sdwebuiapi

To use the Promptgen API, you must install the specific api-implementation branch of the Promptgen extension.

Once installed, you can:

1. List available models using api.list_prompt_gen_models().
2. Generate prompts using api.prompt_gen(text=...).

api.prompt_gen supports parameters like batch_size, min_length, max_length, temperature, and sampling_mode.

# Install via CLI
# cd stable-diffusion-webui/extensions
# git clone -b api-implementation https://github.com/davidmartinrius/stable-diffusion-webui-promptgen.git

# Usage
result = api.prompt_gen(
        text="a box", 
        model_name="AUTOMATIC/promptgen-majinai-unsafe",
        batch_size=10,
        min_length=20,
        max_length=150,
        temperature=1,
        sampling_mode="Top K",
        top_k=12
    )

Before using the client, you must enable API support in your AUTOMATIC1111/stable-diffusion-webui instance by adding the --api flag when running the webui.

If you require authentication, you can use the --api-auth username:password option. Note that because this uses basic HTTP authentication, credentials are transmitted in cleartext and are not protected unless using an encrypted communication channel (HTTPS).

Install the webuiapi client using pip to interact with the AUTOMATIC1111/stable-diffusion-webui API.

pip install webuiapi

ADetailer can be used in both txt2img and img2img calls by passing an ADetailer object in the adetailer list parameter.

import webuiapi

# txt2img with ADetailer
ads = webuiapi.ADetailer(ad_model="face_yolov8n.pt")
result1 = api.txt2img(prompt="cute squirrel",
                    adetailer=[ads],
                    steps=30)

# img2img with ADetailer
ads = webuiapi.ADetailer(ad_model="face_yolov8n.pt")
result1 = api.img2img(
    images=[img], 
    prompt="a cute squirrel", 
    adetailer=[ads],
)

Use the txt2img method to generate images from text prompts. You can specify parameters such as prompt, negative_prompt, seed, styles, cfg_scale, steps, enable_hr, denoising_strength, sampler_name, and scheduler. The result object contains an image attribute which is a PIL Image object.

import webuiapi

result1 = api.txt2img(prompt="cute squirrel",
                    negative_prompt="ugly, out of frame",
                    seed=-1,
                    styles=["anime"],
                    cfg_scale=1,
                    steps=20,
                    enable_hr=False,
                    denoising_strength=0.5,
                    sampler_name= "Euler",
                    scheduler= "Simple"
            )
                    
img = result1.image
img

# OR

file_path = "output_image.png"
result1.image.save(file_path)

Use the following methods to manage the WebUI configuration and models:

• get_options(): Returns a map of current options.
• set_options(options): Updates options. Warning: Do not pass the entire dictionary returned by get_options(); only pass the keys you wish to change to avoid making the WebUI unusable.
• get_sd_models(): Retrieves available Stable Diffusion models.
• get_samplers(), get_schedulers(), get_embeddings(), get_scripts(): Retrieve various WebUI assets.
• interrupt(): Interrupts the current process.
• skip(): Skips the current process.

# change sd model
options = {}
options['sd_model_checkpoint'] = 'model.ckpt [7460a6fa]'
api.set_options(options)

# get available sd models
api.get_sd_models()

Methods like txt2img, img2img, extra_single_image, and extra_batch_images support asynchronous calls. To use this, set use_async=True and ensure you have asyncio and aiohttp installed in your environment.

result = await api.txt2img(prompt="cute kitten",
                    seed=1001,
                    use_async=True
                    )
result.image

The util_ prefixed methods provide helper functions for common tasks:

• util_get_current_model(): Returns the name of the currently loaded model.
• util_get_model_names(): Returns a list of available model names.
• util_set_model(name): Sets the model using the exact name or the closest match.
• refresh_checkpoints(): Refreshes the list of available checkpoints.
• util_wait_for_ready(): Blocks until the WebUI is ready for a new job.

# get list of available models
models = api.util_get_model_names()

# set model (find closest match)
api.util_set_model('robodiffusion')

# wait for job complete
api.util_wait_for_ready()

To perform inpainting, use the img2img method and provide a mask_image (a PIL Image) along with the inpainting_fill parameter.

from PIL import Image, ImageDraw

# Create a mask
mask = Image.new('RGB', result2.image.size, color='black')
draw = ImageDraw.Draw(mask)
draw.ellipse((210, 150, 310, 250), fill='white')

# Run inpainting
inpainting_result = api.img2img(
    images=[result2.image],
    mask_image=mask,
    inpainting_fill=1,
    prompt="cute cat",
    seed=104,
    cfg_scale=5.0,
    denoising_strength=0.7
)

image = inpainting_result.image

The extra_single_image method allows for post-processing a single image, such as upscaling. Use the upscaler_1 parameter with values from webuiapi.Upscaler.

result3 = api.extra_single_image(
    image=result2.image,
    upscaler_1=webuiapi.Upscaler.ESRGAN_4x,
    upscaling_resize=1.5
)

image = result3.image