Add Azure AI MAI image generation support

[emerzon]

Summary

• add Azure AI catalog entries for MAI-Image-2 and MAI-Image-2e with verified pricing, modalities, and support metadata
• route Azure AI MAI image generation through /mai/v1/images/generations with MAI-specific request transformation for size / width / height
• extend Azure AI image cost calculation for token-priced image models and add focused MAI metadata + transport tests

Testing

• uv run black litellm/llms/azure_ai/image_generation/mai_transformation.py litellm/llms/azure_ai/image_generation/__init__.py litellm/images/main.py litellm/types/llms/openai.py tests/test_litellm/llms/azure_ai/image_generation/test_azure_ai_mai_image_generation_metadata.py
• uv run pytest tests/test_litellm/llms/azure_ai/image_generation/test_azure_ai_mai_image_generation_metadata.py tests/test_litellm/images/test_image_generation_extra_headers.py -q
• live litellm.image_generation() calls against both Azure AI MAI deployments (MAI-Image-2, MAI-Image-2e) via the Foundry MAI endpoint

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing emerzon:egomes/azure-mai-image-models (e4172c2) with main (934ecdc)}

[codecov]

Codecov Report

❌ Patch coverage is 71.90083% with 34 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...ms/azure_ai/image_generation/mai_transformation.py	70.88%	23 Missing ⚠️
litellm/llms/azure_ai/image_generation/__init__.py	60.00%	8 Missing ⚠️
.../llms/azure_ai/image_generation/cost_calculator.py	86.36%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!

[greptile-apps]

Greptile Summary

Adds Azure AI MAI image generation support (MAI-Image-2, MAI-Image-2e) by routing through /mai/v1/images/generations with a dedicated transformation class, a metadata-driven config selector driven by supports_mai_endpoint in the model cost JSON, and an extended cost calculator that prefers token-based usage over flat per-image pricing. Previous review comments around normalisation consistency, output_format hardcoding, and _parse_size error messages have all been addressed.

Confidence Score: 5/5

Safe to merge — no functional bugs found, all prior review concerns resolved.

No P0 or P1 findings. The routing logic, cost calculation, dimension validation, and header handling are all correct. The _supports_mai_endpoint two-step lookup (get_model_info → model_cost.get) is necessary because ModelInfoBase TypedDict doesn't expose custom JSON fields like supports_mai_endpoint. Tests cover metadata, pricing math, config selection, request mapping, invalid input, and end-to-end HTTP routing without real network calls.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/azure_ai/image_generation/mai_transformation.py	New MAI transformation class with URL construction, request/response mapping, and dimension validation — clean implementation with good error handling
litellm/llms/azure_ai/image_generation/init.py	Refactored to metadata-driven routing via _supports_mai_endpoint using get_model_info + model_cost lookup; normalization is now consistent via _normalize_model_name
litellm/llms/azure_ai/image_generation/cost_calculator.py	Extended to support token-based pricing with usage-first calculation that falls back to per-image pricing; math is correct for both MAI models
litellm/images/main.py	Added MAI routing branch that dispatches to llm_http_handler when config is AzureFoundryMAIImageGenerationConfig; both extra_headers and headers kwargs are correctly forwarded
tests/test_litellm/llms/azure_ai/image_generation/test_azure_ai_mai_image_generation_metadata.py	New test file covering model metadata, cost calculation, config selection, request mapping, and HTTP routing — all mock-only, no real network calls

_{Reviews (2): Last reviewed commit: "Break Azure AI image cost import cycle" | Re-trigger Greptile}

[Copilot]

Pull request overview

Adds Azure AI Foundry “MAI-Image-2” / “MAI-Image-2e” image generation support to LiteLLM, including catalog metadata, request routing to the MAI-specific endpoint, and token-based image cost calculation.

Changes:

• Add MAI model entries (pricing + modalities + endpoints) to the model cost maps and validate via new metadata/cost tests.
• Introduce an Azure Foundry MAI image-generation provider config that transforms size → width/height and targets /mai/v1/images/generations.
• Extend Azure AI image cost calculation to prefer usage/token-based computation when usage metadata is present.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/test_litellm/llms/azure_ai/image_generation/test_azure_ai_mai_image_generation_metadata.py	Adds tests for MAI metadata, cost calculation, request mapping, and routing behavior.
model_prices_and_context_window.json	Registers azure_ai/MAI-Image-2 and azure_ai/MAI-Image-2e with token-based output pricing metadata.
litellm/model_prices_and_context_window_backup.json	Mirrors the MAI entries for the packaged/local cost map used in tests.
litellm/llms/azure_ai/image_generation/mai_transformation.py	New MAI request/response transformation + URL builder for the Foundry MAI endpoint.
litellm/llms/azure_ai/image_generation/cost_calculator.py	Uses usage-based cost calculation when available, otherwise falls back to per-image pricing.
litellm/llms/azure_ai/image_generation/init.py	Routes MAI model names to the new MAI image generation config.
litellm/images/main.py	Routes Azure AI MAI image generation through llm_http_handler.image_generation_handler.
litellm/types/llms/openai.py	Extends the image generation optional-param typing to include width/height.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[emerzon]

Closing due to partially superseed by #29688