feat: support OpenRouter image generation endpoint

[Robinnnnn]

📝 变更描述 / Description

OpenRouter 现已提供图像生成 API，端点为 POST {base}/v1/images（官方文档），与 OpenAI 风格的 /v1/images/generations 路径不同。目前 OpenRouter 渠道复用 OpenAI adaptor，图像请求会被转发到 https://openrouter.ai/api/v1/images/generations，上游返回 404。

本 PR 在 openai.Adaptor.GetRequestURL 中为 OpenRouter 渠道的 RelayModeImagesGenerations 增加特判，改为请求 {base}/v1/images。请求体（model/prompt/n/size）和响应体（created/data[].b64_json/usage）均为 OpenAI Images 兼容格式，OpenRouter 会忽略未知字段，因此现有的 ImageRequest 转换和 OpenaiImageHandler（含 usage 计费解析）无需改动。

OpenRouter now offers an image generation API at POST {base}/v1/images (docs) — a flat path, unlike OpenAI's /v1/images/generations. Since the OpenRouter channel reuses the OpenAI adaptor, image requests currently go to https://openrouter.ai/api/v1/images/generations and get a 404 upstream. This PR special-cases RelayModeImagesGenerations for OpenRouter channels in GetRequestURL. Request and response shapes are OpenAI-Images-compatible (OpenRouter silently ignores unknown fields, responses carry created / data[].b64_json / usage), so the existing ConvertImageRequest passthrough and OpenaiImageHandler billing path work unchanged.

🚀 变更类型 / Type of change

• 🐛 Bug 修复 (Bug fix)
• ✨ 新功能 (New feature)
• ⚡ 性能优化 / 重构 (Refactor)
• 📝 文档更新 (Documentation)

🔗 关联任务 / Related Issue

• 无 / None

✅ 提交前检查项 / Checklist

• 人工确认: 我已亲自整理并撰写此描述，没有直接粘贴未经处理的 AI 输出。
• 非重复提交: 我已搜索现有的 Issues 与 PRs，确认不是重复提交。
• Bug fix 说明: 不适用（新功能）。
• 变更理解: 我已理解这些更改的工作原理及可能影响。
• 范围聚焦: 本 PR 未包含任何与当前任务无关的代码改动。
• 本地验证: 已在本地运行并通过测试与端到端手动验证（见下）。
• 安全合规: 代码中无敏感凭据，且符合项目代码规范。

📸 运行证明 / Proof of Work

1. 路径验证 / Endpoint verification — OpenRouter 上游确认扁平路径存在，/generations 不存在：

$ curl -s -o /dev/null -w "%{http_code}" -X POST https://openrouter.ai/api/v1/images -d '{...}'
401   # 需要鉴权 → 路径存在
$ curl -s -o /dev/null -w "%{http_code}" -X POST https://openrouter.ai/api/v1/images/generations -d '{}'
404   # 路径不存在

2. 单元测试 / Unit tests（新增 openrouter_image_test.go，覆盖图像 URL 特判 + chat 路径不受影响）：

$ go build ./...   # ok
$ go test ./relay/channel/openai/...
ok  	github.com/QuantumNous/new-api/relay/channel/openai	0.367s

3. 端到端验证 / E2E — 本地启动 new-api（SQLite），配置 type=20 OpenRouter 渠道指向一个记录请求路径的 mock 上游，然后调用 new-api 的 /v1/images/generations：

$ curl -s -X POST http://127.0.0.1:18791/v1/images/generations \
    -H "Authorization: Bearer sk-..." \
    -d '{"model":"google/gemini-2.5-flash-image","prompt":"a cute cat"}'
{"created":1751800000,"data":[{"b64_json":"aGVsbG8="}],"usage":{"completion_tokens":1290,"prompt_tokens":10,"total_tokens":1300}}

# mock 上游日志（确认命中扁平路径 + 渠道密钥透传）:
MOCK UPSTREAM HIT: POST /v1/images Authorization="Bearer sk-or-test"

# new-api 计费日志（usage 正确入账）:
google/gemini-2.5-flash-image  prompt_tokens=10  completion_tokens=1290  quota=48750

Summary by CodeRabbit

• Bug Fixes
  • Fixed OpenRouter image-generation routing to use the correct /v1/images endpoint.
  • Updated OpenRouter image requests to merge additional image parameters from extra fields into the outbound JSON body, while preserving existing/known fields.
  • Ensured non-OpenRouter channels omit these extra/unknown image fields from the serialized payload.
• Tests
  • Added unit tests covering OpenRouter URL rewriting, chat routing staying unchanged, and request-body transformation for both OpenRouter and non-OpenRouter channels.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 88d4449f-3ca4-46d5-9cf0-a20bb8776f68

📥 Commits

Reviewing files that changed from the base of the PR and between 017cf45 and 7684bf2.

📒 Files selected for processing (3)

• relay/channel/openai/adaptor.go
• relay/channel/openai/openrouter_image.go
• relay/channel/openai/openrouter_image_test.go

---

Walkthrough

This PR adds OpenRouter-specific image-generation URL routing and image-body merging. OpenRouter image requests now use /v1/images, and extra image parameters are merged into the serialized request body. Tests cover both the routing and payload behavior.

