refactor: standardize dataTag processing across response translators

- Unified `dataTag` initialization by removing spaces after `data:`.
- Replaced manual slicing with `bytes.TrimSpace` for consistent and robust handling of JSON payloads.

internal/client/gemini-cli_client.go

@@ -554,7 +554,7 @@ func (c *GeminiCLIClient) SendRawMessageStream(ctx context.Context, modelName st
 	rawJSON, _ = sjson.SetBytes(rawJSON, "project", c.GetProjectID())
 	rawJSON, _ = sjson.SetBytes(rawJSON, "model", modelName)
 
-	dataTag := []byte("data: ")
+	dataTag := []byte("data:")
 	errChan := make(chan *interfaces.ErrorMessage)
 	dataChan := make(chan []byte)
 	// log.Debugf(string(rawJSON))
@@ -619,7 +619,7 @@ func (c *GeminiCLIClient) SendRawMessageStream(ctx context.Context, modelName st
 				for scanner.Scan() {
 					line := scanner.Bytes()
 					if bytes.HasPrefix(line, dataTag) {
-						lines := translator.Response(handlerType, c.Type(), newCtx, modelName, originalRequestRawJSON, rawJSON, line[6:], &param)
+						lines := translator.Response(handlerType, c.Type(), newCtx, modelName, originalRequestRawJSON, rawJSON, bytes.TrimSpace(line[5:]), &param)
 						for i := 0; i < len(lines); i++ {
 							dataChan <- []byte(lines[i])
 						}
@@ -630,7 +630,7 @@ func (c *GeminiCLIClient) SendRawMessageStream(ctx context.Context, modelName st
 				for scanner.Scan() {
 					line := scanner.Bytes()
 					if bytes.HasPrefix(line, dataTag) {
-						dataChan <- line[6:]
+						dataChan <- bytes.TrimSpace(line[5:])
 					}
 					c.AddAPIResponseData(ctx, line)
 				}

internal/client/gemini-web/client.go

@@ -33,6 +33,10 @@ type GeminiClient struct {
 	accountLabel    string
 }
 
+var NanoBananaModel = map[string]struct{}{
+	"gemini-2.5-flash-image-preview": {},
+}
+
 // NewGeminiClient creates a client. Pass empty strings to auto-detect via browser cookies (not implemented in Go port).
 func NewGeminiClient(secure1psid string, secure1psidts string, proxy string, opts ...func(*GeminiClient)) *GeminiClient {
 	c := &GeminiClient{
@@ -239,6 +243,14 @@ func (c *GeminiClient) GenerateContent(prompt string, files []string, model Mode
 	}
 }
 
+func ensureAnyLen(slice []any, index int) []any {
+	if index < len(slice) {
+		return slice
+	}
+	gap := index + 1 - len(slice)
+	return append(slice, make([]any, gap)...)
+}
+
 func (c *GeminiClient) generateOnce(prompt string, files []string, model Model, gem *Gem, chat *ChatSession) (ModelOutput, error) {
 	var empty ModelOutput
 	// Build f.req
@@ -266,6 +278,14 @@ func (c *GeminiClient) generateOnce(prompt string, files []string, model Model,
 	}
 
 	inner := []any{item0, nil, item2}
+	requestedModel := strings.ToLower(model.Name)
+	if chat != nil && chat.RequestedModel() != "" {
+		requestedModel = chat.RequestedModel()
+	}
+	if _, ok := NanoBananaModel[requestedModel]; ok {
+		inner = ensureAnyLen(inner, 49)
+		inner[49] = 14
+	}
 	if gem != nil {
 		// pad with 16 nils then gem ID
 		for i := 0; i < 16; i++ {
@@ -674,16 +694,17 @@ func truncateForLog(s string, n int) string {
 
 // StartChat returns a ChatSession attached to the client
 func (c *GeminiClient) StartChat(model Model, gem *Gem, metadata []string) *ChatSession {
-	return &ChatSession{client: c, metadata: normalizeMeta(metadata), model: model, gem: gem}
+	return &ChatSession{client: c, metadata: normalizeMeta(metadata), model: model, gem: gem, requestedModel: strings.ToLower(model.Name)}
 }
 
