316 lines
16 KiB
Dart
316 lines
16 KiB
Dart
import 'dart:convert';
|
||
import 'dart:io';
|
||
|
||
import 'package:archive/archive_io.dart';
|
||
import 'package:crypto/crypto.dart';
|
||
import 'package:flutter_test/flutter_test.dart';
|
||
import 'package:jj_mkvmaker/app/ai_local.dart';
|
||
import 'package:jj_mkvmaker/app/settings.dart';
|
||
import 'package:jj_mkvmaker/core/ai_catalog.dart';
|
||
import 'package:jj_mkvmaker/core/sd_cli.dart';
|
||
import 'package:jj_mkvmaker/services/image_ai.dart';
|
||
import 'package:jj_mkvmaker/services/secret_store.dart';
|
||
import 'package:path/path.dart' as p;
|
||
|
||
/// 1×1 PNG
|
||
final _png = base64Decode('iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8BQDwAEhQGAhKmMIQAAAABJRU5ErkJggg==');
|
||
|
||
void main() {
|
||
group('123: sd-cli 인수 · 진행 읽기', () {
|
||
const m = SdModelPaths(model: 'm.safetensors', loraDir: 'lora', loraName: 'lcm-lora-sdv1-5', taesd: 'taesd.safetensors');
|
||
|
||
test('글 → 그림: LCM-LoRA · f16 · TAESD · 장치 · 여러 장', () {
|
||
final a = sdCliArgs(const ImageGenRequest(prompt: 'a cat', negative: 'blurry', seed: 7, count: 2), m,
|
||
out: 'o_%d.png', backend: 'vulkan0', threads: 8);
|
||
expect(a.join(' '), contains('-m m.safetensors --type f16 --lora-model-dir lora --taesd taesd.safetensors'));
|
||
expect(a[a.indexOf('-p') + 1], 'a cat<lora:lcm-lora-sdv1-5:1>');
|
||
expect(a[a.indexOf('-n') + 1], 'blurry');
|
||
for (final pair in [
|
||
['--sampling-method', 'lcm'],
|
||
['--seed', '7'],
|
||
['-b', '2'],
|
||
['--backend', 'vulkan0'],
|
||
['-t', '8'],
|
||
['-o', 'o_%d.png'],
|
||
['--cfg-scale', '1.0'],
|
||
]) {
|
||
expect(a[a.indexOf(pair[0]) + 1], pair[1], reason: pair[0]);
|
||
}
|
||
expect(a, isNot(contains('-i')));
|
||
});
|
||
|
||
test('그림 → 그림 · LoRA 없이 · 빼고 싶은 것 없으면 -n 없음', () {
|
||
final a = sdCliArgs(const ImageGenRequest(prompt: 'x', initImage: 'in.png', strength: 0.4), const SdModelPaths(model: 'm'),
|
||
out: 'o.png');
|
||
expect(a[a.indexOf('-i') + 1], 'in.png');
|
||
expect(a[a.indexOf('--strength') + 1], '0.4');
|
||
expect(a[a.indexOf('-p') + 1], 'x');
|
||
expect(a, isNot(contains('-n')));
|
||
expect(a, isNot(contains('--sampling-method')));
|
||
});
|
||
|
||
test('해상도 올리기 인수 (121)', () {
|
||
expect(sdUpscaleArgs(model: 'esr.pth', input: 'a.png', out: 'b.png', repeats: 2).join(' '),
|
||
'-M upscale --upscale-model esr.pth -i a.png --upscale-repeats 2 -o b.png');
|
||
});
|
||
|
||
test('진행: 읽기 · 그리기 (단계당 초) · 디코드 · 끝', () {
|
||
final load = parseSdProgress(' |####################### | 90/196 - 628.27MB/s\x1b[K')!;
|
||
expect((load.phase, load.step, load.steps), ('load', 90, 196));
|
||
expect(load.overall, closeTo(0.1 * 90 / 196, 1e-9));
|
||
final s = parseSdProgress(' |=========================> | 2/4 - 25.11s/it')!;
|
||
expect((s.phase, s.step, s.steps, s.secondsPerStep), ('sample', 2, 4, 25.11));
|
||
