VoucherVisionGO Client

This repository contains only the client component of VoucherVisionGO, a tool for automatic label data extraction from museum specimen images.

Purpose

This repository is designed for users who only need the client component without the full VoucherVisionGO codebase, allowing for:

• Easier integration into existing projects
• Smaller footprint
• Focused functionality
• Simple installation process

Information

VoucherVision is designed to transcribe museum specimen labels. Please see the VoucherVision Github for more information.

The University of Michigan provides managed VoucherVision access subject to each account's quota. The API is hosted on demand, so a cold request can take about a minute while later requests are usually much faster. VoucherVisionGO uses supported Google models for OCR and for parsing unformatted label text into structured JSON.

Available LLM Models

Usage limits may exist for certain models if you are using VoucherVision credits. If you provide your own Gemini API Key or link a Vertex AI account, then limits are removed.

Model	VoucherVisionGO tier	Notes
gemini-3.1-flash-lite	General Access	Recommended default
gemini-2.5-flash-lite	General Access	Legacy inexpensive model
gemini-3.5-flash-lite	General Access	Inexpensive Flash-Lite model
gemini-3-flash-preview	Restricted Flash	Higher-cost Flash model
gemini-3.5-flash	Restricted Flash	Higher-cost Flash model
gemini-3.6-flash	Restricted Flash	Higher-cost Flash model
gemini-3.7-flash	Restricted Flash	Best tested performance/cost balance
gemini-3.8-flash	Restricted Flash	Very expensive
gemini-3.1-pro	Gemini Pro	Expensive but good
gemma-4-26b-a4b-it	Server-dependent	May run slowly or be unavailable
gemma-4-31b-it	Server-dependent	May run slowly or be unavailable

For the most up-to-date list of supported models, refer to the Google AI Gemini API documentation

If you want pure speed, use only "flash-lite" models with "low" thinking for both tasks.

If you want to transcribe different fields, reach out and I can help you develop a prompt or upload your existing prompt to make it available on the API.

Requirements

• Python 3.10 or higher
• External dependencies (see installation options below)

Authentication

To use the API you need to apply for an authorization token. Go to the login page and submit your info. Copy the token and store it in a safe location. Never put the token directly into your code. Always use environment variables or secrets.

Current Python client API

Version 0.2 adds a stateful client while retaining the existing functional entry points:

import os
from VoucherVision import VoucherVisionClient

with VoucherVisionClient(
    "https://vouchervision-go-738307415303.us-central1.run.app",
    os.environ["VVGO_API_KEY"],
) as client:
    result = client.process_image(
        fname="specimen",
        image_path="specimen.jpg",
        engines=["gemini-3.1-flash-lite"],
        llm_model="gemini-3.1-flash-lite",
        ocr_thinking_level="low",
        llm_thinking_level="low",
        include_wfo=False,
        include_cop90=False,
    )

Complete VoucherVisionClient payment examples

The examples below show every user-facing constructor option and every option accepted by client.process_image(). They deliberately set options even when the displayed value is already the default, so each example documents a complete calling state. Set ocr_only=True to omit JSON parsing, or set notebook_mode=True for OCR-only Markdown output. Leave both False for the normal OCR-and-parsing workflow shown here.

Only one inference payment method may be selected for a request. Supplying both gemini_api_key and vertex_project is rejected before processing.

VoucherVision Credits

Omit both user-supplied Google credentials to charge the request against the quota associated with the VoucherVision API key:

import os
from VoucherVision import VoucherVisionClient

server_url = "https://vouchervision-go-738307415303.us-central1.run.app"
image_path = "./specimen.jpg"

with VoucherVisionClient(
    server_url=server_url,
    auth_token=os.environ["VVGO_API_KEY"],
    gemini_api_key=None,
    vertex_project=None,
    vertex_region="global",
    timeout=(15, 900),
    session=None,
) as client:
    result = client.process_image(
        fname="specimen",
        image_path=image_path,
        output_dir="./output/credits",
        verbose=True,
        engines=["gemini-3.1-flash-lite"],
        llm_model="gemini-3.1-flash-lite",
        prompt="SLTPvM_full.yaml",
        ocr_only=False,
        notebook_mode=False,
        skip_label_collage=False,
        include_wfo=False,
        include_cop90=True,
        ocr_thinking_level="low",
        llm_thinking_level="low",
        on_error="raise",
    )

print(result["payment_inference"])  # VoucherVisionGO_credits

User-supplied Gemini API key

Set gemini_api_key to bill Gemini inference to a Google AI Studio API key:

import os
from VoucherVision import VoucherVisionClient

server_url = "https://vouchervision-go-738307415303.us-central1.run.app"
image_path = "./specimen.jpg"

with VoucherVisionClient(
    server_url=server_url,
    auth_token=os.environ["VVGO_API_KEY"],
    gemini_api_key=os.environ["GEMINI_API_KEY"],
    vertex_project=None,
    vertex_region="global",
    timeout=(15, 900),
    session=None,
) as client:
    result = client.process_image(
        fname="specimen",
        image_path=image_path,
        output_dir="./output/gemini",
        verbose=True,
        engines=["gemini-3.1-flash-lite"],
        llm_model="gemini-3.1-flash-lite",
        prompt="SLTPvM_full.yaml",
        ocr_only=False,
        notebook_mode=False,
        skip_label_collage=False,
        include_wfo=False,
        include_cop90=True,
        ocr_thinking_level="low",
        llm_thinking_level="low",
        on_error="raise",
    )

print(result["payment_inference"])  # user_supplied_gemini_key

User-supplied Vertex AI account

The project must first be linked to the same VoucherVision account through the website's API Settings tab. The region defaults to global:

import os
from VoucherVision import VoucherVisionClient

server_url = "https://vouchervision-go-738307415303.us-central1.run.app"
image_path = "./specimen.jpg"