 // ChatSession holds conversation metadata
 type ChatSession struct {
-	client     *GeminiClient
-	metadata   []string // cid, rid, rcid
-	lastOutput *ModelOutput
-	model      Model
-	gem        *Gem
+	client         *GeminiClient
+	metadata       []string // cid, rid, rcid
+	lastOutput     *ModelOutput
+	model          Model
+	gem            *Gem
+	requestedModel string
 }
 
 func (cs *ChatSession) String() string {
@@ -710,6 +731,10 @@ func normalizeMeta(v []string) []string {
 
 func (cs *ChatSession) Metadata() []string     { return cs.metadata }
 func (cs *ChatSession) SetMetadata(v []string) { cs.metadata = normalizeMeta(v) }
+func (cs *ChatSession) RequestedModel() string { return cs.requestedModel }
+func (cs *ChatSession) SetRequestedModel(name string) {
+	cs.requestedModel = strings.ToLower(name)
+}
 func (cs *ChatSession) CID() string {
 	if len(cs.metadata) > 0 {
 		return cs.metadata[0]

internal/client/gemini-web/models.go

@@ -116,6 +116,8 @@ func EnsureGeminiWebAliasMap() {
 		for _, m := range registry.GetGeminiModels() {
 			if m.ID == "gemini-2.5-flash-lite" {
 				continue
+			} else if m.ID == "gemini-2.5-flash" {
+				GeminiWebAliasMap["gemini-2.5-flash-image-preview"] = "gemini-2.5-flash"
 			}
 			alias := AliasFromModelID(m.ID)
 			GeminiWebAliasMap[strings.ToLower(alias)] = strings.ToLower(m.ID)
@@ -130,6 +132,13 @@ func GetGeminiWebAliasedModels() []*registry.ModelInfo {
 	for _, m := range registry.GetGeminiModels() {
 		if m.ID == "gemini-2.5-flash-lite" {
 			continue
+		} else if m.ID == "gemini-2.5-flash" {
+			cpy := *m
+			cpy.ID = "gemini-2.5-flash-image-preview"
+			cpy.Name = "gemini-2.5-flash-image-preview"
+			cpy.DisplayName = "Nano Banana"
+			cpy.Description = "Gemini 2.5 Flash Preview Image"
+			aliased = append(aliased, &cpy)
 		}
 		cpy := *m
 		cpy.ID = AliasFromModelID(m.ID)

internal/client/gemini-web_client.go

@@ -207,7 +207,7 @@ func (c *GeminiWebClient) registerModelsOnce() {
 	if c.modelsRegistered {
 		return
 	}
-	c.RegisterModels(GEMINI, geminiWeb.GetGeminiWebAliasedModels())
+	c.RegisterModels(GEMINIWEB, geminiWeb.GetGeminiWebAliasedModels())
 	c.modelsRegistered = true
 }
 
@@ -219,8 +219,8 @@ func (c *GeminiWebClient) EnsureRegistered() {
 	}
 }
 
-func (c *GeminiWebClient) Type() string     { return GEMINI }
-func (c *GeminiWebClient) Provider() string { return GEMINI }
+func (c *GeminiWebClient) Type() string     { return GEMINIWEB }
+func (c *GeminiWebClient) Provider() string { return GEMINIWEB }
 func (c *GeminiWebClient) CanProvideModel(modelName string) bool {
 	geminiWeb.EnsureGeminiWebAliasMap()
 	_, ok := geminiWeb.GeminiWebAliasMap[strings.ToLower(modelName)]
@@ -394,6 +394,7 @@ func (c *GeminiWebClient) prepareChat(ctx context.Context, modelName string, raw
 	c.appendUpstreamRequestLog(ctx, modelName, res.tagged, true, res.prompt, len(uploadedFiles), res.reuse, res.metaLen)
 	gem := c.getConfiguredGem()
 	res.chat = c.gwc.StartChat(model, gem, meta)
+	res.chat.SetRequestedModel(modelName)
 	return res, nil
 }
 

internal/client/gemini_client.go

@@ -298,7 +298,7 @@ func (c *GeminiClient) SendRawMessageStream(ctx context.Context, modelName strin
 	handlerType := handler.HandlerType()
 	rawJSON = translator.Request(handlerType, c.Type(), modelName, rawJSON, true)
 
