a library to read metadata from images created by Stable Diffusion
Project description
[!IMPORTANT]
Version 0.3 featuring a revisited API and extended ComfyUI & InvokeAI parsing logic is available.See release notes for changes.
Features
Supports reading metadata from images generated with:
- Automatic1111's Stable Diffusion web UI
- ComfyUI *
- InvokeAI
- NovelAI
Provides a list of prompts used in the generation of the image, as well as generator-specific metadata.
* Custom ComfyUI nodes might parse incorrectly / with incomplete data.
Installation
pip install sd-parsers
Usage
For a simple query, import ParserManager
from sd_parsers
and use its parse()
method to parse an image. (see examples)
Basic usage:
Read prompt information from a given filename:
from sd_parsers import ParserManager
parser_manager = ParserManager()
prompt_info = parser_manager.parse("image.png")
if prompt_info:
for prompt in prompt_info.prompts:
print(f"Prompt: {prompt.value}")
Read prompt information from an already opened image:
from PIL import Image
from sd_parsers import ParserManager
parser_manager = ParserManager()
with Image.open('image.png') as image:
prompt_info = parser_manager.parse(image)
Each parser module can also be used directly, omitting the use of ParserManager
:
from PIL import Image
from sd_parsers.exceptions import ParserError
from sd_parsers.parsers import AUTOMATIC1111Parser
parser = AUTOMATIC1111Parser()
with Image.open("image.png") as image:
prompt_info, error = parser.read_parameters(image)
if prompt_info:
# the following can be omitted for an equivalent
# of ParserManager(lazy_read=True)
try:
prompt_info.parse()
except ParserError:
...
Output
The output returned from ParserManager
is a PromptInfo
type with the following properties (or None
if no metadata was found):
-
generator
: Specifies the image generator that may have been used for creating the image. -
prompts
: Prompts as found in the parsed metadata. -
negative_prompts
: Negative prompts as found in the parsed metadata. -
samplers
: Samplers used in the image generation process. -
models
: Models used in the image generation process. -
metadata
: Additional metadata which could not be attributed to one of the former described.Highly dependent on the provided data structure of the respective image generator.
-
parameters
: A dictionary of unmodified metadata entries as found in the parsed image (if present).
Contributing
As i don't have the time and resources to keep up with all the available AI-based image generators out there, the scale and features of this library is depending greatly on your help.
If you find the sd-parsers library unable to read metadata from an image, feel free to open an issue.
See CONTRIBUTING.md, if you are willing to help with improving the library itself and/or to create/maintain an additional parser module.
Credits
Idea and motivation using AUTOMATIC1111's stable diffusion webui
Example workflows for testing the ComfyUI parser
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for sd_parsers-0.3.1rc1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | bd7100f06097db580a88198568394373113f3f9ec2fa7346cacc612b92f852a1 |
|
MD5 | 676f8a014859714f8d50b67278060a42 |
|
BLAKE2b-256 | 230ba5acdbfeb0ed4c9485df617d1ce86cc7166c679727dbb2403337335f099c |