LiteLLM cannot handle image responses from TogetherAI's text2image model Stabeldiffusion due to an additional 'ID' field being sent from togetherAI in the response

[JosefAschauer]

TogetherAI sends an additional 'ID field':

litellm_1 | ) litellm_1 | ^ litellm_1 | File "/usr/lib/python3.13/site-packages/litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py", line 278, in convert_to_image_response litellm_1 | model_response_object = ImageResponse(**model_response_dict) litellm_1 | TypeError: ImageResponse.init() got an unexpected keyword argument 'id' litellm_1 | litellm_1 | litellm_1 | received_args={'response_object': {'created': None, 'data': [{'b64_json': '/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAgGBgcGBQgHBwcJCQgKDBQNDAsLDBkSEw8UHRofHh0aHBwgJC4nICIsIxwcKDcpLDAxNDQ0Hyc5PTgyPC4zNDL/2wBDAQkJCQwLDBgNDRgyIRwhMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjIyMjL/wAARCAQABAAD......

model's config:

• model_name: together_ai/stable-diffusion-xl litellm_params: model: "together_ai/stabilityai/stable-diffusion-xl-base-1.0" api_base: "https://api.together.xyz/v1" api_key: tgp_v1_-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx response_format: "base64" # Ensure base64 output model_info: mode: image_generation

Why This Happens Together AI Response Format: Together AI’s /v1/images/generations endpoint returns a JSON object like {"created": null, "data": [{"b64_json": "..."}], "id": "..."}. The 'id' field is part of their response but not part of the standard OpenAI image generation response spec that litellm expects ({"created": ..., "data": [{"b64_json": "..."}]}). litellm Bug: The ImageResponse class in litellm (at convert_dict_to_response.py:278) isn’t designed to handle extra fields like 'id', causing a TypeError. This is a compatibility issue between Together AI’s response and litellm’s parsing logic.