ima2-gen 1.1.4 → 1.1.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/config.js CHANGED
@@ -74,6 +74,11 @@ export const config = {
74
74
  },
75
75
  limits: {
76
76
  maxRefB64Bytes: pickInt(env.IMA2_MAX_REF_B64_BYTES, fileCfg.limits?.maxRefB64Bytes, 7 * 1024 * 1024),
77
+ maxMetadataReadB64Bytes: pickInt(
78
+ env.IMA2_MAX_METADATA_READ_B64_BYTES,
79
+ fileCfg.limits?.maxMetadataReadB64Bytes,
80
+ 12 * 1024 * 1024,
81
+ ),
77
82
  maxRefCount: pickInt(env.IMA2_MAX_REF_COUNT, fileCfg.limits?.maxRefCount, 5),
78
83
  maxParallel: pickInt(env.IMA2_MAX_PARALLEL, fileCfg.limits?.maxParallel, 8),
79
84
  graphMaxNodes: pickInt(env.IMA2_GRAPH_MAX_NODES, fileCfg.limits?.graphMaxNodes, 500),
package/lib/db.js CHANGED
@@ -96,12 +96,50 @@ function migrate(database) {
96
96
  );
97
97
  }
98
98
 
99
+ // ── Prompt Library (schema v4) ──
100
+ database.exec(`
101
+ CREATE TABLE IF NOT EXISTS prompt_folders (
102
+ id TEXT PRIMARY KEY,
103
+ parent_id TEXT NOT NULL,
104
+ name TEXT NOT NULL COLLATE NOCASE,
105
+ created_at INTEGER NOT NULL DEFAULT (unixepoch()),
106
+ updated_at INTEGER NOT NULL DEFAULT (unixepoch()),
107
+ UNIQUE(parent_id, name)
108
+ );
109
+
110
+ CREATE TABLE IF NOT EXISTS prompts (
111
+ id TEXT PRIMARY KEY,
112
+ folder_id TEXT NOT NULL DEFAULT '__root__',
113
+ name TEXT NOT NULL,
114
+ text TEXT NOT NULL,
115
+ tags TEXT,
116
+ mode TEXT,
117
+ is_favorite INTEGER NOT NULL DEFAULT 0,
118
+ favorited_at INTEGER,
119
+ created_at INTEGER NOT NULL DEFAULT (unixepoch()),
120
+ updated_at INTEGER NOT NULL DEFAULT (unixepoch()),
121
+ FOREIGN KEY (folder_id) REFERENCES prompt_folders(id) ON DELETE SET DEFAULT
122
+ );
123
+
124
+ CREATE TABLE IF NOT EXISTS gallery_favorites (
125
+ id TEXT PRIMARY KEY,
126
+ browser_id TEXT NOT NULL,
127
+ filename TEXT NOT NULL,
128
+ favorited_at INTEGER NOT NULL DEFAULT (unixepoch()),
129
+ UNIQUE(browser_id, filename)
130
+ );
131
+
132
+ INSERT OR IGNORE INTO prompt_folders (id, parent_id, name) VALUES
133
+ ('__root__', '__root__', '__root__'),
134
+ ('__trash__', '__root__', '__trash__');
135
+ `);
136
+
99
137
  const row = database.prepare("SELECT value FROM _meta WHERE key = 'schema_version'").get();
100
138
  if (!row) {
101
- database.prepare("INSERT INTO _meta (key, value) VALUES ('schema_version', '3')").run();
102
- } else if (row.value !== "3") {
139
+ database.prepare("INSERT INTO _meta (key, value) VALUES ('schema_version', '4')").run();
140
+ } else if (row.value !== "4") {
103
141
  database
104
- .prepare("UPDATE _meta SET value = '3' WHERE key = 'schema_version'")
142
+ .prepare("UPDATE _meta SET value = '4' WHERE key = 'schema_version'")
105
143
  .run();
106
144
  }
107
145
  }
@@ -12,6 +12,19 @@ const PASSTHROUGH_CODES = new Set([
12
12
 
13
13
  const SAFETY_CODES = new Set(["SAFETY_REFUSAL", "MODERATION_REFUSED", "moderation_blocked"]);
14
14
 
