GenarrativeAI/Genarrative
5
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Switch to VectorEngine gpt-image-2 and edits

Browse Source 
Replace uses of the legacy `gpt-image-2-all` model with `gpt-image-2` and standardize image workflows: no-reference generation uses POST /v1/images/generations, any-reference flows use POST /v1/images/edits with multipart `image` parts. Update SKILLs, generation scripts, decision logs, and docs to reflect the contract change and edits-vs-generations guidance. Apply corresponding changes across backend (api-server match3d/puzzle modules, openai image adapter, mappers, telemetry, spacetime client/module), frontend components and services (Match3D, Puzzle, CreativeImageInputPanel, runtime shells), and add new spritesheet/parser files and tests. Also add media/logo.png. These changes align repository code and documentation with the VectorEngine image API contract and update generation/upload handling (green-screen -> alpha processing, spritesheet handling, and related tests).
This commit is contained in:
高物
2026-05-22 03:06:41 +08:00
parent 321e1ea33a
commit ae014ac881
90 changed files with 7078 additions and 3389 deletions
Download Patch File Download Diff File Expand all files Collapse all files

249
server-rs/crates/api-server/src/puzzle/vector_engine.rs
View File

@@ -12,7 +12,7 @@ impl PuzzleImageModel {
}
pub(crate) fn request_model_name(self) -> &'static str {
VECTOR_ENGINE_GPT_IMAGE_2_MODEL
GPT_IMAGE_2_MODEL
}
pub(crate) fn candidate_source_type(self) -> &'static str {
@@ -95,13 +95,26 @@ pub(crate) struct GeneratedPuzzleUiBackgroundResponse {
pub(crate) object_key: String,
}
#[derive(Clone, Debug)]
pub(crate) struct GeneratedPuzzleLevelAssetResponse {
pub(crate) image_src: String,
pub(crate) object_key: String,
}
#[derive(Clone, Debug)]
pub(crate) struct GeneratedPuzzleLevelAssetBundle {
pub(crate) level_scene: GeneratedPuzzleLevelAssetResponse,
pub(crate) ui_spritesheet: GeneratedPuzzleLevelAssetResponse,
pub(crate) level_background: GeneratedPuzzleLevelAssetResponse,
}
pub(crate) fn resolve_puzzle_image_model(value: Option<&str>) -> PuzzleImageModel {
match value.map(str::trim).filter(|value| !value.is_empty()) {
Some(PUZZLE_IMAGE_MODEL_GEMINI_31_FLASH_PREVIEW) => {
tracing::warn!(
requested_model = PUZZLE_IMAGE_MODEL_GEMINI_31_FLASH_PREVIEW,
effective_model = VECTOR_ENGINE_GPT_IMAGE_2_MODEL,
"拼图 nanobanana2 历史选项已回落到 VectorEngine GPT-image-2-all"
effective_model = GPT_IMAGE_2_MODEL,
"拼图 nanobanana2 历史选项已回落到 VectorEngine GPT-image-2"
);
PuzzleImageModel::Gemini31FlashPreview
}
@@ -150,7 +163,7 @@ pub(crate) fn build_puzzle_image_http_client(
reqwest::Client::builder()
.timeout(Duration::from_millis(request_timeout_ms.max(1)))
// 中文注释：VectorEngine 的图片编辑接口是 multipart 请求；强制 HTTP/1.1 可避开部分网关对 HTTP/2 multipart 流的中断兼容问题。
// 中文注释：参考图走 multipart edits；强制 HTTP/1.1 可降低部分网关对长耗时上传流的兼容风险。
.http1_only()
.build()
.map_err(|error| {
@@ -186,6 +199,20 @@ pub(crate) async fn create_puzzle_vector_engine_image_generation(
candidate_count: u32,
reference_image: Option<&PuzzleResolvedReferenceImage>,
) -> Result<PuzzleGeneratedImages, AppError> {
if let Some(reference_image) = reference_image {
return create_puzzle_vector_engine_image_edit(
http_client,
settings,
image_model,
prompt,
negative_prompt,
size,
candidate_count,
reference_image,
)
.await;
}
let request_body = build_puzzle_vector_engine_image_request_body(
image_model,
prompt,
@@ -262,6 +289,15 @@ pub(crate) async fn create_puzzle_vector_engine_image_generation(
return Ok(images);
}
let b64_images = extract_puzzle_b64_images(&payload);
if !b64_images.is_empty() {
return Ok(puzzle_images_from_base64(
format!("vector-engine-{}", current_utc_micros()),
b64_images,
candidate_count,
));
}
Err(
AppError::from_status(StatusCode::BAD_GATEWAY).with_details(json!({
"provider": VECTOR_ENGINE_PROVIDER,
@@ -273,6 +309,7 @@ pub(crate) async fn create_puzzle_vector_engine_image_generation(
pub(crate) async fn create_puzzle_vector_engine_image_edit(
http_client: &reqwest::Client,
settings: &PuzzleVectorEngineSettings,
