增加MiniMax海螺模型生图功能

[feitianbubu]

官方文档: https://platform.minimaxi.com/docs/api-reference/image-generation-i2i
支持模型: image-01, image-01-live
支持格式: openai的图片生成和图片编辑接口
请求示例:

curl http://localhost:30001/v1/images/generations \
  --request POST \
  --header 'Content-Type: application/json' \
  --data '{
  "model": "image-01-live",
  "prompt": "可爱的中国小女孩在花园里玩耍"
}'

Summary by CodeRabbit

• New Features
  • Added image generation and editing support to the relay channel
  • Introduced two new image models: "image-01" and "image-01-live"
  • Enhanced image request handling with styling, dimensions, and subject reference capabilities
  • Improved response processing for image operations with metadata tracking

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

Walkthrough

This PR adds image generation and editing support to the Minimax relay channel. It introduces new DTOs for image operations, implements conversion functions from OpenAI format to Minimax format, adds image endpoint routing, and updates the adaptor with conditional image handling logic. Two new Minimax image models are registered.

Changes

Cohort / File(s)	Summary
Core DTO and Type Support dto/values.go, relay/channel/minimax/dto_image.go	Adds IntValue type with custom JSON marshalling for flexible numeric/string handling; introduces six new image DTO types (ImageGenerationRequest, ImageDataObject, ImageGenerationResponse, ImageSubjectReference, StyleObject, ImageGenerationMetadata) for structured image generation workflows.
Image Operations Implementation relay/channel/minimax/image.go	New file with four image handling functions: oaiImageToHailuoImageRequest converts OpenAI-style requests to Minimax format, imageEditFromOai adds subject references from base64 images, imageHandler processes HTTP responses with error handling and usage tracking, responseToOpenAIImage formats responses back to OpenAI format.
Channel Configuration relay/channel/minimax/constants.go	Extends ModelList with two new image models: "image-01" and "image-01-live".
Request/Response Routing relay/channel/minimax/adaptor.go, relay/channel/minimax/relay-minimax.go	Updates adaptor to handle ImagesGenerations and ImagesEdits modes: routes to image conversion functions, adds Content-Type/Accept headers, delegates to imageHandler for responses. Adds endpoint routing for image operations in GetRequestURL. Changes import alias from constant to relayconstant throughout.

Sequence Diagram(s)

sequenceDiagram
    participant Client
    participant Adaptor
    participant ImageHandler
    participant Minimax API
    participant ResponseProcessor
    
    Client->>Adaptor: POST (Images request, RelayMode=ImagesGenerations/ImagesEdits)
    
    rect rgb(230, 245, 245)
        Note over Adaptor: ConvertImageRequest
        alt Edits Mode
            Adaptor->>Adaptor: imageEditFromOai() - extract form base64 images
        else GenerationMode
            Adaptor->>Adaptor: oaiImageToHailuoImageRequest() - convert OpenAI format
        end
        Adaptor->>Adaptor: SetupRequestHeader - add Content-Type, Accept
    end
    
    Adaptor->>Minimax API: POST /v1/image_generation (converted request)
    Minimax API-->>ImageHandler: HTTP response (status + body)
    
    rect rgb(240, 250, 240)
        Note over ImageHandler: imageHandler processing
        ImageHandler->>ImageHandler: Read & validate response body
        ImageHandler->>ImageHandler: Unmarshal to ImageGenerationResponse
        ImageHandler->>ResponseProcessor: responseToOpenAIImage()
        ResponseProcessor->>ResponseProcessor: Build ImageResponse with metadata
        ResponseProcessor->>ResponseProcessor: Map URLs and base64 variants
        alt Multiple images generated
            ResponseProcessor->>ImageHandler: Update usage pricing
        end
    end
    
    ImageHandler-->>Client: JSON (OpenAI-format ImageResponse)

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~35 minutes

Possibly related PRs

• 增加MiniMax语音合成TTS支持 #2081: Modifies the same minimax channel including constants and adaptor structures, indicating parallel or dependent feature development.

Suggested reviewers

• creamlike1024
• seefs001
• Calcium-Ion

Poem

🐰 A nimble hop through image dreams,
New DTOs and JSON streams,
From OpenAI's form to Minimax's call,
The adaptor routes them through it all!
With headers set and handlers tight,
Image magic, pure delight! ✨

Pre-merge checks and finishing touches

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The PR title is in Chinese and translates to 'Add MiniMax Hailuo model image generation capability', which directly matches the main objective of adding image-generation support for MiniMax models.

✨ Finishing touches

• 📝 Generate docstrings

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (5)

dto/values.go (1)

10-26: Consider handling JSON null values explicitly.

The current implementation will fail if the JSON value is null - it will try to unmarshal as int (fail), then as string (succeed with empty string), then strconv.Atoi("") will return an error. If null is a valid input for optional fields like Seed, you may want to handle it gracefully.

🔎 Proposed fix to handle null values

 func (i *IntValue) UnmarshalJSON(b []byte) error {
+	if string(b) == "null" {
+		return nil
+	}
 	var n int
 	if err := json.Unmarshal(b, &n); err == nil {
 		*i = IntValue(n)
 		return nil
 	}

relay/channel/minimax/image.go (3)

25-28: Silently ignoring marshal/unmarshal errors may hide issues.