expect(s.overall, closeTo(0.5, 1e-9));
|
||
expect(parseSdProgress('|===> | 1/4 - 2.00it/s')!.secondsPerStep, 0.5);
|
||
expect(parseSdProgress('[I] decode_first_stage completed, taking 2.56s')!.phase, 'decode');
|
||
expect(parseSdProgress('[I] generate_image completed in 49.03s')!.overall, 1);
|
||
expect(parseSdProgress('[I] loading model from x'), isNull);
|
||
});
|
||
|
||
test('장치 목록 읽기', () {
|
||
expect(parseSdDevices('ggml_vulkan: 1 = ...\nVulkan0\tNVIDIA GeForce MX150\nVulkan1\tIntel UHD\nCPU\tIntel i7\n'), [
|
||
('vulkan0', 'NVIDIA GeForce MX150'),
|
||
('vulkan1', 'Intel UHD'),
|
||
('cpu', 'Intel i7'),
|
||
]);
|
||
});
|
||
|
||
test('다시 만들 수 있게 남기는 json (시드 · 모델 · 프롬프트)', () {
|
||
const r = ImageGenRequest(prompt: '고양이', negative: 'x', width: 768, height: 512, steps: 6, seed: -1, initImage: 'a.png', strength: 0.3);
|
||
final back = ImageGenRequest.fromJson(jsonDecode(r.encode(engine: '이 기기', actualSeed: 42)) as Map<String, Object?>);
|
||
expect((back.prompt, back.negative, back.width, back.height, back.steps, back.seed, back.initImage, back.strength),
|
||
('고양이', 'x', 768, 512, 6, 42, 'a.png', 0.3));
|
||
expect(resolveSeed(5), 5);
|
||
expect(resolveSeed(-1), inInclusiveRange(0, (1 << 31) - 1));
|
||
});
|
||
});
|
||
|
||
group('123: 받는 파일 목록', () {
|
||
test('공식 배포처 · SHA-256 고정 · 비상업 라이선스 없음 · CUDA 는 NVIDIA 때만', () {
|
||
for (final f in aiCatalog) {
|
||
expect(f.url, startsWith('https://'), reason: f.id);
|
||
expect(RegExp(r'^[0-9a-f]{64}$').hasMatch(f.sha256), isTrue, reason: f.id);
|
||
expect(f.size, greaterThan(0), reason: f.id);
|
||
expect(f.license.toUpperCase(), isNot(contains('NC')), reason: f.id);
|
||
expect(f.licenseUrl, startsWith('https://'));
|
||
}
|
||
expect(aiFile('sd15').terms, contains('OpenRAIL-M'));
|
||
expect(aiFile('engine-cuda').needs, 'nvidia');
|
||
expect(aiFile('engine-cudart').needs, 'nvidia');
|
||
expect(aiFile('engine-vulkan').needs, isNull);
|
||
expect(aiLicenseTexts().map((e) => e.$1), contains('stable-diffusion.cpp'));
|
||
expect(aiLicenseTexts().firstWhere((e) => e.$1 == 'Stable Diffusion v1.5').$2, contains('부속서 A'));
|
||
});
|
||
});
|
||
|
||
group('123: 받기 (SHA-256 확인 · 풀기)', () {
|
||
late Directory tmp;
|
||
late HttpServer server;
|
||
final files = <String, List<int>>{};
|
||
setUp(() async {
|
||
tmp = Directory.systemTemp.createTempSync('jj_ai_store_');
|
||
server = await HttpServer.bind(InternetAddress.loopbackIPv4, 0);
|
||
server.listen((req) async {
|
||
final b = files[req.uri.path];
|
||
if (b == null) {
|
||
req.response.statusCode = 404;
|
||
} else {
|
||
req.response.add(b);
|
||
}
|
||
await req.response.close();
|
||
});
|
||
});
|
||
tearDown(() async {
|
||
await server.close(force: true);
|
||
tmp.deleteSync(recursive: true);
|
||
});
|
||
|
||
AiFile file(String id, String path, List<int> body, {String kind = 'model', String? sha}) {