15
+ function has4kSize(size) {
16
+ if (typeof size !== "string") return false;
17
+ const [w, h] = size.split("x").map((part) => Number(part));
18
+ return Number.isFinite(w) && Number.isFinite(h) && Math.max(w, h) >= 3840;
19
+ }
20
+
21
+ function diagnosticReasonFrom(err) {
22
+ if (typeof err?.diagnosticReason === "string" && err.diagnosticReason) return err.diagnosticReason;
23
+ if (Number(err?.referenceMismatchCount) > 0) return "reference_mime_mismatch_candidate";
24
+ if (has4kSize(err?.size)) return "experimental_4k_empty_response";
25
+ return null;
26
+ }
27
+
15
28
  export function errorCodeFrom(err) {
16
29
  if (!err) return "UNKNOWN";
17
30
  const upstreamCode = classifyUpstreamErrorCode(err.upstreamCode);
@@ -86,6 +99,17 @@ export function normalizeGenerationFailure(lastErr, options = {}) {
86
99
  if (lastErr.size) err.size = lastErr.size;
87
100
  if (lastErr.quality) err.quality = lastErr.quality;
88
101
  if (lastErr.model) err.model = lastErr.model;
102
+ if (typeof lastErr.eventCount === "number") err.eventCount = lastErr.eventCount;
103
+ if (lastErr.eventTypes) err.eventTypes = lastErr.eventTypes;
104
+ if (typeof lastErr.refsCount === "number") err.refsCount = lastErr.refsCount;
105
+ if (typeof lastErr.inputImageCount === "number") err.inputImageCount = lastErr.inputImageCount;
106
+ if (Array.isArray(lastErr.referenceDiagnostics)) err.referenceDiagnostics = lastErr.referenceDiagnostics;
107
+ if (typeof lastErr.referenceMismatchCount === "number") err.referenceMismatchCount = lastErr.referenceMismatchCount;
108
+ if (lastErr.retryKind) err.retryKind = lastErr.retryKind;
109
+ if (typeof lastErr.referencesDroppedOnRetry === "boolean") err.referencesDroppedOnRetry = lastErr.referencesDroppedOnRetry;
110
+ if (typeof lastErr.developerPromptDroppedOnRetry === "boolean") err.developerPromptDroppedOnRetry = lastErr.developerPromptDroppedOnRetry;
111
+ const diagnosticReason = diagnosticReasonFrom(lastErr);
112
+ if (diagnosticReason) err.diagnosticReason = diagnosticReason;
89
113
  return err;
90
114
  }
91
115
  // Unrecognized errors → UNKNOWN (do not pretend they are safety refusals)
@@ -1,6 +1,7 @@
1
1
  import { mkdir, readFile, readdir, stat } from "fs/promises";
2
2
  import { dirname, join } from "path";
3
3
  import { config } from "../config.js";
4
+ import { readEmbeddedImageMetadataFromFile } from "./imageMetadataStore.js";
4
5
 
5
6
  async function listImageFiles(baseDir) {
6
7
  const out = [];
@@ -28,7 +29,7 @@ export async function listHistoryRows(baseDir = config.storage.generatedDir) {
28
29
  const setRows = await listCardNewsSetRows(baseDir);
29
30
  const rows = await Promise.all(imgs.map(async ({ full, rel, name }) => {
30
31
  const st = await stat(full).catch(() => null);
31
- const meta = await readImageSidecar(full, rel);
32
+ const meta = await readImageMetadata(full, rel);
32
33
  return {
33
34
  filename: rel,
34
35
  url: `/generated/${rel.split("/").map(encodeURIComponent).join("/")}`,
@@ -81,6 +82,18 @@ async function readImageSidecar(full, rel) {
81
82
  return null;
82
83
  }
83
84
 