The Extra field processing ignores both json.Marshal and json.Unmarshal errors. While this is likely intentional for optional fields, consider logging a warning if unmarshalling fails, as it could indicate malformed input.

---

114-114: Debug logging may include large base64 payloads.

Logging the full originBody at debug level could result in very large log entries when base64-encoded images are included. Consider truncating or summarizing the response for debug purposes.

---

51-56: Consider accepting subject reference type from request parameters for flexibility.

The Type field is hardcoded to "character". While this matches the standard use case documented in the MiniMax API, accepting the type from request parameters (if available) would allow for potential future extensibility and make the implementation more flexible.

relay/channel/minimax/dto_image.go (1)

8-21: Consider using pointer types for optional numeric fields.

Width, Height, and N are non-pointer dto.IntValue types. With omitempty, Go's JSON encoder only omits the field if the value is the zero value. However, since IntValue is an int, the zero value 0 will be omitted. If 0 is a valid value the user might want to send, this could cause issues.

For fields where 0 might be meaningful or where explicit omission is important, consider using *dto.IntValue (like Seed).

🔎 Proposed change for consistency

-	Width            dto.IntValue            `json:"width,omitempty"`
-	Height           dto.IntValue            `json:"height,omitempty"`
+	Width            *dto.IntValue           `json:"width,omitempty"`
+	Height           *dto.IntValue           `json:"height,omitempty"`

📜 Review details

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 022fbce and c99ac24.

📒 Files selected for processing (6)

• dto/values.go
• relay/channel/minimax/adaptor.go
• relay/channel/minimax/constants.go
• relay/channel/minimax/dto_image.go
• relay/channel/minimax/image.go
• relay/channel/minimax/relay-minimax.go

🧰 Additional context used

🧬 Code graph analysis (5)

relay/channel/minimax/relay-minimax.go (1)

relay/constant/relay_mode.go (2)

• RelayModeImagesGenerations (14-14)
• RelayModeImagesEdits (15-15)

relay/channel/minimax/adaptor.go (2)

relay/constant/relay_mode.go (3)

• RelayModeAudioSpeech (35-35)
• RelayModeImagesEdits (15-15)
• RelayModeImagesGenerations (14-14)

types/set.go (1)

• Set (3-5)

relay/channel/minimax/dto_image.go (2)

dto/values.go (1)

• IntValue (8-8)

relay/channel/task/hailuo/models.go (2)

• SubjectReference (3-6)
• BaseResp (27-30)

dto/values.go (1)

common/json.go (2)

• Unmarshal (9-11)
• Marshal (21-23)

relay/channel/minimax/image.go (4)

relay/channel/minimax/dto_image.go (4)

• ImageGenerationRequest (8-21)
• StyleObject (28-31)
• ImageSubjectReference (23-26)
• ImageGenerationResponse (33-38)

relay/channel/task/hailuo/models.go (2)

• SubjectReference (3-6)
• BaseResp (27-30)

relay/channel/task/hailuo/constants.go (1)

• StatusSuccess (25-25)

dto/openai_image.go (1)

• ImageData (174-178)

🔇 Additional comments (9)

relay/channel/minimax/constants.go (1)

17-18: LGTM!

The new image model entries follow the existing naming convention and align with the PR objectives for MiniMax image generation support.

relay/channel/minimax/relay-minimax.go (1)

22-23: LGTM!

The new image generation and edit modes correctly route to the /v1/image_generation endpoint, following the existing switch-case pattern for other relay modes.

relay/channel/minimax/adaptor.go (3)

77-82: LGTM!

The branching logic correctly differentiates between image edit and generation modes, routing to the appropriate handler functions.

---

94-97: LGTM!

Setting explicit Content-Type and Accept headers for image modes ensures proper JSON communication with the MiniMax API.

---

124-135: LGTM!

The response routing correctly handles audio, image generation, and image edit modes, with a fallback to the OpenAI adaptor for other cases.

relay/channel/minimax/image.go (2)

60-65: Verify graceful close behavior after ReadAll.

CloseResponseBodyGracefully is called after io.ReadAll. Since ReadAll consumes the entire body, the close should work correctly, but ensure that CloseResponseBodyGracefully handles already-consumed bodies gracefully.

---

101-112: No duplicate entries are possible with MiniMax's API design.

The MiniMax image generation API uses the response_format parameter to control the response format, which can be either "url" or "base64"—not both. When response_format: "url" is used, only ImageUrls is populated; when response_format: "base64" is used, only ImageBase64 is populated. The two formats are mutually exclusive at the API level, so the code cannot produce duplicate entries. The defensive iteration over both fields is safe and reasonable for robustness, but a single field will always be empty in any given response.

Likely an incorrect or invalid review comment.

relay/channel/minimax/dto_image.go (2)

23-26: LGTM!

The ImageSubjectReference struct correctly models the subject reference with type and image file fields matching the MiniMax API requirements.

---

33-43: LGTM!

The response structures properly model the MiniMax API response, reusing BaseResp for consistent error handling and separating image data into URLs and base64 formats.