Changes

OpenRouter image handling

Layer / File(s)	Summary
Image URL rewrite relay/channel/openai/adaptor.go, relay/channel/openai/openrouter_image_test.go	GetRequestURL now routes OpenRouter image-generation requests to {ChannelBaseUrl}/v1/images, and tests verify that image requests rewrite while chat completions do not.
Image body extra merge relay/channel/openai/openrouter_image.go, relay/channel/openai/adaptor.go, relay/channel/openai/openrouter_image_test.go	ConvertImageRequest now uses a JSON merge helper for OpenRouter image generation, preserving known fields and injecting extra request fields into the outbound body; tests verify OpenRouter merging and non-OpenRouter omission.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Possibly related PRs

• QuantumNous/new-api#2036: Also adds RelayModeImagesGenerations handling in an adaptor, including image URL rewriting and custom image request conversion.
• QuantumNous/new-api#2356: Touches dto.ImageRequest, which affects which fields are treated as known versus merged from request.Extra here.
• QuantumNous/new-api#4646: Updates the dto.ImageRequest schema in a way that can change OpenRouter image-body merging behavior.

Suggested reviewers: creamlike1024, seefs001

Poem

A bunny hops through OpenRouter trails,
Images now skip the winding rails.
Extras hop in when the body is spun,
Chat stays steady, and routing is done.
🐇

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: adding OpenRouter support for the image generation endpoint.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@relay/channel/openai/openrouter_image_test.go`:
- Around line 14-62: The new tests in TestGetRequestURLOpenRouterImageGeneration
and TestGetRequestURLOpenRouterChatUnchanged should use testify instead of raw
t.Fatalf. Update the openrouter_image_test.go test setup to import
github.com/stretchr/testify/require and use require.NoError for GetRequestURL
errors, then use assert.Equal for the returned URL comparisons. Keep the
existing test names and Adaptor/GetRequestURL usage, but replace manual fatal
checks with the appropriate require/assert helpers throughout.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: dc97ac6c-a51f-4c6e-adbe-765e46fce95e

📥 Commits

Reviewing files that changed from the base of the PR and between 5cbb7b0 and 6915d61.

📒 Files selected for processing (2)

• relay/channel/openai/adaptor.go
• relay/channel/openai/openrouter_image_test.go

[Robinnnnn]

Addressed the review comment — tests now use require.NoError / assert.Equal from testify.

[Calcium-Ion]

这里把 OpenRouter 图片生成请求改到 /v1/images 是对的，但请求体仍走通用 dto.ImageRequest 序列化。当前 ImageRequest.MarshalJSON 不会把 Extra 中捕获的未知字段合并回去，所以 OpenRouter 图片 API 支持的 resolution、aspect_ratio、seed、input_references、provider.options 等参数会被丢弃，导致请求无法按用户指定配置生成图片。

建议补充 OpenRouter 图片 API 需要的显式字段，或增加 OpenRouter 专用 image DTO/转换路径，并加一个回归测试验证至少一个 OpenRouter 专属参数（如 aspect_ratio 或 provider.options）会被转发到上游。

参考：https://openrouter.ai/docs/guides/overview/multimodal/image-generation.md

[Robinnnnn]

@Calcium-Ion @t0ng7u @seefs001 would appreciate a review!

[Robinnnnn]

感谢审阅——您说得对，仅修复 URL 还不够。通用的 ImageRequest.MarshalJSON 会丢弃 Extra 中捕获的所有字段，导致 aspect_ratio / resolution / seed / input_references / provider 无法到达 OpenRouter。已在最新提交中修复。

实现方式： 我注意到 MarshalJSON 是有意不把 Extra 全局合并回去的（dto/openai_image.go 中被注释掉的那段代码），所以没有改动 DTO 本身。而是在 ConvertImageRequest 中加了一个仅针对该渠道的转换：当渠道为 OpenRouter 且 relay mode 为图片生成时，把请求展平为 map[string]json.RawMessage 并合并 Extra（键冲突时已知字段优先）。其他所有渠道以及 image edits 的 multipart 路径序列化行为完全不变。

测试： TestConvertImageRequestOpenRouterMergesExtra 验证 aspect_ratio、seed、provider.options 能保留到出站请求体中；TestConvertImageRequestNonOpenRouterDropsExtra 保证非 OpenRouter 渠道仍然丢弃未知字段（保持现有行为不变）。

已用真实 API 调用完成端到端验证： 本地运行 new-api 实例并配置 OpenRouter 渠道，向 POST /v1/images/generations 发送带 aspect_ratio: "16:9" 和 seed: 42 的请求。调试日志确认出站请求发往 https://openrouter.ai/api/v1/images，请求体为 {"aspect_ratio":"16:9","model":"google/gemini-2.5-flash-image","n":1,"prompt":"...","seed":42}，返回的图片尺寸为 1344×768——正好是 16:9；而 aspect_ratio: "1:1" 的对照请求返回 1024×1024。provider 选项也以同样方式透传。

这种方式能通用覆盖 OpenRouter 图片生成文档 中的所有参数——OpenRouter 未来新增的参数也无需再改动即可透传。如果您更倾向于显式的类型化字段，我也很乐意调整。