feat: add gpt-image-1-mini model for OpenAI image generation

[Kamilbenkirane]

Summary

Adds the new GPT Image 1 Mini model (gpt-image-1-mini) to the OpenAI image generation provider.

Changes

• Added gpt-image-1-mini model with appropriate parameter constraints:
  • Streaming: Supported
  • Partial Images: 0-3
  • Aspect Ratio: 1024x1024, 1024x1536, 1536x1024, auto
  • Quality: low, medium, high, auto

Model Details

• Model ID: gpt-image-1-mini
• Display Name: GPT Image 1 Mini
• Provider: OpenAI
• Status: Cost-efficient version of GPT-Image-1

Validation

• Model validated in playground
• Parameter constraints verified against OpenAI API documentation

[claude]

Code Review: Add gpt-image-1-mini Model

Thank you for this contribution! Overall, the PR follows the existing patterns well. Here's my detailed review:

✅ Code Quality & Best Practices

Positives:

• Code follows the established pattern for model definitions in this codebase
• Proper use of type hints and data structures (Model, Choice, Range)
• Consistent formatting and style with existing models
• Clear and descriptive display name

Observations:

• The model definition is nearly identical to gpt-image-1 (lines 30-43). While this appears intentional for a "mini" variant, consider if these models should share any configuration to reduce duplication in the future.

⚠️ Potential Issues

1. Streaming Support Validation (CRITICAL)

The model is marked with streaming=True, but the streaming implementation in client.py:117-119 has a hardcoded check:

if self.model.id != "gpt-image-1":
    msg = f"Streaming not supported for model '{self.model.id}'. Only 'gpt-image-1' supports streaming."
    raise ValueError(msg)

Issue: If gpt-image-1-mini supports streaming, this check needs to be updated to:

if self.model.id not in ("gpt-image-1", "gpt-image-1-mini"):

Recommendation: Please verify with OpenAI's documentation whether gpt-image-1-mini actually supports streaming. If it does, update the validation logic in packages/image-generation/src/celeste_image_generation/providers/openai/client.py:117-119.

2. Aspect Ratio Ordering Inconsistency

• gpt-image-1: ["1024x1024", "1536x1024", "1024x1536", "auto"]
• gpt-image-1-mini: ["1024x1024", "1024x1536", "1536x1024", "auto"]

The order of aspect ratios differs between the two models (notice 1024x1536 and 1536x1024 are swapped). Unless this reflects actual API differences, consider using consistent ordering for maintainability.

🧪 Test Coverage

Missing:

• No unit tests added for the new model
• No integration tests for gpt-image-1-mini

Recommendations:

1. Add integration test in packages/image-generation/tests/integration_tests/test_image_generation/test_generate.py:

   (Provider.OPENAI, "gpt-image-1-mini", {"aspect_ratio": "1024x1024"}),

2. If streaming is supported, add streaming tests
3. Consider adding tests to verify parameter constraints are properly enforced

🔒 Security Concerns

No security issues identified. The changes are purely declarative configuration.

⚡ Performance Considerations

No performance concerns. As a "mini" model, this is likely intended to be more cost-efficient than gpt-image-1, which is appropriate for the use case.

📋 Action Items

Before merging:

1. Verify streaming support - Confirm with OpenAI docs whether gpt-image-1-mini supports streaming
2. Update streaming validation - If streaming is supported, update the check in client.py:117-119
3. Fix aspect ratio ordering - Align with gpt-image-1 unless API differs
4. Add tests - At minimum, add the model to integration tests

Nice to have:

• Add a comment or documentation explaining the differences between gpt-image-1 and gpt-image-1-mini
• Consider extracting shared configuration if more GPT Image variants are expected

📝 Summary

The code quality is good and follows existing patterns. The main concern is the streaming validation hardcoded to only allow gpt-image-1, which will cause runtime errors if users try to stream with gpt-image-1-mini. Please verify the streaming capability and update the validation accordingly.

Great work overall! 🎉

[claude]

PR Review: Add GPT Image 1 Mini Model

Summary

This PR adds support for the new gpt-image-1-mini model to the OpenAI image generation provider and removes a redundant streaming validation check. Overall, the implementation is solid and well-structured. The changes are minimal, focused, and follow existing patterns in the codebase.

---

✅ Strengths

1. Excellent Code Consistency: The new model definition follows the exact same pattern as the existing gpt-image-1 model, making it easy to understand and maintain.

2. Smart Refactoring: Removing the hardcoded model ID check in client.py:114-117 is a good improvement. The base Client.stream() method already validates self.model.streaming at line 91-92 in src/celeste/client.py, making this check redundant.

3. Clear Documentation: The PR description and commit messages are well-written and explain the changes clearly.

4. Minimal Surface Area: The changes touch only the necessary files, reducing risk.

---

🔍 Code Quality Observations

Model Definition (models.py:44-58)

The new model is correctly configured with:

• ✅ Appropriate streaming support
• ✅ Correct parameter constraints matching gpt-image-1
• ✅ Proper display name

Aspect Ratio Constraint

I notice the gpt-image-1-mini has slightly different aspect ratios than gpt-image-1:

• gpt-image-1: ["1024x1024", "1536x1024", "1024x1536", "auto"]
• gpt-image-1-mini: ["1024x1024", "1024x1536", "1536x1024", "auto"]

The order is different (1024x1536 and 1536x1024 are swapped). This is likely intentional but worth confirming:

• Question: Is this order difference intentional or should it match gpt-image-1 for consistency?