-	dataTag := []byte("data: ")
+	dataTag := []byte("data:")
 	errChan := make(chan *interfaces.ErrorMessage)
 	dataChan := make(chan []byte)
 	// log.Debugf(string(rawJSON))
@@ -342,7 +342,7 @@ func (c *GeminiClient) SendRawMessageStream(ctx context.Context, modelName strin
 				for scanner.Scan() {
 					line := scanner.Bytes()
 					if bytes.HasPrefix(line, dataTag) {
-						lines := translator.Response(handlerType, c.Type(), newCtx, modelName, originalRequestRawJSON, rawJSON, line[6:], &param)
+						lines := translator.Response(handlerType, c.Type(), newCtx, modelName, originalRequestRawJSON, rawJSON, bytes.TrimSpace(line[5:]), &param)
 						for i := 0; i < len(lines); i++ {
 							dataChan <- []byte(lines[i])
 						}
@@ -353,7 +353,7 @@ func (c *GeminiClient) SendRawMessageStream(ctx context.Context, modelName strin
 				for scanner.Scan() {
 					line := scanner.Bytes()
 					if bytes.HasPrefix(line, dataTag) {
-						dataChan <- line[6:]
+						dataChan <- bytes.TrimSpace(line[5:])
 					}
 					c.AddAPIResponseData(ctx, line)
 				}

internal/client/openai-compatibility_client.go

@@ -291,9 +291,8 @@ func (c *OpenAICompatibilityClient) SendRawMessageStream(ctx context.Context, mo
 	handlerType := handler.HandlerType()
 	rawJSON = translator.Request(handlerType, c.Type(), modelName, rawJSON, true)
 
-	dataTag := []byte("data: ")
-	dataUglyTag := []byte("data:") // Some APIs providers don't add space after "data:", fuck for them all
-	doneTag := []byte("data: [DONE]")
+	dataTag := []byte("data:")
+	doneTag := []byte("[DONE]")
 	errChan := make(chan *interfaces.ErrorMessage)
 	dataChan := make(chan []byte)
 	// log.Debugf(string(rawJSON))
@@ -332,19 +331,10 @@ func (c *OpenAICompatibilityClient) SendRawMessageStream(ctx context.Context, mo
 			for scanner.Scan() {
 				line := scanner.Bytes()
 				if bytes.HasPrefix(line, dataTag) {
-					if bytes.Equal(line, doneTag) {
+					if bytes.Equal(bytes.TrimSpace(line[5:]), doneTag) {
 						break
 					}
-					lines := translator.Response(handlerType, c.Type(), newCtx, modelName, originalRequestRawJSON, rawJSON, line[6:], &param)
-					for i := 0; i < len(lines); i++ {
-						c.AddAPIResponseData(ctx, line)
-						dataChan <- []byte(lines[i])
-					}
-				} else if bytes.HasPrefix(line, dataUglyTag) {
-					if bytes.Equal(line, doneTag) {
-						break
-					}
-					lines := translator.Response(handlerType, c.Type(), newCtx, modelName, originalRequestRawJSON, rawJSON, line[5:], &param)
+					lines := translator.Response(handlerType, c.Type(), newCtx, modelName, originalRequestRawJSON, rawJSON, bytes.TrimSpace(line[5:]), &param)
 					for i := 0; i < len(lines); i++ {
 						c.AddAPIResponseData(ctx, line)
 						dataChan <- []byte(lines[i])
@@ -356,13 +346,10 @@ func (c *OpenAICompatibilityClient) SendRawMessageStream(ctx context.Context, mo
 			for scanner.Scan() {
 				line := scanner.Bytes()
 				if bytes.HasPrefix(line, dataTag) {
-					if bytes.Equal(line, doneTag) {
+					if bytes.Equal(bytes.TrimSpace(line[5:]), doneTag) {
 						break
 					}
-					c.AddAPIResponseData(newCtx, line[6:])
-					dataChan <- line[6:]
-				} else if bytes.HasPrefix(line, dataUglyTag) {
-					c.AddAPIResponseData(newCtx, line[5:])
+					c.AddAPIResponseData(newCtx, bytes.TrimSpace(line[5:]))
 					dataChan <- line[5:]
 				}
 			}

internal/client/qwen_client.go

@@ -215,8 +215,8 @@ func (c *QwenClient) SendRawMessageStream(ctx context.Context, modelName string,
 	handlerType := handler.HandlerType()
 	rawJSON = translator.Request(handlerType, c.Type(), modelName, rawJSON, true)
 
