refactor: let Artifact validator handle base64 decoding across all providers

src/celeste/artifacts.py

@@ -3,7 +3,7 @@
 import base64
 from typing import Any
 
-from pydantic import BaseModel, Field, field_serializer
+from pydantic import BaseModel, Field, field_serializer, field_validator
 
 from celeste.mime_types import (
     AudioMimeType,
@@ -30,6 +30,14 @@ class Artifact(BaseModel):
     mime_type: MimeType | None = None
     metadata: dict[str, Any] = Field(default_factory=dict)
 
+    @field_validator("data", mode="before")
+    @classmethod
+    def decode_base64_data(cls, v: bytes | str | None) -> bytes | None:
+        """Decode base64 string to bytes when constructing from JSON."""
+        if isinstance(v, str):
+            return base64.b64decode(v)
+        return v
+
     @field_serializer("data", when_used="json")
     def serialize_data(self, value: bytes | None) -> str | None:
         """Serialize bytes as base64 string for JSON compatibility."""

src/celeste/modalities/audio/providers/google/client.py

@@ -1,6 +1,5 @@
 """Google audio client."""
 
-import base64
 from typing import Any, Unpack
 
 from celeste.artifacts import AudioArtifact
@@ -62,12 +61,11 @@ def _parse_content(
     ) -> AudioArtifact:
         """Extract audio bytes from response."""
         audio_b64 = super()._parse_content(response_data)
-        audio_bytes = base64.b64decode(audio_b64)
 
         output_format = parameters.get(AudioParameter.OUTPUT_FORMAT)
         mime_type = AudioMimeType(output_format) if output_format else AudioMimeType.MP3
 
-        return AudioArtifact(data=audio_bytes, mime_type=mime_type)
+        return AudioArtifact(data=audio_b64, mime_type=mime_type)
 
     def _parse_finish_reason(self, response_data: dict[str, Any]) -> AudioFinishReason:
         """Parse finish reason from response."""

src/celeste/modalities/images/providers/byteplus/client.py

@@ -1,6 +1,5 @@
 """BytePlus images client."""
 
-import base64
 from typing import Any, Unpack
 
 from celeste.artifacts import ImageArtifact
@@ -80,8 +79,7 @@ def _parse_chunk(self, event_data: dict[str, Any]) -> ImageChunk | None:
         if content_type == "url":
             artifact = ImageArtifact(url=content, mime_type=ImageMimeType.PNG)
         else:  # b64_json
-            image_data = base64.b64decode(content)
-            artifact = ImageArtifact(data=image_data)
+            artifact = ImageArtifact(data=content)
 
         return ImageChunk(
             content=artifact,
@@ -162,9 +160,8 @@ def _parse_content(
                 mime_type=ImageMimeType.PNG,
             )
         if image_data.get("b64_json"):
-            image_bytes = base64.b64decode(image_data["b64_json"])
             return ImageArtifact(
-                data=image_bytes,
+                data=image_data["b64_json"],
                 mime_type=ImageMimeType.PNG,
             )
 

src/celeste/modalities/images/providers/ollama/client.py

@@ -1,6 +1,5 @@
 """Ollama images client."""
 
-import base64
 from typing import Any, Unpack
 
 from celeste.artifacts import ImageArtifact
@@ -53,9 +52,8 @@ def _parse_chunk(self, event_data: dict[str, Any]) -> ImageChunk | None:
                 metadata=self._parse_chunk_metadata(event_data),
             )
 
-        image_bytes = base64.b64decode(b64_image)
         return ImageChunk(
-            content=ImageArtifact(data=image_bytes),
+            content=ImageArtifact(data=b64_image),
             finish_reason=self._parse_chunk_finish_reason(event_data),
             usage=self._parse_chunk_usage(event_data),
             metadata=self._parse_chunk_metadata(event_data),
@@ -115,8 +113,7 @@ def _parse_content(
         Ollama returns base64-encoded image in the 'image' field.
         """
         image_b64 = super()._parse_content(response_data)
-        image_bytes = base64.b64decode(image_b64)
-        return ImageArtifact(data=image_bytes)
+        return ImageArtifact(data=image_b64)
 
     def _parse_finish_reason(self, response_data: dict[str, Any]) -> ImageFinishReason:
         """Parse finish reason from response."""

src/celeste/modalities/images/providers/openai/client.py

@@ -1,6 +1,5 @@
 """OpenAI images client."""
 
-import base64
 from typing import Any, Unpack
 
 from celeste.artifacts import ImageArtifact
@@ -61,8 +60,7 @@ def _parse_chunk(self, event_data: dict[str, Any]) -> ImageChunk | None:
                 metadata={"event_data": event_data},
             )
 
-        image_data = base64.b64decode(b64_json)
-        artifact = ImageArtifact(data=image_data)
+        artifact = ImageArtifact(data=b64_json)
 
         return ImageChunk(
             content=artifact,
@@ -143,8 +141,7 @@ def _parse_content(
 
         b64_json = image_data.get("b64_json")
         if b64_json:
-            image_bytes = base64.b64decode(b64_json)
-            return ImageArtifact(data=image_bytes)
+            return ImageArtifact(data=b64_json)
 
         url = image_data.get("url")
         if url:

src/celeste/modalities/images/providers/xai/client.py

@@ -82,10 +82,7 @@ def _parse_content(
         # xAI returns either b64_json or url
         b64_json = image_data.get("b64_json")
         if b64_json:
-            import base64
-
-            image_bytes = base64.b64decode(b64_json)
-            return ImageArtifact(data=image_bytes)
+            return ImageArtifact(data=b64_json)
 
         url = image_data.get("url")
         if url:

src/celeste/modalities/videos/providers/openai/client.py

@@ -1,6 +1,5 @@
 """OpenAI videos client."""
 
-import base64
 from typing import Any, Unpack
 
 from celeste.artifacts import VideoArtifact
@@ -55,9 +54,8 @@ def _parse_content(
     ) -> VideoArtifact:
         """Parse content from response."""
         video_data_b64 = super()._parse_content(response_data)
-        video_data = base64.b64decode(video_data_b64)
         return VideoArtifact(
-            data=video_data,
+            data=video_data_b64,
             mime_type=VideoMimeType.MP4,
         )