---

🧪 Test Coverage

Concern: No tests were added or modified for the new model.

Recommendations:

1. Add the new model to the integration test in packages/image-generation/tests/integration_tests/test_image_generation/test_generate.py:

   (Provider.OPENAI, "gpt-image-1-mini", {"aspect_ratio": "1024x1024"}),

2. If streaming is a key feature, consider adding a streaming test (though this may already be covered by existing tests).

While the existing test suite will likely catch major issues, explicitly testing the new model would:

• Validate the model works correctly with the API
• Prevent regressions
• Document expected behavior

---

🔒 Security Considerations

✅ No security concerns identified. The changes:

• Don't introduce new input validation requirements
• Don't modify authentication/authorization logic
• Don't change error handling behavior
• Leverage existing validation through parameter constraints

---

⚡ Performance Considerations

✅ No performance concerns. The changes:

• Don't add new loops or complex operations
• Don't change request/response handling
• The removed validation actually marginally improves performance by eliminating a redundant check

---

🐛 Potential Issues

Minor Documentation Inconsistency (parameters.py:77):
The docstring was updated from:

"""Map quality parameter for DALL-E 3 and gpt-image-1."""

to:

"""Map quality parameter"""

While the method docstring (lines 87-93) still lists all models, the class docstring is now less informative. Consider:

"""Map quality parameter for DALL-E 3, gpt-image-1, and gpt-image-1-mini."""

---

📋 Recommendations

Must Have:

• Verify the aspect ratio order for gpt-image-1-mini against OpenAI's API documentation

Should Have:

• Add integration test for the new model
• Update QualityMapper class docstring to list all supported models

Nice to Have:

• If this is a cost-efficient model (as mentioned in the PR description), consider documenting this in a comment or the model's metadata

---

✨ Final Verdict

Approved with minor suggestions. The code is production-ready, but adding tests would significantly improve confidence in the implementation. The refactoring of the streaming validation is a welcome improvement.

Great work maintaining consistency with existing patterns! 🚀

[claude]

PR Review: Add gpt-image-1-mini Model

✅ Summary

This PR successfully adds the gpt-image-1-mini model to the OpenAI image generation provider. The implementation follows existing patterns well and improves code maintainability by removing an overly restrictive streaming check.

---

🎯 Code Quality & Best Practices

Strengths:

• ✅ Consistent with existing model definitions (follows the same pattern as gpt-image-1)
• ✅ Properly uses type-safe parameter constraints with Range and Choice
• ✅ Good use of the unified API design pattern
• ✅ Clean removal of hardcoded model check in client.py:114-117 - this is a positive change that makes the code more maintainable

Observations:

1. Aspect Ratio Order Inconsistency (models.py:52)

   • gpt-image-1: ["1024x1024", "1536x1024", "1024x1536", "auto"]
   • gpt-image-1-mini: ["1024x1024", "1024x1536", "1536x1024", "auto"]

   The order differs between the two models. While functionally this doesn't matter (both are valid Choice options), consistency would be better for maintainability. Consider matching the order used in gpt-image-1.

2. Documentation Updates (parameters.py:77, 89-92)

   • The docstring updates correctly reflect that streaming is no longer limited to one model
   • Good job removing outdated "gpt-image-1 only" comments
   • The quality mapper documentation is comprehensive and clear

---

🐛 Potential Bugs or Issues

None identified. The code changes are safe:

• The removal of the streaming model check is actually an improvement - it was overly restrictive and would have required maintenance for each new streaming-capable model
• The model now properly declares streaming=True, which the system will respect
• Parameter constraints are properly defined and will be validated by the framework

---

⚡ Performance Considerations

No concerns. The changes are purely declarative (model metadata) and don't introduce any performance overhead.

---

🔒 Security Concerns

None identified. The changes:

• Don't introduce any new attack vectors
• Properly validate parameters through the constraint system
• Don't modify authentication or authorization logic

---

🧪 Test Coverage

Observation: No tests were added for the new model. Consider:

1. Integration Test Addition - The file packages/image-generation/tests/integration_tests/test_image_generation/test_generate.py has a parametrized test that includes other OpenAI models:

   @pytest.mark.parametrize(
       ("provider", "model", "parameters"),
       [
           (Provider.OPENAI, "dall-e-2", {}),
           # Consider adding:
           # (Provider.OPENAI, "gpt-image-1-mini", {"aspect_ratio": "1024x1024"}),
       ],
   )

2. Streaming Test - Since this model supports streaming (streaming=True), consider adding a streaming test if one doesn't already exist for OpenAI image generation models.

However, given that:

• The PR description mentions "Model validated in playground"
• The implementation exactly mirrors gpt-image-1
• The framework will enforce parameter constraints automatically

The lack of automated tests is acceptable for this change, though adding them would increase confidence in future refactoring.

---

📝 Recommendations

Minor improvements (optional):

1. Consider standardizing aspect ratio order in models.py:52 to match line 37:

   options=["1024x1024", "1536x1024", "1024x1536", "auto"]

2. Consider adding the model to integration tests for better coverage

Otherwise, this PR is well-structured and ready to merge! The code follows existing patterns, improves maintainability by removing the hardcoded streaming check, and properly implements the new model with appropriate constraints.

---

🎉 Overall Assessment

Approve with minor suggestions. This is a clean, straightforward addition that improves the codebase. The removal of the streaming model check is actually a positive refactor that makes the code more maintainable for future model additions.

Great work! 🚀