image_model: PuzzleImageModel,
prompt: &str,
negative_prompt: &str,
size: &str,
@@ -295,7 +332,7 @@ pub(crate) async fn create_puzzle_vector_engine_image_edit(
})?;
let form = reqwest::multipart::Form::new()
.part("image", image_part)
.text("model", PUZZLE_VECTOR_ENGINE_IMAGE_EDIT_MODEL.to_string())
.text("model", image_model.request_model_name().to_string())
.text(
"prompt",
build_puzzle_vector_engine_prompt(prompt, negative_prompt),
@@ -314,16 +351,14 @@ pub(crate) async fn create_puzzle_vector_engine_image_edit(
.send()
.await
.map_err(|error| {
map_puzzle_vector_engine_reqwest_error(
"创建拼图 VectorEngine 图片编辑任务失败",
&request_url,
error,
)
map_puzzle_vector_engine_request_error(format!(
"创建拼图 VectorEngine 图片编辑任务失败：{error}"
))
})?;
let status = response.status();
tracing::info!(
provider = VECTOR_ENGINE_PROVIDER,
image_model = PUZZLE_VECTOR_ENGINE_IMAGE_EDIT_MODEL,
image_model = image_model.request_model_name(),
endpoint = %request_url,
status = status.as_u16(),
prompt_chars = prompt.chars().count(),
@@ -372,6 +407,17 @@ pub(crate) async fn create_puzzle_vector_engine_image_edit(
)
}
pub(crate) fn build_puzzle_downloaded_image_reference(
image: &PuzzleDownloadedImage,
) -> PuzzleResolvedReferenceImage {
PuzzleResolvedReferenceImage {
mime_type: image.mime_type.clone(),
bytes_len: image.bytes.len(),
bytes: image.bytes.clone(),
signed_read_url: None,
}
}
pub(crate) fn build_puzzle_vector_engine_image_request_body(
image_model: PuzzleImageModel,
prompt: &str,
@@ -380,7 +426,7 @@ pub(crate) fn build_puzzle_vector_engine_image_request_body(
candidate_count: u32,
reference_image: Option<&PuzzleResolvedReferenceImage>,
) -> Value {
let mut body = Map::from_iter([
let body = Map::from_iter([
(
"model".to_string(),
Value::String(image_model.request_model_name().to_string()),
@@ -392,20 +438,7 @@ pub(crate) fn build_puzzle_vector_engine_image_request_body(
("n".to_string(), json!(candidate_count.clamp(1, 1))),
("size".to_string(), Value::String(size.to_string())),
]);
if let Some(reference_image) = reference_image {
if let Some(signed_read_url) = reference_image
.signed_read_url
.as_deref()
.map(str::trim)
.filter(|value| !value.is_empty())
{
body.insert("image".to_string(), json!([signed_read_url]));
} else if let Some(reference_data_url) =
build_puzzle_generation_reference_image_data_url(reference_image)
{
body.insert("image".to_string(), json!([reference_data_url]));
}
}
let _ = reference_image;
Value::Object(body)
}
@@ -429,32 +462,6 @@ pub(crate) fn build_puzzle_vector_engine_generation_prompt(
)
}
pub(crate) fn build_puzzle_generation_reference_image_data_url(
image: &PuzzleResolvedReferenceImage,
) -> Option<String> {
let bytes = resize_puzzle_generation_reference_image_bytes(image.bytes.as_slice())
.unwrap_or_else(|| image.bytes.clone());
let mime_type = if bytes.starts_with(b"\x89PNG\r\n\x1A\n") {
"image/png"
} else {
image.mime_type.as_str()
};
Some(format!(
"data:{};base64,{}",
normalize_puzzle_downloaded_image_mime_type(mime_type),
BASE64_STANDARD.encode(bytes)
))
}
pub(crate) fn resize_puzzle_generation_reference_image_bytes(bytes: &[u8]) -> Option<Vec<u8>> {
let image = image::load_from_memory(bytes).ok()?;
let resized = image.resize(1024, 1024, image::imageops::FilterType::Triangle);
let mut cursor = std::io::Cursor::new(Vec::new());
resized.write_to(&mut cursor, ImageFormat::Png).ok()?;
Some(cursor.into_inner())
}
pub(crate) fn has_puzzle_reference_image(reference_image_src: Option<&str>) -> bool {
reference_image_src
.map(str::trim)
@@ -545,11 +552,11 @@ pub(crate) fn has_puzzle_reference_images(
.is_empty()
}
pub(crate) fn should_use_puzzle_reference_image_edit(
pub(crate) fn should_use_puzzle_reference_image_generation(
reference_image_src: Option<&str>,
use_reference_image_edit: bool,
use_reference_image_generation: bool,
) -> bool {
use_reference_image_edit && has_puzzle_reference_image(reference_image_src)
use_reference_image_generation && has_puzzle_reference_image(reference_image_src)
}
pub(crate) fn build_puzzle_vector_engine_prompt(prompt: &str, negative_prompt: &str) -> String {
