feat(images): add support for configurable GPT Image 2 base model and improved SSE handling

- Introduced `GPTImage2BaseModel` configuration for hosted image generation tools with validation for "gpt-" prefix.
- Added logic to dynamically resolve and apply the base model in Codex executor workflows.
- Enhanced server-sent events (SSE) implementation with keep-alive tickers and error events for stream reliability.
- Updated configuration file examples and internal documentation.

Author: luispater

config.example.yaml

@@ -100,6 +100,10 @@ disable-cooling: false
 # - "chat": disable image_generation injection on non-images endpoints, but keep /v1/images/generations and /v1/images/edits enabled.
 disable-image-generation: false
 
+# Base model used when proxying gpt-image-2 via the hosted image_generation tool (Responses API).
+# Must start with "gpt-" (case-insensitive). If unset or invalid, defaults to "gpt-5.4-mini".
+# gpt-image-2-base-model: "gpt-5.4-mini"
+
 # Core auth auto-refresh worker pool size (OAuth/file-based auth token refresh).
 # When > 0, overrides the default worker count (16).
 # auth-auto-refresh-workers: 16

internal/config/sdk_config.go

@@ -19,6 +19,13 @@ type SDKConfig struct {
 	//     while keeping /v1/images/generations and /v1/images/edits enabled and preserving image_generation there.
 	DisableImageGeneration DisableImageGenerationMode `yaml:"disable-image-generation" json:"disable-image-generation"`
 
+	// GPTImage2BaseModel sets the base (mainline) model used when proxying GPT Image 2
+	// requests via the hosted image_generation tool (e.g. Codex OAuth /v1/images/*).
+	//
+	// The value must start with "gpt-" (case-insensitive). If empty or invalid, the
+	// default base model ("gpt-5.4-mini") is used.
+	GPTImage2BaseModel string `yaml:"gpt-image-2-base-model,omitempty" json:"gpt-image-2-base-model,omitempty"`
+
 	// EnableGeminiCLIEndpoint controls whether Gemini CLI internal endpoints (/v1internal:*) are enabled.
 	// Default is false for safety; when false, /v1internal:* requests are rejected.
 	EnableGeminiCLIEndpoint bool `yaml:"enable-gemini-cli-endpoint" json:"enable-gemini-cli-endpoint"`

internal/runtime/executor/codex_openai_images.go

@@ -63,6 +63,20 @@ func codexIsImagesEndpointPath(path string) bool {
 	return strings.HasSuffix(path, codexImagesGenerationsPath) || strings.HasSuffix(path, codexImagesEditsPath)
 }
 
+func (e *CodexExecutor) resolveGPTImage2BaseModel() string {
+	if e == nil || e.cfg == nil {
+		return codexOpenAIImagesMainModel
+	}
+	model := strings.TrimSpace(e.cfg.GPTImage2BaseModel)
+	if model == "" {
+		return codexOpenAIImagesMainModel
+	}
+	if strings.HasPrefix(strings.ToLower(model), "gpt-") {
+		return model
+	}
+	return codexOpenAIImagesMainModel
+}
+
 func (e *CodexExecutor) executeOpenAIImage(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	prepared, errPrepare := codexPrepareOpenAIImageRequest(req, opts)
 	if errPrepare != nil {
@@ -74,10 +88,11 @@ func (e *CodexExecutor) executeOpenAIImage(ctx context.Context, auth *cliproxyau
 		baseURL = "https://chatgpt.com/backend-api/codex"
 	}
 
-	reporter := helps.NewUsageReporter(ctx, e.Identifier(), codexOpenAIImagesMainModel, auth)
+	mainModel := e.resolveGPTImage2BaseModel()
+	reporter := helps.NewUsageReporter(ctx, e.Identifier(), mainModel, auth)
 	defer reporter.TrackFailure(ctx, &err)
 