85
+ async function readImageMetadata(full, rel) {
86
+ const sidecar = await readImageSidecar(full, rel);
87
+ if (sidecar) return sidecar;
88
+ try {
89
+ const embedded = await readEmbeddedImageMetadataFromFile(full);
90
+ return embedded.metadata;
91
+ } catch (e) {
92
+ if (e.code !== "ENOENT") console.warn("[history] embedded metadata read fail:", rel, e.message);
93
+ return null;
94
+ }
95
+ }
96
+
84
97
  async function listCardNewsSetRows(baseDir) {
85
98
  const root = join(baseDir, "cardnews");
86
99
  const entries = await readdir(root, { withFileTypes: true }).catch(() => []);
@@ -0,0 +1,107 @@
1
+ export const IMA2_METADATA_SCHEMA = "ima2.generation.v1";
2
+ export const IMA2_XMP_NAMESPACE = "https://github.com/lidge-jun/ima2-gen/ns/1.0/";
3
+ export const IMA2_XMP_PROPERTY = "GenerationMetadata";
4
+ export const MAX_EMBEDDED_METADATA_CHARS = 64 * 1024;
5
+
6
+ function isPlainObject(value) {
7
+ return value != null && typeof value === "object" && !Array.isArray(value);
8
+ }
9
+
10
+ function stringOrNull(value, max = 4000) {
11
+ if (typeof value !== "string") return null;
12
+ return value.slice(0, max);
13
+ }
14
+
15
+ function numberOrNull(value) {
16
+ return Number.isFinite(value) ? value : null;
17
+ }
18
+
19
+ function xmlEscape(value) {
20
+ return String(value)
21
+ .replace(/&/g, "&")
22
+ .replace(/"/g, """)
23
+ .replace(/</g, "&lt;")
24
+ .replace(/>/g, "&gt;");
25
+ }
26
+
27
+ function xmlUnescape(value) {
28
+ return String(value)
29
+ .replace(/&quot;/g, "\"")
30
+ .replace(/&lt;/g, "<")
31
+ .replace(/&gt;/g, ">")
32
+ .replace(/&amp;/g, "&");
33
+ }
34
+
35
+ export function buildIma2MetadataPayload(meta = {}, context = {}) {
36
+ const payload = {
37
+ schema: IMA2_METADATA_SCHEMA,
38
+ app: "ima2-gen",
39
+ version: stringOrNull(context.version, 80) || null,
40
+ createdAt: numberOrNull(meta.createdAt) || Date.now(),
41
+ kind: stringOrNull(meta.kind, 80),
42
+ prompt: stringOrNull(meta.prompt),
43
+ userPrompt: stringOrNull(meta.userPrompt) || stringOrNull(meta.prompt),
44
+ revisedPrompt: stringOrNull(meta.revisedPrompt),
45
+ promptMode: meta.promptMode === "direct" ? "direct" : "auto",
46
+ quality: stringOrNull(meta.quality, 40),
47
+ size: stringOrNull(meta.size, 40),
48
+ format: stringOrNull(meta.format, 20),
49
+ moderation: stringOrNull(meta.moderation, 40),
50
+ model: stringOrNull(meta.model, 80),
51
+ provider: stringOrNull(meta.provider, 40),
52
+ sessionId: stringOrNull(meta.sessionId, 120),
53
+ nodeId: stringOrNull(meta.nodeId, 120),
54
+ parentNodeId: stringOrNull(meta.parentNodeId, 120),
55
+ clientNodeId: stringOrNull(meta.clientNodeId, 120),
56
+ requestId: stringOrNull(meta.requestId, 160),
57
+ refsCount: Number.isFinite(meta.refsCount) ? meta.refsCount : 0,
58
+ webSearchCalls: Number.isFinite(meta.webSearchCalls) ? meta.webSearchCalls : 0,
59
+ styleSheetApplied: Boolean(meta.styleSheetApplied),
60
+ };
61
+ return Object.fromEntries(Object.entries(payload).filter(([, value]) => value !== undefined));
62
+ }
63
+
64
+ export function normalizeEmbeddedMetadata(value) {
65
+ if (!isPlainObject(value)) return null;
66
+ if (value.schema !== IMA2_METADATA_SCHEMA || value.app !== "ima2-gen") return null;
67
+ return buildIma2MetadataPayload(value, { version: value.version });
68
+ }
69
+
70
+ export function buildIma2Xmp(metadataPayload) {
71
+ const normalized = normalizeEmbeddedMetadata(metadataPayload);
72
+ if (!normalized) {
73
+ const err = new Error("Invalid ima2 metadata payload");
74
+ err.code = "IMAGE_METADATA_INVALID";
75
+ throw err;
76
+ }
77
+ const json = JSON.stringify(normalized);
78
+ if (json.length > MAX_EMBEDDED_METADATA_CHARS) {
79
+ const err = new Error("ima2 metadata payload is too large");
80
+ err.code = "IMAGE_METADATA_TOO_LARGE";
