Add headless generation of images, and batch processing of prompts in a json file, directly from CLI.

[TimothyAlexisVass]

It's a start. Can be developed further.
Tested in Colab with:
!python entry_with_update.py --headless --prompt "Beautiful woman"
and
!python launch.py --headless --prompt Wow

Will solve #246 , #385 and #484

After adding the --headless option, there could also be a --batch path/to/bulk_commands.json.
That way, the whole environment wouldn't need to be reloaded for every generation.

[TimothyAlexisVass]

Okay, both --headless with and without --batch /path/to/batch_prompts_file.json has now been tested and it works fine for both entry_with_update.py and launch.py.

[baifagg]

Hello, I also have the need to generate pictures in batches. My job is to convert a lot of text usage fooocus given by users into pictures and store them locally. But my fooocus is deployed in the cloud, so I need a software that gets the base64 encoding of the fooocus generated image.

[baifagg]

Your code file directly replaces the corresponding code in the fooocus, can you use it successfully like your example?

[baifagg]

[
{
"prompt": "Note that seed -1 means 'random'. Only 'prompt' is required all other are optional and have default values. This example shows the default values",
"negative_prompt": "",
"styles": ["Fooocus V2", "Default (Slightly Cinematic)"],
"performance": "Speed",
"aspect_ratio": "1024x1024",
"image_number": 2,
"seed": -1,
"sharpness": 2.0,
"base_model": "sd_xl_base_1.0_0.9vae.safetensors",
"refiner_model": "sd_xl_refiner_1.0_0.9vae.safetensors",
"l1": "sd_xl_offset_example-lora_1.0.safetensors",
"w1": 0.5,
"l2": "None",
"w2": 0.5,
"l3": "None",
"w3": 0.5,
"l4": "None",
"w4": 0.5,
"l5": "None",
"w5": 0.5,
"current_tab": "",
"use_input_image": false,
"uov_method": "disabled",
"uov_input_image": null,
"outpaint": [],
"inpaint_input_image": null
},
{
"prompt": "This one will generate four images with aspect_ratio 1280x768",
"image_number": 4,
"aspect_ratio": "1280x768"
},
{
"prompt": "Here is another one with some other parameters set",
"negative_prompt": "You can include as many as you need",
"seed": "12345",
"performance": "Quality",
"sharpness": 10.0
}
]

How should the example call you provided be used? What I need is to change the way style is called every time.

[TimothyAlexisVass]

First of all, note that this is a Pull Request which is still awaiting approval.

Here is an example what you could do, with the same prompt for every batch:

my_batches.json

[
  {
    "prompt": "Your prompt",
    "styles": ["Fooocus V2", "The first style you want"]
  },
  {
    "prompt": "Your prompt",
    "styles": ["Fooocus V2", "The second style you want"]
  },
  {
    "prompt": "Your prompt",
    "styles": ["Fooocus V2", "The third style"]
  },
  ...add as many as you want, with the settings that you need...
]

Then you would run it using either entry_with_update.py or launch.py with the --headless and --batch parameters:
python entry_with_update.py --headless --batch /path/to/where/you/put/my_batches.json

But, you could also combine --headless with --prompt PROMPT and --batch /path/to/batch_file.json like this:

my_batches.json

[
  {
    "styles": ["Fooocus V2", "The first style you want"]
  },
  {
    "styles": ["Fooocus V2", "The second style you want"]
  },
  {
    "styles": ["Fooocus V2", "The third style"]
  }
]

And then set the same prompt for the whole batch like this:
python entry_with_update.py --headless --prompt "The prompt you want" --batch /path/to/my_batches.json

[xtremebeing]

I get an error after merging your PR. Any idea how to fix this?

python entry_with_update.py --headless --prompt "Beautiful woman"
...
LoRAs loaded: [('sd_xl_offset_example-lora_1.0.safetensors', 0.5), ('None', 0.5), ('None', 0.5), ('None', 0.5), ('None', 0.5)]
Fooocus Expansion engine loaded for cuda:0, use_fp16 = True.
loading new
'NoneType' object has no attribute 'local_url'
Traceback (most recent call last):
File "/home/x/Fooocus/modules/async_worker.py", line 439, in worker
handler(task)
File "/home/x/Fooocus/fooocus_env/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/home/x/Fooocus/fooocus_env/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/home/x/Fooocus/modules/async_worker.py", line 50, in handler
prompt, negative_prompt, style_selections, performance_selection,
ValueError: not enough values to unpack (expected 39, got 26)
Total time: 62.78 seconds
Image generation failed.

[TimothyAlexisVass]

Yeah, since there are now more parameters than when this PR was suggested, you are missing some of the parameters as indicated by expected 39, got 26.

[YellowTigerr]

Why do I still get an error when I modify the launch.py and add headless.py according to the method? launch.py: error: unrecognized arguments: --headless --batch