Prompt tweaks

[lstein]

This PR corrects the call signature for do_diffusion_step() and enhances the description of using prompt2prompt in the documentation.

[lstein]

@damian0815:

There's a problem when passing an empty prompt ("") to the prompt parser. I get this stack tracec:

│ /u/lstein/projects/SD/InvokeAI/./scripts/invoke.py:880 in <module>                               │
│                                                                                                  │
│   877 ######################################                                                     │
│   878                                                                                            │
│   879 if __name__ == '__main__':                                                                 │
│ ❱ 880 │   main()                                                                                 │
│   881                                                                                            │
│                                                                                                  │
│ /u/lstein/projects/SD/InvokeAI/./scripts/invoke.py:106 in main                                   │
│                                                                                                  │
│   103 │   │   )                                                                                  │
│   104 │                                                                                          │
│   105 │   try:                                                                                   │
│ ❱ 106 │   │   main_loop(gen, opt)                                                                │
│   107 │   except KeyboardInterrupt:                                                              │
│   108 │   │   print("\ngoodbye!")                                                                │
│   109                                                                                            │
│                                                                                                  │
│ /u/lstein/projects/SD/InvokeAI/./scripts/invoke.py:332 in main_loop                              │
│                                                                                                  │
│   329 │   │   │   │   catch_ctrl_c = infile is None # if running interactively, we catch keybo   │
│   330 │   │   │   │   opt.last_operation='generate'                                              │
│   331 │   │   │   │   try:                                                                       │
│ ❱ 332 │   │   │   │   │   gen.prompt2image(                                                      │
│   333 │   │   │   │   │   │   image_callback=image_writer,                                       │
│   334 │   │   │   │   │   │   step_callback=step_callback,                                       │
│   335 │   │   │   │   │   │   catch_interrupts=catch_ctrl_c,                                     │
│                                                                                                  │
│ /u/lstein/projects/SD/InvokeAI/ldm/generate.py:417 in prompt2image                               │
│                                                                                                  │
│    414 │   │   mask_image = None                                                                 │
│    415 │   │                                                                                     │
│    416 │   │   try:                                                                              │
│ ❱  417 │   │   │   uc, c, extra_conditioning_info = get_uc_and_c_and_ec(                         │
│    418 │   │   │   │   prompt, model =self.model,                                                │
│    419 │   │   │   │   skip_normalize=skip_normalize,                                            │
│    420 │   │   │   │   log_tokens    =self.log_tokenization                                      │
│                                                                                                  │
│ /u/lstein/projects/SD/InvokeAI/ldm/invoke/conditioning.py:65 in get_uc_and_c_and_ec              │
│                                                                                                  │
│    62 │   │   │   this_embedding, _ = build_embeddings_and_tokens_for_flattened_prompt(model,    │
│    63 │   │   │   embeddings_to_blend = this_embedding if embeddings_to_blend is None else tor   │
│    64 │   │   │   │   (embeddings_to_blend, this_embedding))                                     │
│ ❱  65 │   │   conditioning = WeightedFrozenCLIPEmbedder.apply_embedding_weights(embeddings_to_   │
│    66 │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   blend.we   │
│    67 │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   │   normaliz   │
│    68 │   else:                                                                                  │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
AttributeError: 'NoneType' object has no attribute 'unsqueeze'

[damian0815]

empty prompt crash is fixed in dc86fc92ce99af7c0f850a14bc4de845dd917c05. i have a unit test for it but including legacy blend support bypasses that code hah