-	dataTag := []byte("data: ")
-	doneTag := []byte("data: [DONE]")
+	dataTag := []byte("data:")
+	doneTag := []byte("[DONE]")
 	errChan := make(chan *interfaces.ErrorMessage)
 	dataChan := make(chan []byte)
 
@@ -264,7 +264,7 @@ func (c *QwenClient) SendRawMessageStream(ctx context.Context, modelName string,
 			for scanner.Scan() {
 				line := scanner.Bytes()
 				if bytes.HasPrefix(line, dataTag) {
-					lines := translator.Response(handlerType, c.Type(), ctx, modelName, originalRequestRawJSON, rawJSON, line[6:], &param)
+					lines := translator.Response(handlerType, c.Type(), ctx, modelName, originalRequestRawJSON, rawJSON, bytes.TrimSpace(line[5:]), &param)
 					for i := 0; i < len(lines); i++ {
 						dataChan <- []byte(lines[i])
 					}
@@ -274,9 +274,9 @@ func (c *QwenClient) SendRawMessageStream(ctx context.Context, modelName string,
 		} else {
 			for scanner.Scan() {
 				line := scanner.Bytes()
-				if !bytes.HasPrefix(line, doneTag) {
-					if bytes.HasPrefix(line, dataTag) {
-						dataChan <- line[6:]
+				if bytes.HasPrefix(line, dataTag) {
+					if !bytes.Equal(bytes.TrimSpace(line[5:]), doneTag) {
+						dataChan <- bytes.TrimSpace(line[5:])
 					}
 				}
 				c.AddAPIResponseData(ctx, line)

internal/constant/constant.go

@@ -3,6 +3,7 @@ package constant
 const (
 	GEMINI          = "gemini"
 	GEMINICLI       = "gemini-cli"
+	GEMINIWEB       = "gemini-web"
 	CODEX           = "codex"
 	CLAUDE          = "claude"
 	OPENAI          = "openai"

internal/translator/claude/gemini/claude_gemini_response.go

@@ -17,7 +17,7 @@ import (
 )
 
 var (
-	dataTag = []byte("data: ")
+	dataTag = []byte("data:")
 )
 
 // ConvertAnthropicResponseToGeminiParams holds parameters for response conversion
@@ -64,7 +64,7 @@ func ConvertClaudeResponseToGemini(_ context.Context, modelName string, original
 	if !bytes.HasPrefix(rawJSON, dataTag) {
 		return []string{}
 	}
-	rawJSON = rawJSON[6:]
+	rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 	root := gjson.ParseBytes(rawJSON)
 	eventType := root.Get("type").String()
@@ -336,7 +336,7 @@ func ConvertClaudeResponseToGeminiNonStream(_ context.Context, modelName string,
 		line := scanner.Bytes()
 		// log.Debug(string(line))
 		if bytes.HasPrefix(line, dataTag) {
-			jsonData := line[6:]
+			jsonData := bytes.TrimSpace(line[5:])
 			streamingEvents = append(streamingEvents, jsonData)
 		}
 	}

internal/translator/claude/openai/chat-completions/claude_openai_response.go

@@ -18,7 +18,7 @@ import (
 )
 
 var (
-	dataTag = []byte("data: ")
+	dataTag = []byte("data:")
 )
 
 // ConvertAnthropicResponseToOpenAIParams holds parameters for response conversion
@@ -62,7 +62,7 @@ func ConvertClaudeResponseToOpenAI(_ context.Context, modelName string, original
 	if !bytes.HasPrefix(rawJSON, dataTag) {
 		return []string{}
 	}
-	rawJSON = rawJSON[6:]
+	rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 	root := gjson.ParseBytes(rawJSON)
 	eventType := root.Get("type").String()
@@ -289,7 +289,7 @@ func ConvertClaudeResponseToOpenAINonStream(_ context.Context, _ string, origina
 		if !bytes.HasPrefix(line, dataTag) {
 			continue
 		}
-		chunks = append(chunks, line[6:])
+		chunks = append(chunks, bytes.TrimSpace(rawJSON[5:]))
 	}
 
 	// Base OpenAI non-streaming response template

internal/translator/claude/openai/responses/claude_openai-responses_response.go

@@ -34,7 +34,7 @@ type claudeToResponsesState struct {
 	ReasoningIndex     int
 }
 