with VoucherVisionClient(
    server_url=server_url,
    auth_token=os.environ["VVGO_API_KEY"],
    gemini_api_key=None,
    vertex_project=os.environ["VVGO_VERTEX_PROJECT"],
    vertex_region=os.environ.get("VVGO_VERTEX_REGION", "global"),
    timeout=(15, 900),
    session=None,
) as client:
    result = client.process_image(
        fname="specimen",
        image_path=image_path,
        output_dir="./output/vertex",
        verbose=True,
        engines=["gemini-3.1-flash-lite"],
        llm_model="gemini-3.1-flash-lite",
        prompt="SLTPvM_full.yaml",
        ocr_only=False,
        notebook_mode=False,
        skip_label_collage=False,
        include_wfo=False,
        include_cop90=True,
        ocr_thinking_level="low",
        llm_thinking_level="low",
        on_error="raise",
    )

print(result["payment_inference"])  # user_supplied_vertex_account

session=None lets VoucherVisionClient create and close its own requests.Session. Advanced callers may instead pass an existing session for connection pooling. timeout may be one number or a (connect, read) tuple.

For a hosted image, keep the same constructor for the selected payment method and call process_url() instead. It accepts the same processing options except for the local-output arguments fname, image_path, and output_dir:

result = client.process_url(
    "https://example.org/specimen.jpg",
    verbose=True,
    engines=["gemini-3.1-flash-lite"],
    llm_model="gemini-3.1-flash-lite",
    prompt="SLTPvM_full.yaml",
    ocr_only=False,
    notebook_mode=False,
    skip_label_collage=False,
    include_wfo=False,
    include_cop90=True,
    ocr_thinking_level="low",
    llm_thinking_level="low",
    on_error="raise",
)

Thinking levels are independent for OCR and parsing. The accepted values are low, medium, and high; Python/API requests default to low. Thinking tokens are returned separately in ocr_info and parsing_info, are priced at the model's output-token rate, and are included in total_request_cost_usd.

Successful responses can include:

{
  "payment_inference": "VoucherVisionGO_credits",
  "host": "VoucherVisionGO",
  "total_request_cost_usd": 0.0123,
  "ocr_info": {
    "gemini-3.1-flash-lite": {
      "tokens_in": 1200,
      "tokens_out": 300,
      "thinking_tokens": 1800,
      "cost_in": 0.0003,
      "cost_out": 0.00045,
      "thinking_cost": 0.0027,
      "total_cost": 0.00345
    }
  }
}

Structured errors

The stateful client raises VoucherVisionAPIError, exposing status_code, error_code, request_id, details, host, and payment_inference without printing raw provider responses. Existing functional calls retain their print-and-return-None behavior by default and accept on_error="raise" or on_error="return" when structured handling is preferred.

URL processing and filenames

HTTP and HTTPS image inputs are sent to /process-url; the server owns URL validation, streaming, retries and filename resolution. A filename returned by the server takes precedence over the client's local URL parser.

Quota and Vertex helpers

quota = client.quota_status()
projects = client.vertex_projects()
client.link_vertex_project("my-gcp-project", nickname="Herbarium billing")
client.revoke_vertex_project("my-gcp-project")

Asynchronous PDF jobs

The existing process_vouchers() path still renders local PDFs into page images. For server-side asynchronous processing:

job = client.submit_pdf(
    "labels.pdf",
    ocr_thinking_level="low",
    llm_thinking_level="low",
)
completed = client.wait_for_pdf(job["job_id"], poll_interval=5)
client.download_pdf(job["job_id"], "./output/labels-results.zip")

Async PDFs support VoucherVision Credits or a linked Vertex project. They do not currently support a user-supplied Gemini API key.

Using Curl

curl can interact with VoucherVisionGO directly, without Python or the VoucherVisionGO client package. Local files are sent to /process as multipart form data. The examples below include every processing field accepted by that route. Although curl infers POST from -F, -X POST is shown explicitly for clarity.

Set shell variables first so credentials do not appear directly in scripts:

export VVGO_SERVER_URL="https://vouchervision-go-738307415303.us-central1.run.app"
export VVGO_API_KEY="your-vouchervision-api-key"
export GEMINI_API_KEY="your-google-ai-studio-key"
export VVGO_VERTEX_PROJECT="your-linked-gcp-project"
export VVGO_VERTEX_REGION="global"
export IMAGE_PATH="./specimen.jpg"
export IMAGE_URL="https://example.org/specimen.jpg"

VoucherVision Credits with curl

Omit Gemini and Vertex billing credentials:

curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -F "file=@$IMAGE_PATH" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite" \
  -F "prompt=SLTPvM_full.yaml" \
  -F "ocr_only=false" \
  -F "notebook_mode=false" \
  -F "skip_label_collage=false" \
  -F "include_wfo=false" \
  -F "include_cop90=true" \
  -F "ocr_thinking_level=low" \
  -F "llm_thinking_level=low" \
  -o credits-response.json

User-supplied Gemini API key with curl

The preferred credential form is the dedicated request header:

curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -H "X-Gemini-API-Key: $GEMINI_API_KEY" \
  -F "file=@$IMAGE_PATH" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite" \
  -F "prompt=SLTPvM_full.yaml" \
  -F "ocr_only=false" \
  -F "notebook_mode=false" \
  -F "skip_label_collage=false" \
  -F "include_wfo=false" \
  -F "include_cop90=true" \
  -F "ocr_thinking_level=low" \
  -F "llm_thinking_level=low" \
  -o gemini-response.json

User-supplied Vertex AI account with curl

The project must already be linked to the VoucherVision account that owns VVGO_API_KEY:

curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -H "X-Vertex-Project: $VVGO_VERTEX_PROJECT" \
  -H "X-Vertex-Region: $VVGO_VERTEX_REGION" \
  -F "file=@$IMAGE_PATH" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite" \
  -F "prompt=SLTPvM_full.yaml" \
  -F "ocr_only=false" \
  -F "notebook_mode=false" \
  -F "skip_label_collage=false" \
  -F "include_wfo=false" \
  -F "include_cop90=true" \
  -F "ocr_thinking_level=low" \
  -F "llm_thinking_level=low" \
  -o vertex-response.json

Alternate curl input forms

The three complete examples above use the recommended headers. Billing values may alternatively be supplied through multipart form fields, query parameters, or—on /process-url—a JSON body. These names are accepted:

Value	Snake-case field/query name	Camel-case alias	Header
Gemini API key	gemini_api_key	geminiApiKey	X-Gemini-API-Key
Vertex project	vertex_project	vertexProject	X-Vertex-Project
Vertex region	vertex_region	vertexRegion	X-Vertex-Region

If the same billing value is supplied more than once, precedence is multipart form field, JSON field, query parameter, then header. Never send Gemini and Vertex credentials in the same request.

