Python models (pydantic, attrs, dataclasses or custom) generator from JSON data with typing module support

attrs, dataclasses, datasets, generator, json, models, pydantic, python, typing
pip install json2python-models==0.2.1



PyPI version Build Status Coverage Status Codacy Badge


json2python-models is a Python tool that can generate Python models classes (pydantic, dataclasses, attrs) from JSON dataset.


  • Full typing module support
  • Types merging - if some field contains data of different types this will be represent as Union type
  • Fields and models names generation (unicode support included)
  • Similar models generalization
  • Handling recursive data structures (i.e family tree)
  • Detecting string serializable types (i.e. datetime or just stringify numbers)
  • Detecting fields containing string constants (Literal['foo', 'bar'])
  • Generation models as list (flat models structure) or tree (nested models)
  • Specifying when dictionaries should be processed as dict type (by default every dict is considered as some model)
  • CLI API with a lot of options

Table of Contents


Part of Path of Exile public items API

from pydantic import BaseModel, Field
from typing import List, Optional
from typing_extensions import Literal

class Tab(BaseModel):
    id_: str = Field(..., alias="id")
    public: bool
    stash_type: Literal["CurrencyStash", "NormalStash", "PremiumStash"] = Field(..., alias="stashType")
    items: List['Item']
    account_name: Optional[str] = Field(None, alias="accountName")
    last_character_name: Optional[str] = Field(None, alias="lastCharacterName")
    stash: Optional[str] = None
    league: Optional[Literal["Hardcore", "Standard"]] = None

F1 Season Results

----- Show -----

        "season": "2019",
        "round": "3",
        "DriverStandings": [
                "position": "1",
                "positionText": "1",
                "points": "68",
                "wins": "2",
                "Driver": {
                    "driverId": "hamilton",
                    "permanentNumber": "44",
                    "code": "HAM",
                    "url": "",
                    "givenName": "Lewis",
                    "familyName": "Hamilton",
                    "dateOfBirth": "1985-01-07",
                    "nationality": "British"
                "Constructors": [
                        "constructorId": "mercedes",
                        "url": "",
                        "name": "Mercedes",
                        "nationality": "German"
json2models -f pydantic -l DriverStandings - driver_standings.json
generated by json2python-models v0.2.0 at Mon May  4 17:46:30 2020
command: /opt/projects/json2python-models/venv/bin/json2models -f pydantic -s flat -l DriverStandings - driver_standings.json
from pydantic import BaseModel, Field
from typing import List
from typing_extensions import Literal

class DriverStandings(BaseModel):
    season: int
    round_: int = Field(..., alias="round")
    DriverStandings: List['DriverStanding']

class DriverStanding(BaseModel):
    position: int
    position_text: int = Field(..., alias="positionText")
    points: int
    wins: int
    driver: 'Driver' = Field(..., alias="Driver")
    constructors: List['Constructor'] = Field(..., alias="Constructors")

class Driver(BaseModel):
    driver_id: str = Field(..., alias="driverId")
    permanent_number: int = Field(..., alias="permanentNumber")
    code: str
    url: str
    given_name: str = Field(..., alias="givenName")
    family_name: str = Field(..., alias="familyName")
    date_of_birth: str = Field(..., alias="dateOfBirth")
    nationality: str

class Constructor(BaseModel):
    constructor_id: str = Field(..., alias="constructorId")
    url: str
    name: str
    nationality: Literal["Austrian", "German", "American", "British", "Italian", "French"]


----- Show -----

swagger.json from any online API (I tested file generated by drf-yasg and another one for Spotify API)

It requires a lit bit of tweaking:

  • Some fields store routes/models specs as dicts
  • There is a lot of optinal fields so we reduce merging threshold
  • Disable string literals
json2models -f dataclasses -m Swagger testing_tools/swagger.json \
    --dict-keys-fields securityDefinitions paths responses definitions properties \
    --merge percent_50 number --max-strings-literals 0
generated by json2python-models v0.2.0 at Mon May  4 18:08:09 2020
command: /opt/projects/json2python-models/json_to_models/ -s flat -f dataclasses -m Swagger testing_tools/swagger.json --max-strings-literals 0 --dict-keys-fields securityDefinitions paths responses definitions properties --merge percent_50 number
from dataclasses import dataclass, field
from json_to_models.dynamic_typing import FloatString
from typing import Any, Dict, List, Optional, Union

class Swagger:
    swagger: FloatString
    info: 'Info'
    host: str
    schemes: List[str]
    base_path: str
    consumes: List[str]
    produces: List[str]
    security_definitions: Dict[str, 'Parameter_SecurityDefinition']
    security: List['Security']
    paths: Dict[str, 'Path']
    definitions: Dict[str, 'Definition_Schema']

class Info:
    title: str
    description: str
    version: str

class Security:
    api_key: Optional[List[Any]] = field(default_factory=list)
    basic: Optional[List[Any]] = field(default_factory=list)

class Path:
    parameters: List['Parameter_SecurityDefinition']
    post: Optional['Delete_Get_Patch_Post_Put'] = None
    get: Optional['Delete_Get_Patch_Post_Put'] = None
    put: Optional['Delete_Get_Patch_Post_Put'] = None
    patch: Optional['Delete_Get_Patch_Post_Put'] = None
    delete: Optional['Delete_Get_Patch_Post_Put'] = None

class Property:
    type_: str
    format_: Optional[str] = None
    xnullable: Optional[bool] = None
    items: Optional['Item_Schema'] = None

class Property_2E:
    type_: str
    title: Optional[str] = None
    read_only: Optional[bool] = None
    max_length: Optional[int] = None
    min_length: Optional[int] = None
    items: Optional['Item'] = None
    enum: Optional[List[str]] = field(default_factory=list)
    maximum: Optional[int] = None
    minimum: Optional[int] = None
    format_: Optional[str] = None

class Item:
    title: Optional[str] = None
    type_: Optional[str] = None
    ref: Optional[str] = None
    max_length: Optional[int] = None
    min_length: Optional[int] = None

class Parameter_SecurityDefinition:
    name: Optional[str] = None
    in_: Optional[str] = None
    required: Optional[bool] = None
    schema: Optional['Item_Schema'] = None
    description: Optional[str] = None
    type_: Optional[str] = None

class Delete_Get_Patch_Post_Put:
    operation_id: str
    description: str
    parameters: List['Parameter_SecurityDefinition']
    responses: Dict[str, 'Response']
    tags: List[str]

class Item_Schema:
    ref: str

class Response:
    description: str
    schema: Optional[Union['Item_Schema', 'Definition_Schema']] = None

class Definition_Schema:
    type_: str
    required: Optional[List[str]] = field(default_factory=list)
    properties: Optional[Dict[str, Union['Property', 'Property_2E']]] = field(default_factory=dict)
    ref: Optional[str] = None


Be ware: this project supports only python3.7 and higher.

To install it, use pip:

pip install json2python-models

Or you can build it from source:

git clone
cd json2python-models
python install



For regular usage CLI tool is the best option. After you install this package you could use it as json2models <arguments> or python -m json_to_models <arguments>. I.e.:

json2models -m Car car_*.json -f attrs >


  • -h, --help - Show help message and exit

  • -m, --model - Model name and its JSON data as path or unix-like path pattern. *, ** or ? patterns symbols are supported.

    • Format: -m <Model name> [<JSON files> ...]
    • Example: -m Car audi.json reno.json or -m Car audi.json -m Car reno.json (results will be the same)
  • -l, --list - Like -m but given json file should contain list of model data (dataset). If this file contains dict with nested list than you can pass <JSON key> to lookup. Deep lookups are supported by dot-separated path. If no lookup needed pass - as <JSON key>.

    • Format: -l <Model name> <JSON key> <JSON file>
    • Example: -l Car - cars.json -l Person fetch_results.items.persons result.json
    • Note: Models names under this arguments should be unique.
  • -o, --output - Output file

    • Format: -o <FILE>
    • Example: -o
  • -f, --framework - Model framework for which python code is generated. base (default) mean no framework so code will be generated without any decorators and additional meta-data.