-var dataTag = []byte("data: ")
+var dataTag = []byte("data:")
 
 func emitEvent(event string, payload string) string {
 	return fmt.Sprintf("event: %s\ndata: %s\n\n", event, payload)
@@ -51,7 +51,7 @@ func ConvertClaudeResponseToOpenAIResponses(ctx context.Context, modelName strin
 	if !bytes.HasPrefix(rawJSON, dataTag) {
 		return []string{}
 	}
-	rawJSON = rawJSON[6:]
+	rawJSON = bytes.TrimSpace(rawJSON[5:])
 	root := gjson.ParseBytes(rawJSON)
 	ev := root.Get("type").String()
 	var out []string

internal/translator/codex/claude/codex_claude_response.go

@@ -16,7 +16,7 @@ import (
 )
 
 var (
-	dataTag = []byte("data: ")
+	dataTag = []byte("data:")
 )
 
 // ConvertCodexResponseToClaude performs sophisticated streaming response format conversion.
@@ -45,7 +45,7 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 	if !bytes.HasPrefix(rawJSON, dataTag) {
 		return []string{}
 	}
-	rawJSON = rawJSON[6:]
+	rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 	output := ""
 	rootResult := gjson.ParseBytes(rawJSON)

internal/translator/codex/gemini/codex_gemini_response.go

@@ -16,7 +16,7 @@ import (
 )
 
 var (
-	dataTag = []byte("data: ")
+	dataTag = []byte("data:")
 )
 
 // ConvertCodexResponseToGeminiParams holds parameters for response conversion.
@@ -53,7 +53,7 @@ func ConvertCodexResponseToGemini(_ context.Context, modelName string, originalR
 	if !bytes.HasPrefix(rawJSON, dataTag) {
 		return []string{}
 	}
-	rawJSON = rawJSON[6:]
+	rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 	rootResult := gjson.ParseBytes(rawJSON)
 	typeResult := rootResult.Get("type")
@@ -161,7 +161,7 @@ func ConvertCodexResponseToGeminiNonStream(_ context.Context, modelName string,
 		if !bytes.HasPrefix(line, dataTag) {
 			continue
 		}
-		rawJSON = line[6:]
+		rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 		rootResult := gjson.ParseBytes(rawJSON)
 

internal/translator/codex/openai/chat-completions/codex_openai_response.go

@@ -16,7 +16,7 @@ import (
 )
 
 var (
-	dataTag = []byte("data: ")
+	dataTag = []byte("data:")
 )
 
 // ConvertCliToOpenAIParams holds parameters for response conversion.
@@ -54,7 +54,7 @@ func ConvertCodexResponseToOpenAI(_ context.Context, modelName string, originalR
 	if !bytes.HasPrefix(rawJSON, dataTag) {
 		return []string{}
 	}
-	rawJSON = rawJSON[6:]
+	rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 	// Initialize the OpenAI SSE template.
 	template := `{"id":"","object":"chat.completion.chunk","created":12345,"model":"model","choices":[{"index":0,"delta":{"role":null,"content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}]}`
@@ -175,7 +175,7 @@ func ConvertCodexResponseToOpenAINonStream(_ context.Context, _ string, original
 		if !bytes.HasPrefix(line, dataTag) {
 			continue
 		}
-		rawJSON = line[6:]
+		rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 		rootResult := gjson.ParseBytes(rawJSON)
 		// Verify this is a response.completed event

internal/translator/codex/openai/responses/codex_openai-responses_response.go

@@ -13,8 +13,8 @@ import (
 // ConvertCodexResponseToOpenAIResponses converts OpenAI Chat Completions streaming chunks
 // to OpenAI Responses SSE events (response.*).
 func ConvertCodexResponseToOpenAIResponses(ctx context.Context, modelName string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
-	if bytes.HasPrefix(rawJSON, []byte("data: ")) {
-		rawJSON = rawJSON[6:]
+	if bytes.HasPrefix(rawJSON, []byte("data:")) {
+		rawJSON = bytes.TrimSpace(rawJSON[5:])
 		if typeResult := gjson.GetBytes(rawJSON, "type"); typeResult.Exists() {
 			typeStr := typeResult.String()
 			if typeStr == "response.created" || typeStr == "response.in_progress" || typeStr == "response.completed" {
@@ -32,14 +32,14 @@ func ConvertCodexResponseToOpenAIResponsesNonStream(_ context.Context, modelName
 	scanner := bufio.NewScanner(bytes.NewReader(rawJSON))
 	buffer := make([]byte, 10240*1024)
 	scanner.Buffer(buffer, 10240*1024)
-	dataTag := []byte("data: ")
+	dataTag := []byte("data:")
 	for scanner.Scan() {
 		line := scanner.Bytes()
 
