I agree this is a good idea, should be fairly straightforward in v2.

I'll just repeat some of my comment from #1361, (with some updates):

I have no problems with this, but I think that we should give the model-config-key and field-info-key different names — rather than using title_generator for both, we should use class_title_generator or similar for BaseModels/dataclasses/etc., and field_title_generator for field titles, otherwise I think it will lead to confusion.

I'll also note that this will require some care around the use of something like field_title_priority with similar behavior to alias_priority, and perhaps similar for model_title_priority.

Is there any workaround for this today? I've tried using __init_subclass__ to give all my models a custom title generated from cls.__module__ and cls.__name__.

Using just the class name, like the default behavior does, doesn't work in large projects. If you use e.g. FastAPI to generate an OpenAPI spec (which uses JSON schema) then we can't generate a reasonable OpenAPI client library from the OpenAPI spec. The reason is that we get a huge flat namespace of all model names, which we can't map back into e.g. namespaces. In other words the current approach is lossy.

In v1 the following workaround seems to work:

class _Base(pydantic.BaseModel):
    class Config:
        @staticmethod
        def schema_extra(schema: dict[str, Any], model: type['_Base']) -> None:
            schema['title'] = f'{model.__module__}.{model.__name__}'

Thinking about this some more it's not actually the title I want to control over but the ID generated as the key in the components map. The title is just for human consumption, I want to preserve the model's module name in the generate OpenAPI spec so I can generate an OpenAPI client with a similar structure.

Is there any news for this title generation feature?

When creating model json schemas, the current default title generation often does not play nicely when using aliases or alias generators. Consider the following toy example:

from pydantic import BaseModel, ConfigDict
from pydantic.alias_generators import to_camel


class MyModel(BaseModel):
    my_field: str
    my_other_field: int

    model_config = ConfigDict(alias_generator=to_camel)

When calling MyModel.model_json_schema(), the fields are converted to camel case as expected, but the titles come back as Myfield and Myotherfield. My guess is that the title generation in pydantic.json_schema.GenerateJsonSchema.get_title_from_name assumes a certain naming style to generate sensible titles.

As a workaround I am currently using a custom GenerateJsonSchema class that overrides get_title_from_name, which is then passed to model_json_schema(schema_generator=...) to better suit my case.

Having a title generator on the model config would be a convenient solution to this problem. The definitions for both the alias generator and title generator would be closer to each other, and it is more obvious what kind of values the title generator should mostly expect.