    • Format: -f {base, pydantic, attrs, dataclasses, custom}
    • Example: -f pydantic
    • Default: -f base
  • -s, --structure - Models composition style.

    • Format: -s {flat, nested}
    • Example: -s nested
    • Default: -s flat
  • --datetime - Enable datetime/date/time strings parsing.

    • Default: disabled
    • Warning: This can lead to 6-7 times slowdown on large datasets. Be sure that you really need this option.
  • --disable-unicode-conversion, --no-unidecode - Disable unicode conversion in field labels and class names

    • Default: enabled
  • --strings-converters - Enable generation of string types converters (i.e. IsoDatetimeString or BooleanString).

    • Default: disabled
  • --max-strings-literals - Generate Literal['foo', 'bar'] when field have less than NUMBER string constants as values.

    • Format: --max-strings-literals <NUMBER>
    • Default: 10 (generator classes could override it)
    • Example: --max-strings-literals 5 - only 5 literals will be saved and used to code generation
    • Note: There could not be more than 15 literals per field (for performance reasons)
    • Note: attrs code generator do not use Literals and just generate str fields instead
  • --merge - Merge policy settings. Possible values are:

    • Format: --merge MERGE_POLICY [MERGE_POLICY ...]
    • Possible values (MERGE_POLICY):
      • percent[_<percent>] - two models had a certain percentage of matched field names. Custom value could be i.e. percent_95.
      • number[_<number>] - two models had a certain number of matched field names.
      • exact - two models should have exact same field names to merge.
    • Example: --merge percent_95 number_20 - merge if 95% of fields are matched or 20 of fields are matched
    • Default: --merge percent_70 number_10
  • --dict-keys-regex, --dkr - List of regular expressions (Python syntax). If all keys of some dict are match one of the pattern then this dict will be marked as dict field but not nested model.

    • Format: --dkr RegEx [RegEx ...]
    • Example: --dkr node_\d+ \d+_\d+_\d+
    • Note: ^ and $ (string borders) tokens will be added automatically but you have to escape other special characters manually.
    • Optional
  • --dict-keys-fields, --dkf - List of model fields names that will be marked as dict fields

    • Format: --dkf FIELD_NAME [FIELD_NAME ...]
    • Example: --dkf "dict_data" "mapping"
    • Optional
  • --code-generator - Absolute import path to GenericModelCodeGenerator subclass.

    • Format: --code-generator CODE_GENERATOR
    • Example: -f mypackage.mymodule.DjangoModelsGenerator
    • Note: Is ignored without -f custom but is required with it.
  • --code-generator-kwargs - List of GenericModelCodeGenerator subclass arguments (for __init__ method, see docs of specific subclass). Each argument should be in following format: argument_name=value or "argument_name=value with space". Boolean values should be passed in JS style: true or false

    • Format: --code-generator-kwargs [NAME=VALUE [NAME=VALUE ...]]
    • Example: --code-generator-kwargs kwarg1=true kwarg2=10 "kwarg3=It is string with spaces"
    • Optional

One of model arguments (-m or -l) is required.

Low level API



To run tests you should clone project and run script:

git clone
cd json2python-models
python test -a '<pytest additional arguments>'

Also I would recommend you to install pytest-sugar for pretty printing test results

Test examples

You can find out some examples of usage of this project at testing_tools/real_apis/...

Each file contains functions to download data from some online API (references included at the top of file) and main function that generates and prints code. Some examples may print debug data before actual code. Downloaded data will be saved at testing_tools/real_apis/<name of example>/<dataset>.json

Built With

Test tools:


Feel free to open pull requests with new features or bug fixes. Just follow few rules:

  1. Always use some code formatter (black or PyCharm built-in)
  2. Keep code coverage above 95-98%
  3. All existing tests should be passed (including test examples from testing_tools/real_apis)
  4. Use typing module
  5. Fix codacy issues from your PR


This project is licensed under the MIT License - see the LICENSE file for details