Billing credentials as multipart form fields

Gemini:

curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -F "file=@$IMAGE_PATH" \
  -F "gemini_api_key=$GEMINI_API_KEY" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite" \
  -F "prompt=SLTPvM_full.yaml" \
  -F "ocr_only=false" \
  -F "notebook_mode=false" \
  -F "skip_label_collage=false" \
  -F "include_wfo=false" \
  -F "include_cop90=true" \
  -F "ocr_thinking_level=low" \
  -F "llm_thinking_level=low"

Vertex, using the accepted camel-case aliases:

curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -F "file=@$IMAGE_PATH" \
  -F "vertexProject=$VVGO_VERTEX_PROJECT" \
  -F "vertexRegion=$VVGO_VERTEX_REGION" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite" \
  -F "prompt=SLTPvM_full.yaml" \
  -F "ocr_only=false" \
  -F "notebook_mode=false" \
  -F "skip_label_collage=false" \
  -F "include_wfo=false" \
  -F "include_cop90=true" \
  -F "ocr_thinking_level=low" \
  -F "llm_thinking_level=low"

Billing credentials as query parameters

Query parameters are supported but are not recommended for secrets because URLs are commonly retained in shell history, proxy logs, and request logs:

# Gemini query parameter
curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process?gemini_api_key=$GEMINI_API_KEY" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -F "file=@$IMAGE_PATH" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite"

# Vertex query parameters, using the accepted camel-case aliases
curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process?vertexProject=$VVGO_VERTEX_PROJECT&vertexRegion=$VVGO_VERTEX_REGION" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -F "file=@$IMAGE_PATH" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite"

URL input as multipart form data

The /process-url route accepts the same processing fields but replaces the uploaded file with image_url:

curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process-url" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -H "X-Gemini-API-Key: $GEMINI_API_KEY" \
  -F "image_url=$IMAGE_URL" \
  -F "engines=gemini-3.1-flash-lite" \
  -F "llm_model=gemini-3.1-flash-lite" \
  -F "prompt=SLTPvM_full.yaml" \
  -F "ocr_only=false" \
  -F "notebook_mode=false" \
  -F "skip_label_collage=false" \
  -F "include_wfo=false" \
  -F "include_cop90=true" \
  -F "ocr_thinking_level=low" \
  -F "llm_thinking_level=low"

URL input as JSON

JSON is supported by /process-url, including billing fields. It is not supported by /process, because a local file must be uploaded as multipart form data:

curl --max-time 900 --fail-with-body -sS \
  -X POST "$VVGO_SERVER_URL/process-url" \
  -H "X-API-Key: $VVGO_API_KEY" \
  -H "Content-Type: application/json" \
  --data "{
    \"image_url\": \"$IMAGE_URL\",
    \"engines\": [\"gemini-3.1-flash-lite\"],
    \"llm_model\": \"gemini-3.1-flash-lite\",
    \"prompt\": \"SLTPvM_full.yaml\",
    \"ocr_only\": false,
    \"notebook_mode\": false,
    \"skip_label_collage\": false,
    \"include_wfo\": false,
    \"include_cop90\": true,
    \"ocr_thinking_level\": \"low\",
    \"llm_thinking_level\": \"low\",
    \"vertex_project\": \"$VVGO_VERTEX_PROJECT\",
    \"vertex_region\": \"$VVGO_VERTEX_REGION\"
  }"

VoucherVision authentication itself also has alternatives. The preferred form is X-API-Key. WE DO NOT RECOMMEND USING THE "FIREBASE ID" TOKEN METHOD FOR AUTH. USE THE API KEY. An API key may instead be sent as ?api_key=.... Firebase ID tokens may be sent as Authorization: Bearer ... or ?token=...:

# VoucherVision API key in the query string (less secure than the header)
curl --fail-with-body -sS \
  "$VVGO_SERVER_URL/auth-check?api_key=$VVGO_API_KEY"

# Firebase ID token in an Authorization header
curl --fail-with-body -sS \
  -H "Authorization: Bearer $FIREBASE_ID_TOKEN" \
  "$VVGO_SERVER_URL/auth-check"

# Firebase ID token in the query string (less secure than the header)
curl --fail-with-body -sS \
  "$VVGO_SERVER_URL/auth-check?token=$FIREBASE_ID_TOKEN"

Client-only options such as output_dir, verbose, save_to_xlsx, and max_workers do not exist as HTTP fields. They control local Python behavior, not server processing.

Bill Vertex AI to your own Google Cloud project (optional)

If Google AI Studio API keys aren't available in your region, or your institution requires that AI costs land on your own Google Cloud account, you can have Gemini inference billed to your GCP project via Vertex AI. Pass vertex_project (and optionally vertex_region, which defaults to "global") instead of gemini_api_key.

This requires a one-time setup in Google Cloud and linking your project ID to your VoucherVisionGO account. The full walkthrough lives in the API Settings tab at leafmachine.org/vouchervisiongo. The client will only accept vertex_project values that you have linked there.

