docs: update README to include support for Gemini multimodal output and add relevant images

fix: ensure nil check for content text in token counting function

README.md

@@ -36,6 +36,7 @@ Also welcome to register and use my deployed one-api gateway, which supports var
       - [Reasoning Format - thinking](#reasoning-format---thinking)
     - [Support AWS cross-region inferences](#support-aws-cross-region-inferences)
     - [Support OpenAI web search models](#support-openai-web-search-models)
+    - [Support gemini multimodal output #2197](#support-gemini-multimodal-output-2197)
   - [Bug fix](#bug-fix)
 
 ## Turtorial
@@ -223,6 +224,16 @@ Supports two URL parameters: `thinking` and `reasoning_format`.
 
 support `gpt-4o-search-preview` & `gpt-4o-mini-search-preview`
 
+![](https://s3.laisky.com/uploads/2025/03/openai-websearch-models-1.png)
+
+![](https://s3.laisky.com/uploads/2025/03/openai-websearch-models-2.png)
+
+### Support gemini multimodal output #2197
+
+- [feature: support gemini multimodal output #2197](https://github.com/songquanpeng/one-api/pull/2197)
+
+![](https://s3.laisky.com/uploads/2025/03/gemini-multimodal.png)
+
 ## Bug fix
 
 - [BUGFIX: Several issues when updating tokens #1933](https://github.com/songquanpeng/one-api/pull/1933)

relay/adaptor/openai/token.go

@@ -103,7 +103,9 @@ func CountTokenMessages(ctx context.Context,
 		for _, content := range contents {
 			switch content.Type {
 			case model.ContentTypeText:
-				tokenNum += getTokenNum(tokenEncoder, content.Text)
+				if content.Text != nil {
+					tokenNum += getTokenNum(tokenEncoder, *content.Text)
+				}
 			case model.ContentTypeImageURL:
 				imageTokens, err := countImageTokens(
 					content.ImageURL.Url,