@@ -1072,6 +1079,57 @@ pub(crate) async fn persist_puzzle_ui_background_image(
})
}
pub(crate) async fn persist_puzzle_level_asset_image(
state: &PuzzleApiState,
owner_user_id: &str,
session_id: &str,
level_name: &str,
task_id: &str,
path_segment: &str,
asset_kind: &str,
slot: &str,
file_stem: &str,
image: PuzzleDownloadedImage,
) -> Result<GeneratedPuzzleLevelAssetResponse, AppError> {
let oss_client = state.oss_client().ok_or_else(|| {
AppError::from_status(StatusCode::SERVICE_UNAVAILABLE).with_details(json!({
"provider": "aliyun-oss",
"reason": "OSS 未完成环境变量配置",
}))
})?;
let http_client = reqwest::Client::new();
let put_result = oss_client
.put_object(
&http_client,
OssPutObjectRequest {
prefix: LegacyAssetPrefix::PuzzleAssets,
path_segments: vec![
sanitize_path_segment(session_id, "session"),
sanitize_path_segment(level_name, "puzzle"),
sanitize_path_segment(path_segment, "level-asset"),
sanitize_path_segment(task_id, "task"),
],
file_name: format!("{file_stem}.{}", image.extension),
content_type: Some(image.mime_type.clone()),
access: OssObjectAccess::Private,
metadata: build_puzzle_level_asset_metadata(
owner_user_id,
session_id,
asset_kind,
slot,
),
body: image.bytes,
},
)
.await
.map_err(map_puzzle_asset_oss_error)?;
Ok(GeneratedPuzzleLevelAssetResponse {
image_src: put_result.legacy_public_path,
object_key: put_result.object_key,
})
}
pub(crate) fn handle_puzzle_asset_spacetime_index_error(
error: SpacetimeClientError,
owner_user_id: &str,
@@ -1126,6 +1184,21 @@ pub(crate) fn build_puzzle_ui_background_asset_metadata(
])
}
pub(crate) fn build_puzzle_level_asset_metadata(
owner_user_id: &str,
session_id: &str,
asset_kind: &str,
slot: &str,
) -> BTreeMap<String, String> {
BTreeMap::from([
("asset_kind".to_string(), asset_kind.to_string()),
("owner_user_id".to_string(), owner_user_id.to_string()),
("entity_kind".to_string(), PUZZLE_ENTITY_KIND.to_string()),
("entity_id".to_string(), session_id.to_string()),
("slot".to_string(), slot.to_string()),
])
}
pub(crate) fn parse_puzzle_json_payload(
raw_text: &str,
fallback_message: &str,
@@ -1331,72 +1404,6 @@ pub(crate) fn map_puzzle_vector_engine_request_error(message: String) -> AppErro
}))
}
pub(crate) fn map_puzzle_vector_engine_reqwest_error(
context: &str,
request_url: &str,
error: reqwest::Error,
) -> AppError {
let message = format!(
"{context}：{}",
normalize_puzzle_reqwest_error_message(&error)
);
let is_timeout = error.is_timeout() || is_puzzle_request_timeout_message(message.as_str());
let is_connect = error.is_connect();
let status = if is_timeout {
StatusCode::GATEWAY_TIMEOUT
} else {
StatusCode::BAD_GATEWAY
};
let source = error.source().map(ToString::to_string).unwrap_or_default();
tracing::warn!(
provider = VECTOR_ENGINE_PROVIDER,
endpoint = %request_url,
timeout = is_timeout,
connect = is_connect,
request = error.is_request(),
body = error.is_body(),
source = %source,
message = %message,
"拼图 VectorEngine 请求发送失败"
);
AppError::from_status(status).with_details(json!({
"provider": VECTOR_ENGINE_PROVIDER,
"message": message,
"reason": resolve_puzzle_vector_engine_request_failure_reason(&error),
"endpoint": request_url,
"timeout": is_timeout,
"connect": is_connect,
"request": error.is_request(),
"body": error.is_body(),
"source": source,
}))
}
pub(crate) fn normalize_puzzle_reqwest_error_message(error: &reqwest::Error) -> String {
error
.to_string()
.split_whitespace()
.collect::<Vec<_>>()
.join(" ")
}
pub(crate) fn resolve_puzzle_vector_engine_request_failure_reason(
error: &reqwest::Error,
) -> &'static str {
if error.is_timeout() {
return "VectorEngine 图片编辑请求超时，请稍后重试或调大 VECTOR_ENGINE_IMAGE_REQUEST_TIMEOUT_MS";
}
if error.is_connect() {
return "无法连接 VectorEngine 图片编辑接口，请检查服务器网络、DNS、防火墙或代理配置";
}
if error.is_body() {
return "发送 VectorEngine 图片编辑 multipart 请求体失败，请重试并检查参考图大小";
}
"VectorEngine 图片编辑请求发送失败，请查看 source 字段中的底层网络错误"
}
pub(crate) fn is_puzzle_request_timeout_message(message: &str) -> bool {
let lower = message.to_ascii_lowercase();
lower.contains("timed out")