Python:

import os
from VoucherVision import process_vouchers

auth_token = os.environ["your_auth_token"]

process_vouchers(
    server="https://vouchervision-go-738307415303.us-central1.run.app/",
    output_dir="./output",
    image="path/to/image.jpg",
    auth_token=auth_token,
    vertex_project="your-gcp-project-id",  # vertex_region defaults to "global"
)

CLI:

vouchervision \
  --server https://vouchervision-go-738307415303.us-central1.run.app \
  --auth-token "$VVGO_TOKEN" \
  --image path/to/image.jpg \
  --output-dir ./output \
  --vertex-project your-gcp-project-id

Pick one auth method per request — supplying both gemini_api_key and vertex_project returns HTTP 400.

Installation

Choose one of the following installation methods:

Option 1: Install in your own Python environment from the PyPi repo

Install

pip install vouchervision-go-client[full]

Upgrade

pip install --upgrade vouchervision-go-client[full]

Note: You may need to install these packages too:

pip install requests pandas termcolor tabulate tqdm

Option 2: Using pip (Install from source locally)

# Clone
git clone https://github.com/Gene-Weaver/VoucherVisionGO-client.git
cd VoucherVisionGO-client
# Create a virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Option 3: Using conda (Install from source locally)

# Clone
git clone https://github.com/Gene-Weaver/VoucherVisionGO-client.git
cd VoucherVisionGO-client
# Create a virtual environment
conda create -n vvgo-client python=3.10
conda activate vvgo-client

# Install dependencies
pip install -r requirements.txt

Usage Guide (Option 1)

Programmatic Usage

You can also use the client functions in your own Python code. Install VoucherVisionGO-client from PyPi:

import os
from VoucherVision import process_vouchers

if __name__ == '__main__':
  auth_token = os.environ.get("your_auth_token") # Add auth token as an environment variable or secret

  process_vouchers(
    server="https://vouchervision-go-738307415303.us-central1.run.app/", 
    output_dir="./output", 
    prompt="SLTPvM_full_chromosome.yaml", 
    image="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg", 
    llm_model="gemini-3.1-flash-lite",
    ocr_thinking_level="low",
    llm_thinking_level="low",
    directory=None, 
    file_list=None, 
    verbose=True, 
    save_to_xlsx=True, 
    max_workers=4,
    auth_token=auth_token)  

  process_vouchers(
    server="https://vouchervision-go-738307415303.us-central1.run.app/", 
    output_dir="./output2", 
    prompt="SLTPvM_full_chromosome.yaml", 
    image=None, 
    llm_model=None, # Use the default LLM
    directory="D:/Dropbox/VoucherVisionGO/demo/images", 
    file_list=None, 
    verbose=True, 
    save_to_xlsx=True, 
    max_workers=4,
    auth_token=auth_token)  

To get the JSON packet for a single specimen record:

import os
from VoucherVision import process_image, ordereddict_to_json, get_output_filename

if __name__ == '__main__':
  auth_token = os.environ.get("your_auth_token") # Add auth token as an environment variable or secret

  image_path = "https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg"
  output_dir = "./output"
  output_file, _ = get_output_filename(image_path, output_dir)  # returns (json_path, md_path)
  fname = os.path.basename(output_file).split(".")[0]

  result = process_image(fname=fname,
    server_url="https://vouchervision-go-738307415303.us-central1.run.app/", 
    image_path=image_path, 
    output_dir=output_dir, 
    verbose=True, 
    engines=["gemini-3.1-flash-lite"],
    ocr_thinking_level="high",
    llm_thinking_level="high",
    prompt="SLTPvM_full_chromosome.yaml",
    auth_token=auth_token)

  # Convert to JSON string
  output_str = ordereddict_to_json(result, output_type="json")
  print(output_str)

  # Or keep it as a python dict
  output_dict = ordereddict_to_json(result, output_type="dict")
  print(output_dict)

Processing Images from URLs Programmatically

Use process_vouchers_urls when your images are hosted online and you want to process them by URL rather than downloading them first:

import os
from VoucherVision import process_vouchers_urls

if __name__ == '__main__':
  auth_token = os.environ.get("your_auth_token")

  # Process a single image URL
  process_vouchers_urls(
    server="https://vouchervision-go-738307415303.us-central1.run.app/",
    output_dir="./output_urls",
    image_url="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg",
    prompt="SLTPvM_full.yaml",
    llm_model="gemini-3.1-flash-lite",
    verbose=True,
    save_to_xlsx=True,
    auth_token=auth_token)

  # Process a list of image URLs from a file (txt, csv, or xlsx — one URL per line/row)
  process_vouchers_urls(
    server="https://vouchervision-go-738307415303.us-central1.run.app/",
    output_dir="./output_urls_bulk",
    url_list="./demo/txt/url_list.txt",
    prompt="SLTPvM_full.yaml",
    llm_model="gemini-3.1-flash-lite",
    verbose=False,
    save_to_xlsx=True,
    max_workers=8,
    auth_token=auth_token)

Viewing prompts from the command line if you install using PyPi

To see an overview of available prompts:

vv-prompts --server https://vouchervision-go-738307415303.us-central1.run.app/ --view --auth-token "your_auth_token"

To see the entire chosen prompt:

vv-prompts --server https://vouchervision-go-738307415303.us-central1.run.app/ --prompt "SLTPvM_full.yaml" --raw --auth-token "your_auth_token"

Running VoucherVision from the command line if you install using PyPi

Process a single image

vouchervision --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg 
  --output-dir ./output 
  --prompt SLTPvM_full_chromosome.yaml 
  --verbose 
  --save-to-xlsx
  --auth-token "your_auth_token"

Process a directory of images

vouchervision --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --directory ./demo/images 
  --output-dir ./output2 
  --prompt SLTPvM_full_chromosome.yaml 
  --verbose 
  --save-to-xlsx 
  --max-workers 4
  --auth-token "your_auth_token"

Changing OCR engine

vouchervision --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg 
  --output-dir ./output3 
  --engines "gemini-3.1-flash-lite"
  --ocr-thinking-level low
  --llm-thinking-level low
  --auth-token "your_auth_token"