 		if !bytes.HasPrefix(line, dataTag) {
 			continue
 		}
-		rawJSON = line[6:]
+		rawJSON = bytes.TrimSpace(rawJSON[5:])
 
 		rootResult := gjson.ParseBytes(rawJSON)
 		// Verify this is a response.completed event

internal/translator/gemini-web/openai/chat-completions/init.go

@@ -0,0 +1,20 @@
+package chat_completions
+
+import (
+	. "github.com/luispater/CLIProxyAPI/v5/internal/constant"
+	"github.com/luispater/CLIProxyAPI/v5/internal/interfaces"
+	geminiChat "github.com/luispater/CLIProxyAPI/v5/internal/translator/gemini/openai/chat-completions"
+	"github.com/luispater/CLIProxyAPI/v5/internal/translator/translator"
+)
+
+func init() {
+	translator.Register(
+		OPENAI,
+		GEMINIWEB,
+		geminiChat.ConvertOpenAIRequestToGemini,
+		interfaces.TranslateResponse{
+			Stream:    geminiChat.ConvertGeminiResponseToOpenAI,
+			NonStream: geminiChat.ConvertGeminiResponseToOpenAINonStream,
+		},
+	)
+}

internal/translator/gemini-web/openai/responses/init.go

@@ -0,0 +1,20 @@
+package responses
+
+import (
+	. "github.com/luispater/CLIProxyAPI/v5/internal/constant"
+	"github.com/luispater/CLIProxyAPI/v5/internal/interfaces"
+	geminiResponses "github.com/luispater/CLIProxyAPI/v5/internal/translator/gemini/openai/responses"
+	"github.com/luispater/CLIProxyAPI/v5/internal/translator/translator"
+)
+
+func init() {
+	translator.Register(
+		OPENAI_RESPONSE,
+		GEMINIWEB,
+		geminiResponses.ConvertOpenAIResponsesRequestToGemini,
+		interfaces.TranslateResponse{
+			Stream:    geminiResponses.ConvertGeminiResponseToOpenAIResponses,
+			NonStream: geminiResponses.ConvertGeminiResponseToOpenAIResponsesNonStream,
+		},
+	)
+}

internal/translator/gemini/openai/chat-completions/gemini_openai_request.go

@@ -170,6 +170,31 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 					node := []byte(`{"role":"model","parts":[{"text":""}]}`)
 					node, _ = sjson.SetBytes(node, "parts.0.text", content.String())
 					out, _ = sjson.SetRawBytes(out, "contents.-1", node)
+				} else if content.IsArray() {
+					// Assistant multimodal content (e.g. text + image) -> single model content with parts
+					node := []byte(`{"role":"model","parts":[]}`)
+					p := 0
+					for _, item := range content.Array() {
+						switch item.Get("type").String() {
+						case "text":
+							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", item.Get("text").String())
+							p++
+						case "image_url":
+							// If the assistant returned an inline data URL, preserve it for history fidelity.
+							imageURL := item.Get("image_url.url").String()
+							if len(imageURL) > 5 { // expect data:...
+								pieces := strings.SplitN(imageURL[5:], ";", 2)
+								if len(pieces) == 2 && len(pieces[1]) > 7 {
+									mime := pieces[0]
+									data := pieces[1][7:]
+									node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".inlineData.mime_type", mime)
+									node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".inlineData.data", data)
+									p++
+								}
+							}
+						}
+					}
+					out, _ = sjson.SetRawBytes(out, "contents.-1", node)
 				} else if !content.Exists() || content.Type == gjson.Null {
 					// Tool calls -> single model content with functionCall parts
 					tcs := m.Get("tool_calls")

internal/translator/gemini/openai/chat-completions/gemini_openai_response.go

@@ -8,6 +8,7 @@ package chat_completions
 import (
 	"bytes"
 	"context"
+	"encoding/json"
 	"fmt"
 	"time"
 
