[ADD] T2I finetuning with SD1 and SD2 to contrib

[Aziily]

Text to image fine-tuning support for Stable Diffusion 1 and 2 series.

[Aziily]

I add a link for a small subset for dreambooth's dog as the example,
dog_t2i_data_example

[research4pan]

Some improvements that can be made before merging into main.

contrib/text2image/finetune_t2i.sh

• [Style] line 7: I am wondering if an example file can be provided here.
• [Style] line 3, 7: Better export --model_name_or_path, --dataset_path for customization.

README

• A README is preferred for the usage of the shell scripts, which 1) illustrates the dataset format, 2) and specifies environments (such as operating system, GPU version), 3) also how to customize the script.

contrib/text2image/requirements.txt

• Better specify the lowest required version of diffuser, e.g. diffuser >= x.xx.xx.

[Aziily]

Thanks for your advice. Here, I add script arguments according to scripts/run_finetune.sh and version requirements. Also, a readme is added for env/data preparation and script usage.

[research4pan]

LGTM