ONLY produce OCR text

vouchervision --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg 
  --output-dir ./output3 
  --engines "gemini-3.1-flash-lite"
  --auth-token "your_auth_token"
  --ocr-only

Usage Guide (Options 2 & 3)

The VoucherVisionGO client provides several ways to process specimen images through the VoucherVision API. Here are the main usage patterns:

Basic Command Structure

(Don't include the '<' or '>' in the actual commands)

python VoucherVision.py --server <SERVER_URL> 
                 --output-dir <OUTPUT_DIR> 
                 --image <SINGLE_IMAGE_PATH_OR_URL> OR --directory <DIRECTORY_PATH> OR --file-list <FILE_LIST_PATH> 
                 --verbose
                 --save-to-xlsx
                 --engines <ENGINE1> <ENGINE2>
                 --prompt <PROMPT_FILE>
                 --max-workers <NUM_WORKERS>
                 --auth-token <YOUR_AUTH_TOKEN>

Required Arguments

The server url:

• --server: URL of the VoucherVision API server

Authentication:

• --auth-token: Your authentication token (obtained from the login page)

One of the following input options:

• --image: Path to a single image file or URL
• --directory: Path to a directory containing images
• --file-list: Path to a file containing a list of image paths or URLs

The path to your local output folder:

• --output-dir: Directory to save the output JSON results

Optional Arguments

• --engines: OCR engine options. Omit this to use gemini-3.1-flash-lite.
• --llm-model: Parsing model. Defaults to gemini-3.1-flash-lite.
• --ocr-thinking-level: OCR thinking level: low, medium, or high. Defaults to low.
• --llm-thinking-level: Parsing thinking level: low, medium, or high. Defaults to low.
• --prompt: Custom prompt file to use. We include a few for you to use. If you created a custom prompt, submit a pull request to add it to VoucherVisionGO or reach out and I can add it for you. (default: "SLTPvM_full.yaml")
• --verbose: Print all output to console. Turns off when processing bulk images, only available for single image calls.
• --save-to-xlsx: Save all results to an XLSX file in the output directory. Recommended over CSV to prevent Excel from auto-converting fields like dates.
• --max-workers: Maximum number of parallel workers. If you are processing 100s/1,000s of images increase this to 8, 16, or 32. Otherwise just skip this and let it use default values. (default: 4, max: 32)
• --ocr-only: Run only the OCR portion of VoucherVision. This will return the same final JSON packet, but with an empty "formatted_json" field.
• --notebook-mode: Run OCR only, skip the text label collage step, use the full image as input, and return OCR output formatted as Markdown. Useful for downstream document processing workflows.
• --skip-label-collage: Skip the text label collage pre-processing step and send the full original image directly to OCR. Use this if your images are already cropped to the label or if the collage step produces poor results for your collection.
• --gemini-api-key: (Optional) Provide your own Gemini API key obtained from Google AI Studio. When provided, API calls to Gemini are billed to your own Google account rather than the shared server key.
• --include-cop90: Add Copernicus GLO-90 elevation data to results. When enabled, if decimalLatitude and decimalLongitude are present in the formatted JSON, the response will include a supplemental COP90 elevation value (in meters). This does not replace any verbatim elevation data from the label — it is purely supplemental.

View Available Prompts

View the prompts in a web GUI

List all prompts

First row linux/Mac, second row Windows

curl -H "Authorization: Bearer your_auth_token" "https://vouchervision-go-738307415303.us-central1.run.app/prompts?format=text"
(curl -H "Authorization: Bearer your_auth_token" "https://vouchervision-go-738307415303.us-central1.run.app/prompts?format=text").Content

View a specific prompt

curl -H "Authorization: Bearer your_auth_token" "https://vouchervision-go-738307415303.us-central1.run.app/prompts?prompt=SLTPvM_full.yaml&format=text"
(curl -H "Authorization: Bearer your_auth_token" "https://vouchervision-go-738307415303.us-central1.run.app/prompts?prompt=SLTPvM_full.yaml&format=text").Content

Getting a specific prompt in JSON format (default)

curl -H "Authorization: Bearer your_auth_token" "https://vouchervision-go-738307415303.us-central1.run.app/prompts?prompt=SLTPvM_full.yaml"
(curl -H "Authorization: Bearer your_auth_token" "https://vouchervision-go-738307415303.us-central1.run.app/prompts?prompt=SLTPvM_full.yaml").Content

Example Calls

Processing a Single Local Image

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/single_image" 
  --verbose
  --auth-token "your_auth_token"

Processing an Image from URL

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "https://swbiodiversity.org/imglib/h_seinet/seinet/KHD/KHD00041/KHD00041592_lg.jpg" 
  --output-dir "./results/url_image" 
  --verbose
  --auth-token "your_auth_token"

Processing All Images in a Directory

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --directory "./demo/images" 
  --output-dir "./results/multiple_images" 
  --max-workers 4
  --auth-token "your_auth_token"

Processing Images from a CSV List

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --file-list "./demo/csv/file_list.csv" 
  --output-dir "./results/from_csv" 
  --max-workers 8
  --auth-token "your_auth_token"

Processing Images from a Text File List

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --file-list "./demo/txt/file_list.txt" 
  --output-dir "./results/from_txt" 
  --auth-token "your_auth_token"

Using a Custom Prompt

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "https://swbiodiversity.org/imglib/h_seinet/seinet/KHD/KHD00041/KHD00041592_lg.jpg" 
  --output-dir "./results/custom_prompt" 
  --prompt "SLTPvM_full_chromosome.yaml" 
  --verbose
  --auth-token "your_auth_token"

Saving Results to XLSX

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --directory "./demo/images" 
  --output-dir "./results/with_xlsx" 
  --save-to-xlsx
  --auth-token "your_auth_token"

Running in OCR-only mode

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --directory "./demo/images" 
  --output-dir "./results/ocr_only" 
  --save-to-xlsx
  --auth-token "your_auth_token"
  --ocr-only

Output

The client saves the following outputs:

• Individual JSON files for each processed image in the specified output directory.
• A consolidated XLSX file when --save-to-xlsx is used. The results sheet contains specimen data, while request_metadata contains token, thinking-token, billing, and cost totals. XLSX is strongly recommended over CSV to prevent Excel from auto-converting dates and catalog numbers.
• Terminal output with processing details if --verbose option is used.