-	body, errBuild := e.prepareCodexOpenAIImageBody(prepared.Body, req, opts)
+	body, errBuild := e.prepareCodexOpenAIImageBody(prepared.Body, req, opts, mainModel)
 	if errBuild != nil {
 		return resp, errBuild
 	}
@@ -161,10 +176,11 @@ func (e *CodexExecutor) executeOpenAIImageStream(ctx context.Context, auth *clip
 		baseURL = "https://chatgpt.com/backend-api/codex"
 	}
 
-	reporter := helps.NewUsageReporter(ctx, e.Identifier(), codexOpenAIImagesMainModel, auth)
+	mainModel := e.resolveGPTImage2BaseModel()
+	reporter := helps.NewUsageReporter(ctx, e.Identifier(), mainModel, auth)
 	defer reporter.TrackFailure(ctx, &err)
 
-	body, errBuild := e.prepareCodexOpenAIImageBody(prepared.Body, req, opts)
+	body, errBuild := e.prepareCodexOpenAIImageBody(prepared.Body, req, opts, mainModel)
 	if errBuild != nil {
 		return nil, errBuild
 	}
@@ -277,18 +293,22 @@ func (e *CodexExecutor) executeOpenAIImageStream(ctx context.Context, auth *clip
 	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
-func (e *CodexExecutor) prepareCodexOpenAIImageBody(body []byte, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) ([]byte, error) {
+func (e *CodexExecutor) prepareCodexOpenAIImageBody(body []byte, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, mainModel string) ([]byte, error) {
 	out := body
+	mainModel = strings.TrimSpace(mainModel)
+	if mainModel == "" {
+		mainModel = codexOpenAIImagesMainModel
+	}
 	var errThinking error
-	out, errThinking = thinking.ApplyThinking(out, codexOpenAIImagesMainModel, codexOpenAIImageSourceFormat, "codex", e.Identifier())
+	out, errThinking = thinking.ApplyThinking(out, mainModel, codexOpenAIImageSourceFormat, "codex", e.Identifier())
 	if errThinking != nil {
 		return nil, errThinking
 	}
 
 	requestedModel := helps.PayloadRequestedModel(opts, req.Model)
 	requestPath := helps.PayloadRequestPath(opts)
-	out = helps.ApplyPayloadConfigWithRequest(e.cfg, codexOpenAIImagesMainModel, "codex", codexOpenAIImageSourceFormat, "", out, body, requestedModel, requestPath, opts.Headers)
-	out, _ = sjson.SetBytes(out, "model", codexOpenAIImagesMainModel)
+	out = helps.ApplyPayloadConfigWithRequest(e.cfg, mainModel, "codex", codexOpenAIImageSourceFormat, "", out, body, requestedModel, requestPath, opts.Headers)
+	out, _ = sjson.SetBytes(out, "model", mainModel)
 	out, _ = sjson.SetBytes(out, "stream", true)
 	out, _ = sjson.DeleteBytes(out, "previous_response_id")
 	out, _ = sjson.DeleteBytes(out, "prompt_cache_retention")

internal/watcher/diff/config_diff.go

@@ -48,6 +48,9 @@ func BuildConfigChangeDetails(oldCfg, newCfg *config.Config) []string {
 	if oldCfg.DisableImageGeneration != newCfg.DisableImageGeneration {
 		changes = append(changes, fmt.Sprintf("disable-image-generation: %v -> %v", oldCfg.DisableImageGeneration, newCfg.DisableImageGeneration))
 	}
+	if strings.TrimSpace(oldCfg.GPTImage2BaseModel) != strings.TrimSpace(newCfg.GPTImage2BaseModel) {
+		changes = append(changes, fmt.Sprintf("gpt-image-2-base-model: %s -> %s", strings.TrimSpace(oldCfg.GPTImage2BaseModel), strings.TrimSpace(newCfg.GPTImage2BaseModel)))
+	}
 	if oldCfg.RequestLog != newCfg.RequestLog {
 		changes = append(changes, fmt.Sprintf("request-log: %t -> %t", oldCfg.RequestLog, newCfg.RequestLog))
 	}