其他插件参数，如示例，我放入了一个lora模型， 1，1为两个权重值，一般只用到前面的权重值1
|
||
sampler_index: 'Euler', // 采样方法
|
||
}
|
||
|
||
// with api /sdapi/v1/txt2img
|
||
|
||
export interface Txt2ImgResponse {
|
||
images: string[]
|
||
}
|
||
|
||
const stuffItems = [
|
||
...vegetable,
|
||
...meat,
|
||
...staple,
|
||
]
|
||
|
||
export default defineEventHandler(async (e) => {
|
||
const body = await readBody(e)
|
||
|
||
const zhFoods = body.foods as string[]
|
||
const enFoods = zhFoods.map((food) => {
|
||
const item = stuffItems.find(item => item.name === food)
|
||
if (item)
|
||
return item.en
|
||
return ''
|
||
})
|
||
|
||
// TODO: 过滤 prompt 只能是食材
|
||
payload.prompt = `<lora:TODO:1>,food focus,transparent background,${enFoods.join(',')}`
|
||
|
||
// console.log(payload.prompt)
|
||
|
||
const data = await $fetch<Txt2ImgResponse>('/sdapi/v1/txt2img', {
|
||
baseURL: sdBaseUrl,
|
||
body: payload,
|
||
method: 'POST',
|
||
})
|
||
return data
|
||
})
|