An example of the JSON packet returned by the VVGO API

{
  "filename": "31234100396116",
  "payment_inference": "VoucherVisionGO_credits",
  "host": "VoucherVisionGO",
  "total_request_cost_usd": 0.0042,
  "ocr_info": {
    "gemini-3.1-flash-lite": {
      "ocr_text": "EASTERN KENTUCKY UNIVERSITY\nHERBARIUM\n060934\n\nKentucky\nLetcher County\nDiapensiaceae\n*Galax aphylla* auct. non L.\nAbove falls.\n\nWhitesburg Q.; Bad Branch. 1.5 miles NE\nof Eolia.\n\nR. Hannan & L. R.\nPhillippe 2022                                      May 31, 1979\n\nIK\n3 1234 10039611 6\nEastern Kentucky University Herbarium\n\n\n*Galax aphylla*\n\n",
      "cost_in": 0.00077875,
      "cost_out": 0.00062,
      "thinking_cost": 0.0015,
      "total_cost": 0.00289875,
      "rates_in": 1.25,
      "rates_out": 5.0,
      "tokens_in": 623,
      "tokens_out": 124,
      "thinking_tokens": 1000
    },
    "gemini-3.7-flash": {
      "ocr_text": "EASTERN\nKENTUCKY\nUNIVERSITY\nHERBARIUM\n060934\nINCH\nOPTIRECTILINEAR\nU.S.A.\nKentucky\nEKY\nLetcher County\nDiapensiaceae\nGalax aphylla auct. non L.\nAbove falls.\nWhitesburg Q.; Bad Branch. 1.5 miles NE\nof Eolia.\nR. Hannnan & L. R.\nPhillippe 2022\nMay. 31, 1979\nIK\n3 1234 10039611 6\nEastern Kentucky University Herbarium\n\n\nGalax aphylla\n\n",
      "cost_in": 0.0006815,
      "cost_out": 5.68e-05,
      "thinking_cost": 0.0,
      "total_cost": 0.0007383,
      "rates_in": 0.1,
      "rates_out": 0.4,
      "tokens_in": 6815,
      "tokens_out": 142,
      "thinking_tokens": 0
    }
  },
  "parsing_info": {
    "model": "gemini-3.1-flash-lite",
    "input": 2136,
    "output": 437,
    "cost_in": 0.0002136,
    "thinking_tokens": 400,
    "cost_out": 0.00017480000000000002,
    "thinking_cost": 0.0006,
    "total_cost": 0.0009884
  },
  "ocr": "\ngemini-1.5-pro OCR:\nEASTERN KENTUCKY UNIVERSITY\nHERBARIUM\n060934\n\nKentucky\nLetcher County\nDiapensiaceae\n*Galax aphylla* auct. non L.\nAbove falls.\n\nWhitesburg Q.; Bad Branch. 1.5 miles NE\nof Eolia.\n\nR. Hannan & L. R.\nPhillippe 2022                                      May 31, 1979\n\nIK\n3 1234 10039611 6\nEastern Kentucky University Herbarium\n\n\n*Galax aphylla*\n\n\ngemini-2.0-flash OCR:\nEASTERN\nKENTUCKY\nUNIVERSITY\nHERBARIUM\n060934\nINCH\nOPTIRECTILINEAR\nU.S.A.\nKentucky\nEKY\nLetcher County\nDiapensiaceae\nGalax aphylla auct. non L.\nAbove falls.\nWhitesburg Q.; Bad Branch. 1.5 miles NE\nof Eolia.\nR. Hannnan & L. R.\nPhillippe 2022\nMay. 31, 1979\nIK\n3 1234 10039611 6\nEastern Kentucky University Herbarium\n\n\nGalax aphylla\n\n",
  "formatted_json": {
    "catalogNumber": "060934",
    "scientificName": "Galax aphylla",
    "genus": "Galax",
    "specificEpithet": "aphylla",
    "scientificNameAuthorship": "auct. non L.",
    "collectedBy": "R. Hannan & L. R. Phillippe",
    "collectorNumber": "2022",
    "identifiedBy": "IK",
    "identifiedDate": "",
    "identifiedConfidence": "",
    "identifiedRemarks": "",
    "identificationHistory": "",
    "verbatimCollectionDate": "May 31, 1979",
    "collectionDate": "1979-05-31",
    "collectionDateEnd": "",
    "habitat": "Above falls.",
    "chromosomeCount": "",
    "guardCell": "",
    "specimenDescription": "",
    "cultivated": "",
    "continent": "North america",
    "country": "Usa",
    "stateProvince": "Kentucky",
    "county": "Letcher County",
    "locality": "Whitesburg Q.; Bad Branch. 1.5 miles NE of Eolia.",
    "verbatimCoordinates": "",
    "decimalLatitude": "",
    "decimalLongitude": "",
    "minimumElevationInMeters": "",
    "maximumElevationInMeters": "",
    "elevationUnits": "",
    "additionalText": "EASTERN KENTUCKY UNIVERSITY\nHERBARIUM\nEastern Kentucky University Herbarium"
  }
}

Advanced Usage

Using Different OCR Engines

Using two models for OCR

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/custom_engines" 
  --engines "gemini-3.1-flash-lite" "gemini-3.7-flash"
  --verbose
  --auth-token "your_auth_token"

Using only 1 of the best Gemini models for OCR.

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/custom_engines" 
  --engines "gemini-3.1-flash-lite"
  --verbose
  --auth-token "your_auth_token"

Using Different LLM Models

In addition to selecting OCR engines, you can specify which LLM model to use for parsing the OCR text into structured JSON data.

