The text provider now includes reference_images alongside inputs when
building prompts. Image files are sent as base64 data URLs via
ImageURLChunk for actual multimodal vision support, replacing the
previous [Attached image: ...] placeholder text.
Replace singular reference_image field with reference_images list to
support an arbitrary number of reference images. Map them to the correct
BFL API parameter names based on model family:
- flux-2-*: input_image, input_image_2, ..., input_image_8
- flux-kontext-*: input_image, input_image_2, ..., input_image_4
- flux 1.x: image_prompt (single)
BREAKING CHANGE: reference_image config field renamed to reference_images (list).
Implement bulkgen/providers/bfl.py with a fully async httpx-based client
that supports all current and future BFL models (including flux-2-*).
Remove the blackforest dependency and simplify the image provider by
eliminating the asyncio.to_thread wrapper.
- Replace pyright with basedpyright in devenv.nix (custom hook)
- Add basedpyright to devenv packages
- Fix all basedpyright warnings: add DiGraph[str] type args, annotate
class attributes, narrow SyncResponse, handle unused call results,
suppress unavoidable Any from yaml.safe_load and untyped blackforest
- Replace pydantic-settings[yaml] with direct pyyaml dependency
- Update CLAUDE.md to reflect basedpyright and dependency changes
- Uses native async API (chat.complete_async)
- Appends text input file contents to prompt with headers
- Notes image inputs as attached references
- Writes raw LLM response directly to output file
- Create bulkgen/ package with __init__.py
- Create bulkgen/providers/ subpackage with __init__.py
- Add mistralai>=1.0.0 to dependencies
- Add [project.scripts] entry point and [build-system] with hatchling
- Add .bulkgen.state.yaml to .gitignore