81
+ throw err;
82
+ }
83
+ const escaped = xmlEscape(json);
84
+ return [
85
+ "<?xpacket begin=\"\" id=\"W5M0MpCehiHzreSzNTczkc9d\"?>",
86
+ "<x:xmpmeta xmlns:x=\"adobe:ns:meta/\">",
87
+ "<rdf:RDF xmlns:rdf=\"http://www.w3.org/1999/02/22-rdf-syntax-ns#\">",
88
+ `<rdf:Description xmlns:ima2="${IMA2_XMP_NAMESPACE}" ima2:${IMA2_XMP_PROPERTY}="${escaped}"/>`,
89
+ "</rdf:RDF>",
90
+ "</x:xmpmeta>",
91
+ "<?xpacket end=\"w\"?>",
92
+ ].join("");
93
+ }
94
+
95
+ export function parseIma2Xmp(xmpString) {
96
+ if (typeof xmpString !== "string" || xmpString.length === 0) return null;
97
+ const attrPattern = new RegExp(`ima2:${IMA2_XMP_PROPERTY}="([^"]*)"`);
98
+ const attrMatch = attrPattern.exec(xmpString);
99
+ const raw = attrMatch?.[1];
100
+ if (!raw) return null;
101
+ try {
102
+ const parsed = JSON.parse(xmlUnescape(raw));
103
+ return normalizeEmbeddedMetadata(parsed);
104
+ } catch {
105
+ return null;
106
+ }
107
+ }
@@ -0,0 +1,67 @@
1
+ import { readFile } from "node:fs/promises";
2
+ import sharp from "sharp";
3
+ import { buildIma2MetadataPayload, buildIma2Xmp, parseIma2Xmp } from "./imageMetadata.js";
4
+
5
+ const SUPPORTED_FORMATS = new Set(["png", "jpeg", "jpg", "webp"]);
6
+
7
+ export function normalizeImageMetadataFormat(format) {
8
+ const normalized = String(format || "").toLowerCase();
9
+ if (normalized === "jpg") return "jpeg";
10
+ return normalized;
11
+ }
12
+
13
+ export function isSupportedMetadataFormat(format) {
14
+ return SUPPORTED_FORMATS.has(String(format || "").toLowerCase());
15
+ }
16
+
17
+ export async function embedImageMetadata(buffer, format, metadata, context = {}) {
18
+ const normalizedFormat = normalizeImageMetadataFormat(format);
19
+ if (!isSupportedMetadataFormat(normalizedFormat)) {
20
+ const err = new Error(`Unsupported image metadata format: ${format}`);
21
+ err.code = "IMAGE_METADATA_UNSUPPORTED_FORMAT";
22
+ throw err;
23
+ }
24
+ const payload = buildIma2MetadataPayload(metadata, context);
25
+ const xmp = buildIma2Xmp(payload);
26
+ const next = await sharp(buffer)
27
+ .toFormat(normalizedFormat)
28
+ .withXmp(xmp)
29
+ .toBuffer();
30
+ return { buffer: next, embedded: true, metadata: payload };
31
+ }
32
+
33
+ export async function embedImageMetadataBestEffort(buffer, format, metadata, context = {}) {
34
+ try {
35
+ return await embedImageMetadata(buffer, format, metadata, context);
36
+ } catch (error) {
37
+ return {
38
+ buffer,
39
+ embedded: false,
40
+ warning: error?.message || "metadata embedding failed",
41
+ code: error?.code || "IMAGE_METADATA_EMBED_FAILED",
42
+ };
43
+ }
44
+ }
45
+
46
+ export async function readEmbeddedImageMetadata(buffer) {
47
+ const meta = await sharp(buffer, { failOn: "none" }).metadata();
48
+ const xmpString = meta.xmpAsString || (meta.xmp ? meta.xmp.toString("utf8") : "");
49
+ const xmp = parseIma2Xmp(xmpString);
50
+ if (xmp) return { metadata: xmp, source: "xmp", warnings: [] };
51
+
52
+ for (const comment of meta.comments || []) {
53
+ const text = comment?.text || "";
54
+ const parsed = parseIma2Xmp(text);
55
+ if (parsed) return { metadata: parsed, source: "png-comment", warnings: [] };
56
+ }
57
+
58
+ return {
59
+ metadata: null,
60
+ source: null,
61
+ warnings: ["No ima2 metadata found in this image."],
62
+ };
63
+ }
64
+
65
+ export async function readEmbeddedImageMetadataFromFile(path) {
66
+ return readEmbeddedImageMetadata(await readFile(path));
67
+ }
package/lib/nodeStore.js CHANGED
@@ -2,6 +2,7 @@ import { writeFile, readFile, access, mkdir } from "fs/promises";
2
2
  import { join, resolve, sep } from "path";
3
3
  import { randomBytes } from "crypto";
4
4
  import { config } from "../config.js";
5
+ import { embedImageMetadataBestEffort } from "./imageMetadataStore.js";
5
6
 