From the command line

# Specify a specific LLM model for processing
python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/custom_llm" 
  --llm-model "gemini-3.7-flash"
  --verbose
  --auth-token "your_auth_token"

From PyPi

import os
from VoucherVision import process_vouchers

auth_token = os.environ.get("your_auth_token")

process_vouchers(
  server="https://vouchervision-go-738307415303.us-central1.run.app/", 
  output_dir="./output", 
  prompt="SLTPvM_full.yaml", 
  image="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg", 
  llm_model="gemini-3.1-pro",
  llm_thinking_level="low",
  verbose=True, 
  save_to_xlsx=True, 
  auth_token=auth_token
)

Using Your Own Gemini API Key

By default, all API calls to Gemini are made using the shared server key provided by the University of Michigan. If you have your own Gemini API key from Google AI Studio, you can supply it so that usage is billed to your own Google account. This is useful for users with high-volume needs or who want to use their own quota.

Never put your API key directly in your code. Always load it from an environment variable or a secrets manager.

From the command line

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/own_key" 
  --gemini-api-key "your_gemini_api_key"
  --verbose
  --auth-token "your_auth_token"

From PyPi

import os
from VoucherVision import process_vouchers

auth_token = os.environ.get("your_auth_token")
gemini_api_key = os.environ.get("your_gemini_api_key")

process_vouchers(
  server="https://vouchervision-go-738307415303.us-central1.run.app/", 
  output_dir="./output", 
  prompt="SLTPvM_full.yaml", 
  image="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg", 
  verbose=True, 
  save_to_xlsx=True, 
  auth_token=auth_token,
  gemini_api_key=gemini_api_key  # Optional: use your own Gemini API key
)

Single image with your own key

import os
from VoucherVision import process_image, ordereddict_to_json, get_output_filename

auth_token = os.environ.get("your_auth_token")
gemini_api_key = os.environ.get("your_gemini_api_key")

image_path = "https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg"
output_dir = "./output"
output_file, _ = get_output_filename(image_path, output_dir)
fname = os.path.basename(output_file).split(".")[0]

result = process_image(
  fname=fname,
  server_url="https://vouchervision-go-738307415303.us-central1.run.app/",
  image_path=image_path,
  output_dir=output_dir,
  verbose=True,
  engines=["gemini-3.1-flash-lite"],
  prompt="SLTPvM_full.yaml",
  auth_token=auth_token,
  gemini_api_key=gemini_api_key  # Optional
)

Using Notebook Mode

Notebook mode runs OCR only (no JSON parsing), skips the text label collage pre-processing step, sends the full original image to the OCR model, and returns the OCR output formatted as Markdown. This is useful when you want clean, structured text output for downstream document processing, note-taking tools, or when you need to inspect raw OCR quality.

When notebook mode is enabled, the formatted_json field in the response will be empty and the OCR result will appear in the formatted_md field as Markdown. A .md file will also be saved alongside the .json file in your output directory.

From the command line

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/notebook" 
  --notebook-mode
  --verbose
  --auth-token "your_auth_token"

From PyPi

import os
from VoucherVision import process_vouchers

auth_token = os.environ.get("your_auth_token")

process_vouchers(
  server="https://vouchervision-go-738307415303.us-central1.run.app/", 
  output_dir="./output_notebook", 
  image="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg", 
  notebook_mode=True,  # Returns OCR as Markdown, skips JSON parsing
  verbose=True, 
  auth_token=auth_token
)

Skipping the Label Collage Step

By default, the server runs a pre-processing step that detects and crops label regions from the image before passing them to OCR (the "text collage"). This improves accuracy for herbarium sheet images where the specimen and labels share the same image.

Use --skip-label-collage to bypass this step and send the full original image directly to OCR. This is useful when:

• Your images are already tightly cropped to the label
• The collage detection is producing poor results for your collection type
• You want faster processing and your images are clean single-label shots

From the command line

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/no_collage" 
  --skip-label-collage
  --verbose
  --auth-token "your_auth_token"

From PyPi

import os
from VoucherVision import process_vouchers

auth_token = os.environ.get("your_auth_token")

process_vouchers(
  server="https://vouchervision-go-738307415303.us-central1.run.app/", 
  output_dir="./output_no_collage", 
  image="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg", 
  skip_label_collage=True,  # Skip collage, use full image
  verbose=True, 
  save_to_xlsx=True,
  auth_token=auth_token
)

Using World Flora Online (WFO) Validation

The --include-wfo flag enables taxonomic validation against the World Flora Online database. This feature validates plant names and provides additional taxonomic information in the results.

When WFO validation is enabled, the results will include a WFO_info field containing taxonomic validation data and any corrections or additional information from the World Flora Online database.

From the Command Line (Options 2 & 3)

Single image with WFO validation:

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg" 
  --output-dir "./results/with_wfo" 
  --include-wfo 
  --verbose
  --auth-token "your_auth_token"

Directory processing with WFO validation:

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --directory "./demo/images" 
  --output-dir "./results/bulk_wfo" 
  --include-wfo 
  --max-workers 4
  --auth-token "your_auth_token"

Combining with custom prompt and LLM model:

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image "https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg" 
  --output-dir "./results/advanced_wfo" 
  --prompt "SLTPvM_full_chromosome.yaml" 
  --llm-model "gemini-3.1-pro"
  --include-wfo 
  --verbose
  --auth-token "your_auth_token"

From PyPi (Option 1)

Command line with PyPi installation:

vouchervision --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --image https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg 
  --output-dir ./output 
  --include-wfo 
  --verbose 
  --auth-token "your_auth_token"

Programmatic usage with PyPi:

import os
from VoucherVision import process_vouchers

auth_token = os.environ.get("your_auth_token")

process_vouchers(
  server="https://vouchervision-go-738307415303.us-central1.run.app/", 
  output_dir="./output", 
  prompt="SLTPvM_full.yaml", 
  image="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg", 
  llm_model="gemini-3.1-pro",
  include_wfo=True,  # Enable WFO validation
  verbose=True, 
  save_to_xlsx=True, 
  auth_token=auth_token
)