@@ -99,6 +100,10 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 			partResult := partResults[i]
 			partTextResult := partResult.Get("text")
 			functionCallResult := partResult.Get("functionCall")
+			inlineDataResult := partResult.Get("inlineData")
+			if !inlineDataResult.Exists() {
+				inlineDataResult = partResult.Get("inline_data")
+			}
 
 			if partTextResult.Exists() {
 				// Handle text content, distinguishing between regular content and reasoning/thoughts.
@@ -124,6 +129,34 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 				}
 				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
 				template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallTemplate)
+			} else if inlineDataResult.Exists() {
+				data := inlineDataResult.Get("data").String()
+				if data == "" {
+					continue
+				}
+				mimeType := inlineDataResult.Get("mimeType").String()
+				if mimeType == "" {
+					mimeType = inlineDataResult.Get("mime_type").String()
+				}
+				if mimeType == "" {
+					mimeType = "image/png"
+				}
+				imageURL := fmt.Sprintf("data:%s;base64,%s", mimeType, data)
+				imagePayload, err := json.Marshal(map[string]any{
+					"type": "image_url",
+					"image_url": map[string]string{
+						"url": imageURL,
+					},
+				})
+				if err != nil {
+					continue
+				}
+				imagesResult := gjson.Get(template, "choices.0.delta.images")
+				if !imagesResult.Exists() || !imagesResult.IsArray() {
+					template, _ = sjson.SetRaw(template, "choices.0.delta.images", `[]`)
+				}
+				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+				template, _ = sjson.SetRaw(template, "choices.0.delta.images.-1", string(imagePayload))
 			}
 		}
 	}
@@ -193,6 +226,10 @@ func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, origina
 			partResult := partsResults[i]
 			partTextResult := partResult.Get("text")
 			functionCallResult := partResult.Get("functionCall")
+			inlineDataResult := partResult.Get("inlineData")
+			if !inlineDataResult.Exists() {
+				inlineDataResult = partResult.Get("inline_data")
+			}
 
 			if partTextResult.Exists() {
 				// Append text content, distinguishing between regular content and reasoning.
@@ -217,9 +254,34 @@ func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, origina
 				}
 				template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
 				template, _ = sjson.SetRaw(template, "choices.0.message.tool_calls.-1", functionCallItemTemplate)
-			} else {
-				// If no usable content is found, return an empty string.
-				return ""
+			} else if inlineDataResult.Exists() {
+				data := inlineDataResult.Get("data").String()
+				if data == "" {
+					continue
+				}
+				mimeType := inlineDataResult.Get("mimeType").String()
+				if mimeType == "" {
+					mimeType = inlineDataResult.Get("mime_type").String()
+				}
+				if mimeType == "" {
+					mimeType = "image/png"
+				}
+				imageURL := fmt.Sprintf("data:%s;base64,%s", mimeType, data)
+				imagePayload, err := json.Marshal(map[string]any{
+					"type": "image_url",
+					"image_url": map[string]string{
+						"url": imageURL,
+					},
+				})
+				if err != nil {
+					continue
+				}
+				imagesResult := gjson.Get(template, "choices.0.message.images")
+				if !imagesResult.Exists() || !imagesResult.IsArray() {
+					template, _ = sjson.SetRaw(template, "choices.0.message.images", `[]`)
+				}
+				template, _ = sjson.Set(template, "choices.0.message.role", "assistant")
+				template, _ = sjson.SetRaw(template, "choices.0.message.images.-1", string(imagePayload))
 			}
 		}
 	}

internal/translator/init.go

@@ -23,6 +23,9 @@ import (
 	_ "github.com/luispater/CLIProxyAPI/v5/internal/translator/gemini/openai/chat-completions"
 	_ "github.com/luispater/CLIProxyAPI/v5/internal/translator/gemini/openai/responses"
 
+	_ "github.com/luispater/CLIProxyAPI/v5/internal/translator/gemini-web/openai/chat-completions"
+	_ "github.com/luispater/CLIProxyAPI/v5/internal/translator/gemini-web/openai/responses"
+
 	_ "github.com/luispater/CLIProxyAPI/v5/internal/translator/openai/claude"
 	_ "github.com/luispater/CLIProxyAPI/v5/internal/translator/openai/gemini"
 	_ "github.com/luispater/CLIProxyAPI/v5/internal/translator/openai/gemini-cli"