|
||
files[path] = body;
|
||
return AiFile(
|
||
id: id,
|
||
name: id,
|
||
url: 'http://127.0.0.1:${server.port}$path',
|
||
sha256: sha ?? sha256.convert(body).toString(),
|
||
size: body.length,
|
||
kind: kind,
|
||
license: 'MIT',
|
||
licenseUrl: 'https://x');
|
||
}
|
||
|
||
test('맞는 파일은 받아 둠 · SHA-256 이 다르면 지우고 알림 · 지우기', () async {
|
||
final store = AiStore(tmp.path);
|
||
final good = file('m1', '/m1.safetensors', List.filled(5000, 7));
|
||
await store.installFile(good);
|
||
expect(store.isFileInstalled(good), isTrue);
|
||
expect(File(store.pathOf(good)).lengthSync(), 5000);
|
||
final bad = file('m2', '/m2.safetensors', [1, 2, 3], sha: '0' * 64);
|
||
await expectLater(store.installFile(bad), throwsA(isA<ImageAiException>()));
|
||
expect(store.isFileInstalled(bad), isFalse);
|
||
expect(tmp.listSync(recursive: true).whereType<File>().where((f) => f.path.contains('.part')), isEmpty);
|
||
});
|
||
|
||
test('실행 파일 zip 은 풀어 둔다', () async {
|
||
final store = AiStore(tmp.path);
|
||
final arc = Archive()..addFile(ArchiveFile.bytes('sd-cli.exe', utf8.encode('fake')));
|
||
final zip = ZipEncoder().encode(arc);
|
||
final eng = file('engine-x', '/e.zip', zip, kind: 'engine');
|
||
await store.installFile(eng);
|
||
expect(File(p.join(store.pathOf(eng), 'sd-cli.exe')).existsSync(), isTrue);
|
||
});
|
||
});
|
||
|
||
group('123: 사용자가 추가한 서버 · 서비스', () {
|
||
late HttpServer server;
|
||
final seen = <String>[];
|
||
Map<String, Object?>? lastBody;
|
||
setUp(() async {
|
||
seen.clear();
|
||
server = await HttpServer.bind(InternetAddress.loopbackIPv4, 0);
|
||
var polls = 0;
|
||
server.listen((req) async {
|
||
seen.add('${req.method} ${req.uri.path} ${req.headers.value('authorization') ?? ''}');
|
||
final text = await utf8.decoder.bind(req).join();
|
||
if (text.isNotEmpty && req.headers.contentType?.mimeType == 'application/json') lastBody = jsonDecode(text) as Map<String, Object?>;
|
||
final res = req.response..headers.contentType = ContentType.json;
|
||
switch (req.uri.path) {
|
||
case '/sdapi/v1/txt2img' || '/sdapi/v1/img2img':
|
||
res.write(jsonEncode({'images': [base64Encode(_png), base64Encode(_png)]}));
|
||
case '/v1/images/generations':
|
||
res.write(jsonEncode({'data': [{'b64_json': base64Encode(_png)}]}));
|
||
case '/prompt':
|
||
res.write(jsonEncode({'prompt_id': 'p1'}));
|
||
case '/history/p1':
|
||
polls++;
|
||
res.write(polls < 2
|
||
? '{}'
|
||
: jsonEncode({'p1': {'outputs': {'9': {'images': [{'filename': 'a.png', 'subfolder': '', 'type': 'output'}]}}}}));
|
||
case '/view':
|
||
res.headers.contentType = ContentType('image', 'png');
|
||
res.add(_png);
|
||
default:
|
||
res.statusCode = 404;
|
||
}
|
||
await res.close();
|
||
});
|
||
});
|
||
tearDown(() => server.close(force: true));
|
||
|
||
Future<List<GeneratedImage>> run(AiService s, ImageGenRequest r) async {
|
||