6
7
  export function newNodeId() {
7
8
  return "n_" + randomBytes(config.ids.nodeHexBytes).toString("hex");
@@ -11,7 +12,18 @@ export async function saveNode(rootDir, { nodeId, b64, meta, ext = "png", genera
11
12
  void rootDir;
12
13
  const filename = `${nodeId}.${ext}`;
13
14
  await mkdir(generatedDir, { recursive: true });
14
- await writeFile(join(generatedDir, filename), Buffer.from(b64, "base64"));
15
+ const imageMeta = {
16
+ ...meta,
17
+ kind: meta?.kind || "node",
18
+ nodeId: meta?.nodeId || nodeId,
19
+ format: meta?.format || ext,
20
+ };
21
+ const rawBuffer = Buffer.from(b64, "base64");
22
+ const embedded = await embedImageMetadataBestEffort(rawBuffer, ext, imageMeta);
23
+ if (!embedded.embedded) {
24
+ console.warn("[nodeStore] metadata embed skipped:", embedded.warning);
25
+ }
26
+ await writeFile(join(generatedDir, filename), embedded.buffer);
15
27
  await writeFile(join(generatedDir, filename + ".json"), JSON.stringify(meta, null, 2));
16
28
  return { filename };
17
29
  }
package/lib/oauthProxy.js CHANGED
@@ -3,6 +3,7 @@ import { config } from "../config.js";
3
3
  import { logEvent } from "./logger.js";
4
4
  import { classifyUpstreamError, classifyUpstreamErrorCode } from "./errorClassify.js";
5
5
  import { compressReferenceB64ForOAuth } from "./referenceImageCompress.js";
6
+ import { detectImageMimeFromB64, safeReferenceDiagnostics } from "./refs.js";
6
7
 