Single image processing with WFO:

import os
from VoucherVision import process_image, ordereddict_to_json, get_output_filename

auth_token = os.environ.get("your_auth_token")

image_path = "https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg"
output_dir = "./output"
output_file, _ = get_output_filename(image_path, output_dir)  # returns (json_path, md_path)
fname = os.path.basename(output_file).split(".")[0]

result = process_image(
  fname=fname,
  server_url="https://vouchervision-go-738307415303.us-central1.run.app/", 
  image_path=image_path, 
  output_dir=output_dir, 
  verbose=True, 
  engines=["gemini-3.1-flash-lite"],
  prompt="SLTPvM_full.yaml",
  include_wfo=True,  # Enable WFO validation
  auth_token=auth_token
)

# The result will now include WFO validation data in the WFO_info field
output_dict = ordereddict_to_json(result, output_type="dict")
print("WFO Validation Results:", output_dict.get('WFO_info', 'No WFO data'))

API Usage

Using form data:

curl -X POST "https://vouchervision-go-738307415303.us-central1.run.app/process" \
  -H "Authorization: Bearer your_auth_token" \
  -F "file=@image.jpg" \
  -F "include_wfo=true"

Using URL processing:

curl -X POST "https://vouchervision-go-738307415303.us-central1.run.app/process-url" \
  -H "Authorization: Bearer your_auth_token" \
  -H "Content-Type: application/json" \
  -d '{
    "image_url": "https://example.com/specimen.jpg",
    "include_wfo": true,
    "prompt": "SLTPvM_full.yaml"
  }'

Using Copernicus GLO-90 Elevation Data

The --include-cop90 flag enriches results with elevation data from the Copernicus GLO-90 Digital Surface Model (90 m resolution), derived from the TanDEM-X mission (DLR/Airbus) and distributed by ESA via OpenTopography.

When enabled, if decimalLatitude and decimalLongitude are present in the formatted JSON, the response will include the COP90 elevation (in meters) for those coordinates. This is supplemental data — it does not replace any verbatim elevation transcribed from the specimen label.

Contains modified Copernicus data (2011–2015). © DLR e.V. 2010–2014 and © Airbus Defence and Space GmbH 2014–2018, provided under Copernicus by the European Union and ESA.

From the Command Line (Options 2 & 3)

Single image with COP90 elevation:

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/
  --image "./demo/images/MICH_16205594_Poaceae_Jouvea_pilosa.jpg"
  --output-dir "./results/with_cop90"
  --include-cop90
  --verbose
  --auth-token "your_auth_token"

Directory processing with COP90 elevation:

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/
  --directory "./demo/images"
  --output-dir "./results/bulk_cop90"
  --include-cop90
  --max-workers 4
  --auth-token "your_auth_token"

Combining with WFO validation and COP90 elevation:

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/
  --image "https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg"
  --output-dir "./results/wfo_cop90"
  --include-wfo
  --include-cop90
  --verbose
  --auth-token "your_auth_token"

From PyPi (Option 1)

Command line with PyPi installation:

vouchervision --server https://vouchervision-go-738307415303.us-central1.run.app/
  --image https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg
  --output-dir ./output
  --include-cop90
  --verbose
  --auth-token "your_auth_token"

Programmatic usage with PyPi:

import os
from VoucherVision import process_vouchers

auth_token = os.environ.get("your_auth_token")

process_vouchers(
  server="https://vouchervision-go-738307415303.us-central1.run.app/",
  output_dir="./output",
  prompt="SLTPvM_full.yaml",
  image="https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg",
  include_cop90=True,  # Add COP90 elevation data
  verbose=True,
  save_to_xlsx=True,
  auth_token=auth_token
)

Single image processing with COP90:

import os
from VoucherVision import process_image, ordereddict_to_json, get_output_filename

auth_token = os.environ.get("your_auth_token")

image_path = "https://swbiodiversity.org/imglib/seinet/sernec/EKY/31234100396/31234100396116.jpg"
output_dir = "./output"
output_file, _ = get_output_filename(image_path, output_dir)
fname = os.path.basename(output_file).split(".")[0]

result = process_image(
  fname=fname,
  server_url="https://vouchervision-go-738307415303.us-central1.run.app/",
  image_path=image_path,
  output_dir=output_dir,
  verbose=True,
  engines=["gemini-3.1-flash-lite"],
  prompt="SLTPvM_full.yaml",
  include_cop90=True,  # Add COP90 elevation data
  auth_token=auth_token
)

output_dict = ordereddict_to_json(result, output_type="dict")
print("COP90 Elevation:", output_dict.get('COP90_info', 'No COP90 data'))

API Usage

Using form data:

curl -X POST "https://vouchervision-go-738307415303.us-central1.run.app/process" \
  -H "Authorization: Bearer your_auth_token" \
  -F "file=@image.jpg" \
  -F "include_cop90=true"

Using URL processing:

curl -X POST "https://vouchervision-go-738307415303.us-central1.run.app/process-url" \
  -H "Authorization: Bearer your_auth_token" \
  -H "Content-Type: application/json" \
  -d '{
    "image_url": "https://example.com/specimen.jpg",
    "include_cop90": true,
    "prompt": "SLTPvM_full.yaml"
  }'

Processing Large Batches with Parallel Workers

For large datasets, you can adjust the number of parallel workers:

python VoucherVision.py --server https://vouchervision-go-738307415303.us-central1.run.app/ 
  --file-list "./demo/txt/file_list32.txt" 
  --output-dir "./results/parallel" 
  --max-workers 32 
  --save-to-xlsx
  --auth-token "your_auth_token"

Contributing

If you encounter any issues or have suggestions for improvements, please open an issue in the main repository VoucherVisionGO.