Add Text2Image into Taskflow

[JunnYu]

PR types

New features

PR changes

Others

Description

Add Text2Image into Taskflow

import paddle
from paddlenlp import Taskflow
from IPython.display import display

model_text = {
    "pai-painter-painting-base-zh": ["风阁水帘今在眼，且来先看早梅红", "见说春风偏有贺，露花千朵照庭闹"],
    "pai-painter-scenery-base-zh": ["海边的日出", "黄昏日落"],
    "pai-painter-commercial-base-zh":
    ["女童套头毛衣打底衫秋冬针织衫童装儿童内搭上衣", "春夏真皮工作鞋女深色软皮久站舒适上班面试职业皮鞋"],
}
for model, prompts in model_text.items():
    text_to_image = Taskflow("text_to_image", model=model)
    images = text_to_image(prompts)
    for image in images:
        display(image)
    paddle.device.cuda.empty_cache()

import paddle
from paddlenlp import Taskflow
from IPython.display import display

prompts = [
    "New York Skyline with 'Google Research Pizza Cafe' written with fireworks on the sky.",
    "Dali painting of WALL·E"
]
model_text = {
    "dalle-mini": prompts,
    "dalle-mega-v16": prompts,
    "dalle-mega": prompts
}

for model, prompts in model_text.items():
    text_to_image = Taskflow("text_to_image", model=model)
    images = text_to_image(prompts)
    for image in images:
        display(image)
    paddle.device.cuda.empty_cache()

[guoshengCS]

这里也尽量隐去 artist 吧，这个模型名确实不见得有认可度，先用 pai这些模型权重来体现吧

[JunnYu]

use_faster：是否使用faster_generation，默认为False，目前支持faster_generation的模型有["pai-painter-painting-base-zh", "pai-painter-scenery-base-zh", "pai-painter-commercial-base-zh"]。

[guoshengCS]

这些后处理如果是和模型绑定的话后面也可以看看将这些方法放在transformers模型代码中提供出来，这里进行统一调用，方便更多模型扩展

[JunnYu]

done, 现已添加到model的generate内部。