7
8
  const RESEARCH_SUFFIX = config.oauth.researchSuffix;
8
9
 
@@ -41,6 +42,51 @@ export function buildEditResearchTextPrompt(userPrompt, mode) {
41
42
  return `Edit this image: ${userPrompt}${RESEARCH_SUFFIX}${AUTO_PROMPT_FIDELITY_SUFFIX}`;
42
43
  }
43
44
 
45
+ function summarizeEventTypes(eventTypes = {}) {
46
+ const entries = Object.entries(eventTypes || {});
47
+ const countFor = (needle) =>
48
+ entries.reduce((sum, [key, value]) => sum + (key.includes(needle) && Number.isFinite(value) ? value : 0), 0);
49
+ return {
50
+ eventTypeCount: entries.length,
51
+ eventTypeKeys: entries.slice(0, 12).map(([key]) => key).join(","),
52
+ imageEventCount: countFor("image"),
53
+ partialEventCount: countFor("partial"),
54
+ completedEventCount: countFor("completed"),
55
+ };
56
+ }
57
+
58
+ function supportedImageMime(mime) {
59
+ return mime === "image/png" || mime === "image/jpeg" || mime === "image/webp";
60
+ }
61
+
62
+ function normalizeReferenceForOAuth(ref, index) {
63
+ const b64 = typeof ref === "string" ? ref : ref?.b64;
64
+ const declaredMime = typeof ref === "object" && ref ? ref.declaredMime || null : null;
65
+ const detectedMime = typeof ref === "object" && ref
66
+ ? ref.detectedMime || detectImageMimeFromB64(b64)
67
+ : detectImageMimeFromB64(b64);
68
+ const warnings = Array.isArray(ref?.warnings) ? [...ref.warnings] : [];
69
+ if (declaredMime && detectedMime && declaredMime !== detectedMime && !warnings.includes("mime_mismatch")) {
70
+ warnings.push("mime_mismatch");
71
+ }
72
+ const requestMime = supportedImageMime(detectedMime)
73
+ ? detectedMime
74
+ : supportedImageMime(declaredMime)
75
+ ? declaredMime
76
+ : "image/png";
77
+ return {
78
+ index,
79
+ b64,
80
+ declaredMime,
81
+ detectedMime,
82
+ requestMime,
83
+ b64Chars: typeof b64 === "string" ? b64.length : 0,
84
+ approxBytes: Number.isFinite(ref?.approxBytes) ? ref.approxBytes : null,
85
+ source: ref?.source || (declaredMime ? "dataUrl" : "rawBase64"),
86
+ warnings,
87
+ };
88
+ }
89
+
44
90
  function getOAuthUrl(ctx = {}) {
45
91
  return ctx.oauthUrl || `http://127.0.0.1:${config.oauth.proxyPort}`;
46
92
  }
@@ -293,16 +339,29 @@ export async function generateViaOAuth(
293
339
  ];
294
340
 
295
341
  const textPrompt = buildUserTextPrompt(prompt, mode);
296
- const userContent = references.length
342
+ const referenceInputs = references.map(normalizeReferenceForOAuth);
343
+ const referenceDiagnostics = safeReferenceDiagnostics(referenceInputs);
344
+ const referenceMismatchCount = referenceDiagnostics.filter((ref) => ref.warnings.includes("mime_mismatch")).length;
345
+ const userContent = referenceInputs.length
297
346
  ? [
298
- ...references.map((b64) => ({
347
+ ...referenceInputs.map(({ b64, requestMime }) => ({
299
348
  type: "input_image",
300
- image_url: `data:image/png;base64,${b64}`,
349
+ image_url: `data:${requestMime};base64,${b64}`,
301
350
  })),
302
351
  { type: "input_text", text: textPrompt },
303
352
  ]
304
353
  : textPrompt;
305
354
 
355
+ if (referenceInputs.length > 0) {
356
+ logEvent("oauth", "reference_diagnostics", {
357
+ requestId,
358
+ refsCount: referenceInputs.length,
359
+ referenceMismatchCount,
360
+ refDetectedMimes: [...new Set(referenceDiagnostics.map((ref) => ref.detectedMime).filter(Boolean))].join(","),
361
+ refDeclaredMimes: [...new Set(referenceDiagnostics.map((ref) => ref.declaredMime).filter(Boolean))].join(","),
362
+ });
363
+ }
364
+
306
365
  const res = await fetchOAuth(`${oauthUrl}/v1/responses`, {
307
366
  method: "POST",
308
367
  headers: { "Content-Type": "application/json", Accept: "text/event-stream" },
@@ -357,10 +416,20 @@ export async function generateViaOAuth(
357
416
  scope: "oauth",
358
417
  onPartialImage: options.onPartialImage,
359
418
  });
360
- logEvent("oauth", "stream_end", { requestId, events: eventCount, hasImage: !!imageB64, eventTypes });
419
+ logEvent("oauth", "stream_end", {
420
+ requestId,
421
+ events: eventCount,
422
+ hasImage: !!imageB64,
423
+ ...summarizeEventTypes(eventTypes),
424
+ });
361
425
 
