github/sub2api
1
0
Fork 0
mirror of https://github.com/Wei-Shaw/sub2api.git synced 2026-10-07 15:38:22 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(openai): preserve OAuth image prompts verbatim

Browse Source
This commit is contained in:
SipengXie2024
2026-08-24 14:37:48 +00:00
parent e2d9b823f6
commit 329b92ef04
3 changed files with 23 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/internal/service/openai_images.go
+7 -6
View File
@@ -34,12 +34,13 @@ const (
openAIImagesGenerationsURL = "https://api.openai.com/v1/images/generations"
openAIImagesEditsURL = "https://api.openai.com/v1/images/edits"
openAIChatGPTStartURL = "https://chatgpt.com/"
openAIChatGPTFilesURL = "https://chatgpt.com/backend-api/files"
openAIImageBackendUserAgent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36"
openAIImageMaxDownloadBytes = 20 << 20 // 20MB per image download
openAIImageMaxUploadPartSize = 20 << 20 // 20MB per multipart upload part
openAIImagesResponsesMainModel = "gpt-5.4-mini"
openAIChatGPTStartURL = "https://chatgpt.com/"
openAIChatGPTFilesURL = "https://chatgpt.com/backend-api/files"
openAIImageBackendUserAgent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36"
openAIImageMaxDownloadBytes = 20 << 20 // 20MB per image download
openAIImageMaxUploadPartSize = 20 << 20 // 20MB per multipart upload part
openAIImagesResponsesMainModel = "gpt-5.4-mini"
openAIImagesVerbatimPromptInstructions = "When invoking the image_generation tool, use the user's image prompt verbatim. Do not rewrite, expand, summarize, embellish, translate, normalize punctuation, or add or remove visual details or constraints. Preserve the original language, wording, capitalization, quotes, and punctuation exactly."
)
type OpenAIImagesCapability string
backend/internal/service/openai_images_responses.go
+1
View File
@@ -356,6 +356,7 @@ func buildOpenAIImagesResponsesRequest(parsed *OpenAIImagesRequest, toolModel st
req := []byte(`{"instructions":"","stream":true,"reasoning":{"effort":"medium","summary":"auto"},"parallel_tool_calls":true,"include":["reasoning.encrypted_content"],"model":"","store":false,"tool_choice":{"type":"image_generation"}}`)
req, _ = sjson.SetBytes(req, "model", openAIImagesResponsesMainModel)
req, _ = sjson.SetBytes(req, "instructions", openAIImagesVerbatimPromptInstructions)
input := []byte(`[{"type":"message","role":"user","content":[{"type":"input_text","text":""}]}]`)
input, _ = sjson.SetBytes(input, "0.content.0.text", prompt)
backend/internal/service/openai_images_test.go
+15
View File
@@ -2011,6 +2011,21 @@ func TestBuildOpenAIImagesResponsesRequest_StripsInputFidelity(t *testing.T) {
require.Equal(t, "edit", gjson.GetBytes(body, "tools.0.action").String())
}
func TestBuildOpenAIImagesResponsesRequest_RequiresVerbatimUserPrompt(t *testing.T) {
prompt := "画一个蓝色马克杯，杯身只写“SkelOT”，保持大小写；白色背景，不要增加其他文字。"
parsed := &OpenAIImagesRequest{
Endpoint: openAIImagesGenerationsEndpoint,
Model: "gpt-image-2",
Prompt: prompt,
N: 1,
}
body, err := buildOpenAIImagesResponsesRequest(parsed, "gpt-image-2")
require.NoError(t, err)
require.Equal(t, openAIImagesVerbatimPromptInstructions, gjson.GetBytes(body, "instructions").String())
require.Equal(t, prompt, gjson.GetBytes(body, "input.0.content.0.text").String())
}
func TestCollectOpenAIImagesFromResponsesBody_FallsBackToOutputItemDone(t *testing.T) {
body := []byte(
"data: {\"type\":\"response.created\",\"response\":{\"created_at\":1710000004}}\n\n" +