Utility for extracting prompt metadata from Civitai AI images, auto-downloading the resources used to generate them, and outputting/formatting the prompt information in a manner defined by user-created templates.
This was originally conceived as a companion tool for Dream Factory, so the included example template will work with Dream Factory out of the box. Templates can easily be created for use with Automatic1111 and other tools.
- Batch extracts prompts and all associated metadata (dimensions, model, sampler, steps, scale, etc) from images downloaded from civitai.com. Also reads metadata from images created with Dream Factory.
- Automatically batch downloads necessary resources (models, LoRAs, embeddings) required to generate downloaded images. You may select the type(s) of resources that you want downloaded, a maximum file size for downloads, and blacklist resources that you never want downloaded.
- Outputs all extracted prompt information and associated metadata into user-specified format driven by simple template system (includes sample template for Dream Factory).
- LoRA references embedded in prompts that have user-specified paths are automatically modified to match your path structure (e.g.:
<lora:some\user\specified\path\lora_name:1.0>becomes<lora:lora_name:1.0>). - LoRAs that are referenced in the 'resources' section of Civitai's prompt metadata but aren't embedded in the actual prompt are automatically added.
- Optional specification of min/max values for certain prompt metadata values (e.g.: you may set max steps to 60, values over this will be set to 60).
- Optional filtering for automatic removal of unwanted words from positive/negative prompts.
- Optional filtering for automatic removal of unwanted LoRAs from prompts.
- Optional image dimension re-sizing to the closest "official" resolution for the base model that each image is generated from (while preserving aspect ratio).
- Sampler names that are unsupported in Automatic1111's webui are automatically translated into an appropriate supported sampler (e.g.: 'dpmpp_3m_sde' becomes 'DPM++ 3M SDE').
You'll just need a working Python environment (tested on 3.10). The setup instructions below assume that you're working within a suitable environment.
[1] Navigate to wherever you want Civitai Companion to live, then clone this repository and switch to its directory:
git clone https://github.com/rbbrdckybk/civitai-companion
cd civitai-companion
If you're already using Dream Factory, you should be good to go and can probably skip the following step.
[2] Install a few required packages:
pip install --no-input requests tqdm pillow
You can verify that everything works properly by running (or, you can simply type start if you're running Windows):
python civitai_reader.py --config_file config-example.txt
If everything is working, you should see Civitai Companion start, scan the seven sample images in the img-examples directory, extract their metadata, automatically download several LoRAs that are referenced by them, and create a civitai_[date]_[time].prompts output file that is usable by Dream Factory to create new images similar to the seven samples (check the template-example.txt file in the inc folder if you want to make changes to the output format for a different tool).
You'll notice errors that a few LoRAs couldn't be downloaded because they require a civitai.com API key. You can read Civitai's guide here to get your own API key (skip to the "How do I get an API token/key?" section). After you have an API key, add it to your config-example.txt file and re-run Civitai Companion. You should see that the LoRAs that previously could not be downloaded complete successfully now.
Read through the rest of the config-example.txt file and modify settings to suit your needs.
The following tokens may be placed into template files; Civitai Companion will replace these tokens with the values extracted from image metadata.
An example template file (template-example.txt) is located in the inc folder and will output prompt files compatible with Dream Factory.
[PROMPT]The positive prompt used to produce this image.[PROMPT_RAW]The positive prompt before any filters are applied.[NEG_PROMPT]The negative prompt used to produce this image.[NEG_PROMPT_RAW]The negative prompt before any filters are applied.[MODEL]The model used to produce this image.[SEED]The seed value used to produce this image.[SAMPLER]The sampler used to produce this image.[CLIP_SKIP]The clip skip setting used to produce this image.[WIDTH]The width of this image in pixels.[HEIGHT]The height of this image in pixels.[SCALE]The cfg scale value used to produce this image.[STEPS]The number of steps used to produce this image.[STRENGTH]The denoising strength used to produce this image (very few images include this).[BASE_MODEL]The base model (e.g.: SDXL 1.0, SD 1.5, Pony, etc) that the [MODEL] is based on.[MODEL_HASH]The hash of the [MODEL].[FILENAME]The original filename that this metadata was extracted from.[FILEPATH]The original full path and filename that this metadata was extracted from.[REF_NUM]A sequential reference number assigned to this image at processing time.[RAW_METADATA]The full raw (unprocessed) metadata extracted from this image.
- Configuration values may be specified in a config file and passed to Civitai Companion via the
--config_fileargument as demonstrated in the usage example or values may be passed via the command line. Values passed via command-line arguments override values set in a config file. This may be useful if you want to share a single config file between multiple workflows and simply override a few values on the command line for each. - If you need to do any troubleshooting or encounter errors, there is a log.txt file in the logs folder that contains details of the last run in much greater verbosity than what is displayed on the console.
- If you need to blacklist (e.g.: prevent from ever being downloaded again) a resource for whatever reason, add its civitai.com version ID to the do_not_download.txt file in the cache folder. If the resource has already been downloaded, its version ID should be in the civitai_version_ids.txt file. Simply search for the resource's name and copy the number at the start of the line (before the first comma) into the do_not_download.txt file.