362
426
  if (!imageB64) {
363
- logEvent("oauth", "retry_json", { requestId });
427
+ logEvent("oauth", "retry_json", {
428
+ requestId,
429
+ retryKind: "prompt_only",
430
+ referencesDroppedOnRetry: referenceInputs.length > 0,
431
+ developerPromptDroppedOnRetry: true,
432
+ });
364
433
  const retryRes = await fetchOAuth(`${oauthUrl}/v1/responses`, {
365
434
  method: "POST",
366
435
  headers: { "Content-Type": "application/json" },
@@ -376,9 +445,23 @@ export async function generateViaOAuth(
376
445
  const json = await retryRes.json();
377
446
  for (const item of json.output || []) {
378
447
  if (item.type === "image_generation_call" && item.result) {
379
- logEvent("oauth", "retry_image", { requestId, imageChars: item.result.length });
448
+ logEvent("oauth", "retry_image", {
449
+ requestId,
450
+ imageChars: item.result.length,
451
+ retryKind: "prompt_only",
452
+ referencesDroppedOnRetry: referenceInputs.length > 0,
453
+ });
380
454
  const retryRevised = typeof item.revised_prompt === "string" ? item.revised_prompt : null;
381
- return { b64: item.result, usage: json.usage, webSearchCalls, revisedPrompt: retryRevised };
455
+ return {
456
+ b64: item.result,
457
+ usage: json.usage,
458
+ webSearchCalls,
459
+ revisedPrompt: retryRevised,
460
+ retryKind: "prompt_only",
461
+ referencesDroppedOnRetry: referenceInputs.length > 0,
462
+ developerPromptDroppedOnRetry: true,
463
+ initialEventCount: eventCount,
464
+ };
382
465
  }
383
466
  }
384
467
  } else {
@@ -397,6 +480,13 @@ export async function generateViaOAuth(
397
480
  emptyErr.size = size;
398
481
  emptyErr.quality = quality;
399
482
  emptyErr.model = model;
483
+ emptyErr.refsCount = referenceInputs.length;
484
+ emptyErr.inputImageCount = referenceInputs.length;
485
+ emptyErr.referenceDiagnostics = referenceDiagnostics;
486
+ emptyErr.referenceMismatchCount = referenceMismatchCount;
487
+ emptyErr.retryKind = "prompt_only";
488
+ emptyErr.referencesDroppedOnRetry = referenceInputs.length > 0;
489
+ emptyErr.developerPromptDroppedOnRetry = true;
400
490
  throw emptyErr;
401
491
  }
402
492
 
@@ -417,8 +507,8 @@ export async function editViaOAuth(prompt, imageB64, quality, size, moderation =
417
507
  });
418
508
  const references = Array.isArray(options.references) ? options.references : [];
419
509
  const referenceImagesForRequest = await Promise.all(
420
- references.map((b64) =>
421
- compressReferenceB64ForOAuth(b64, {
510
+ references.map((ref) =>
511
+ compressReferenceB64ForOAuth(typeof ref === "string" ? ref : ref?.b64, {
422
512
  maxB64Bytes: ctx.config?.limits?.maxRefB64Bytes,
423
513
  force: true,
424
514
  }),
@@ -486,11 +576,25 @@ export async function editViaOAuth(prompt, imageB64, quality, size, moderation =
486
576
 
487
577
  if (requestId) setJobPhase(requestId, "streaming");
488
578
 
489
- const { imageB64: resultB64, usage, revisedPrompt, webSearchCalls } = await readImageStream(res, {
579
+ const { imageB64: resultB64, usage, revisedPrompt, webSearchCalls, eventCount, eventTypes } = await readImageStream(res, {
490
580
  scope: "oauth-edit",
491
581
  requestId,
492
582
  });
493
- logEvent("oauth-edit", "stream_end", { requestId, hasImage: !!resultB64 });
583
+ logEvent("oauth-edit", "stream_end", {
584
+ requestId,
585
+ events: eventCount,
586
+ hasImage: !!resultB64,
587
+ ...summarizeEventTypes(eventTypes),
588
+ });
494
589
  if (resultB64) return { b64: resultB64, usage, revisedPrompt, webSearchCalls };
495
- throw new Error("No image data received from OAuth edit");
590
+ const emptyErr = new Error("No image data received from OAuth edit");
591
+ emptyErr.eventCount = eventCount;
592
+ emptyErr.eventTypes = eventTypes;
593
+ emptyErr.size = size;
594
+ emptyErr.quality = quality;
595
+ emptyErr.model = model;
596
+ emptyErr.refsCount = references.length;
597
+ emptyErr.inputImageCount = 1 + references.length;
598
+ emptyErr.parentImagePresent = true;
599
+ throw emptyErr;
496
600
  }
package/lib/refs.js CHANGED
@@ -4,6 +4,51 @@
4
4
  import { config } from "../config.js";
5
5
 
6
6
  const BASE64_RE = /^[A-Za-z0-9+/]+=*$/;
7
+ const DATA_URL_RE = /^data:([^;,]+);base64,/i;
8
+
9
+ function approxBase64Bytes(b64) {
10
+ try {
11
+ return Buffer.from(b64, "base64").length;
12
+ } catch {
13
+ return Math.floor((b64.length * 3) / 4);
14
+ }
15
+ }
16
+
17
+ export function detectImageMimeFromB64(b64) {
18
+ let buf;
19
+ try {
20
+ buf = Buffer.from(b64, "base64");
21
+ } catch {
22
+ return null;
23
+ }
24
+ if (buf.length >= 4 && buf[0] === 0x89 && buf[1] === 0x50 && buf[2] === 0x4e && buf[3] === 0x47) {
25
+ return "image/png";
26
+ }
27
+ if (buf.length >= 3 && buf[0] === 0xff && buf[1] === 0xd8 && buf[2] === 0xff) {
28
+ return "image/jpeg";
29
+ }
30
+ if (
31
+ buf.length >= 12 &&
32
+ buf.toString("ascii", 0, 4) === "RIFF" &&
33
+ buf.toString("ascii", 8, 12) === "WEBP"
34
+ ) {
35
+ return "image/webp";
36
+ }
37
+ return null;
38
+ }
39
+
40
+ export function safeReferenceDiagnostics(refDetails = []) {
41
+ if (!Array.isArray(refDetails)) return [];
42
+ return refDetails.map((ref) => ({
43
+ index: ref.index,
44
+ declaredMime: ref.declaredMime || null,
45
+ detectedMime: ref.detectedMime || null,
46
+ b64Chars: ref.b64Chars,
47
+ approxBytes: ref.approxBytes,
48
+ source: ref.source,
49
+ warnings: ref.warnings || [],
50
+ }));
51
+ }
7
52
 
8
53
  export function validateAndNormalizeRefs(references, {
9
54
  maxCount = config.limits.maxRefCount,
@@ -16,12 +61,15 @@ export function validateAndNormalizeRefs(references, {
16
61
  return { error: `references may not exceed ${maxCount} items`, code: "REF_TOO_MANY" };
17
62
  }
18
63
  const out = [];
64
+ const refDetails = [];
19
65
  for (let i = 0; i < references.length; i++) {
20
66
  const r = references[i];
21
67
  if (typeof r !== "string") {
22
68
  return { error: `references[${i}] must be a string`, code: "REF_NOT_STRING" };
23
69
  }
24
- const b64 = r.replace(/^data:[^;]+;base64,/, "");
70
+ const dataUrlMatch = r.match(DATA_URL_RE);
71
+ const declaredMime = dataUrlMatch?.[1]?.toLowerCase() || null;
72
+ const b64 = r.replace(DATA_URL_RE, "");
25
73
  if (!b64) return { error: `references[${i}] is empty`, code: "REF_EMPTY" };
26
74
  if (b64.length > maxB64Bytes) {
27
75
  return { error: `references[${i}] exceeds ${maxB64Bytes} bytes`, code: "REF_TOO_LARGE" };
@@ -29,7 +77,22 @@ export function validateAndNormalizeRefs(references, {
29
77
  if (!BASE64_RE.test(b64)) {
30
78
  return { error: `references[${i}] is not valid base64`, code: "REF_NOT_BASE64" };
31
79
  }
80
+ const detectedMime = detectImageMimeFromB64(b64);
81
+ const warnings = [];
82
+ if (declaredMime && detectedMime && declaredMime !== detectedMime) {
83
+ warnings.push("mime_mismatch");
84
+ }
32
85
  out.push(b64);
86
+ refDetails.push({
87
+ index: i,
88
+ b64,
89
+ declaredMime,
90
+ detectedMime,
91
+ b64Chars: b64.length,
92
+ approxBytes: approxBase64Bytes(b64),
93
+ source: declaredMime ? "dataUrl" : "rawBase64",
94
+ warnings,
95
+ });
33
96
  }
34
- return { refs: out };
97
+ return { refs: out, refDetails, referenceDiagnostics: safeReferenceDiagnostics(refDetails) };
35
98
  }
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "ima2-gen",
3
- "version": "1.1.4",
3
+ "version": "1.1.5",
4
4
  "description": "Local OAuth image generation studio with classic and node workflows",
5
5
  "type": "module",
6
6
  "bin": {