final dir = Directory.systemTemp.createTempSync('jj_ai_remote_');
|
||
addTearDown(() => dir.deleteSync(recursive: true));
|
||
final made = await RemoteImageEngine(s).generate(r, outDir: dir.path, baseName: 'ai');
|
||
for (final g in made) {
|
||
expect(File(g.path).readAsBytesSync(), _png);
|
||
}
|
||
return made;
|
||
}
|
||
|
||
test('Automatic1111: 글 → 그림 · 시드 · 아이디:비밀번호는 Basic', () async {
|
||
final s = AiService(id: 'a', name: 'GPU PC', url: 'http://127.0.0.1:${server.port}/', kind: 'a1111', apiKey: 'u:p');
|
||
final made = await run(s, const ImageGenRequest(prompt: 'cat', seed: 10, count: 2));
|
||
expect(made.map((g) => g.seed), [10, 11]);
|
||
expect(seen.single, startsWith('POST /sdapi/v1/txt2img Basic '));
|
||
expect(lastBody!['seed'], 10);
|
||
expect(lastBody!['batch_size'], 2);
|
||
expect(RemoteImageEngine(s).leavesDevice, isTrue);
|
||
});
|
||
|
||
test('OpenAI 형식: Bearer 키 · 크기', () async {
|
||
final s = AiService(id: 'o', name: 'svc', url: 'http://127.0.0.1:${server.port}', kind: 'openai', model: 'm', apiKey: 'sk-test');
|
||
await run(s, const ImageGenRequest(prompt: 'cat', width: 512, height: 768));
|
||
expect(seen.single, 'POST /v1/images/generations Bearer sk-test');
|
||
expect(lastBody!['size'], '512x768');
|
||
expect(lastBody!['model'], 'm');
|
||
});
|
||
|
||
test('ComfyUI: 작업 흐름을 보내고 끝날 때까지 기다려 결과를 받는다', () async {
|
||
final s = AiService(id: 'c', name: 'comfy', url: 'http://127.0.0.1:${server.port}', kind: 'comfy', model: 'sd15.safetensors');
|
||
final made = await run(s, const ImageGenRequest(prompt: 'cat', seed: 3));
|
||
expect(made, hasLength(1));
|
||
final flow = lastBody!['prompt'] as Map;
|
||
expect((flow['4'] as Map)['inputs'], {'ckpt_name': 'sd15.safetensors'});
|
||
expect(((flow['3'] as Map)['inputs'] as Map)['seed'], 3);
|
||
expect(seen.where((x) => x.startsWith('GET /history')).length, greaterThanOrEqualTo(2));
|
||
});
|
||
|
||
test('설정 파일에는 API 키를 쓰지 않고 안전 저장소에 (124)', () async {
|
||
final dir = Directory.systemTemp.createTempSync('jj_ai_key_');
|
||
addTearDown(() => dir.deleteSync(recursive: true));
|
||
final secrets = MemorySecretStore();
|
||
final st = SettingsStore(p.join(dir.path, 'settings.json'), secrets);
|
||
final s = await st.load();
|
||
s.aiServices = [const AiService(id: 'x', name: 'n', url: 'http://h', kind: 'a1111', apiKey: 'sk-secret')];
|
||
await st.save(s);
|
||
expect(File(p.join(dir.path, 'settings.json')).readAsStringSync(), isNot(contains('sk-secret')));
|
||
expect(secrets.values['ai.key.x'], 'sk-secret');
|
||
final again = await SettingsStore(p.join(dir.path, 'settings.json'), secrets).load();
|
||
expect(again.aiServices.single.apiKey, 'sk-secret');
|
||
s.aiServices = [];
|
||
await st.save(s);
|
||
expect(secrets.values.containsKey('ai.key.x'), isFalse, reason: '지운 서비스의 키는 저장소에서도 지움');
|
||
});
|
||
});
|
||
|
||
group('123: 기기 안 엔진 (가짜 sd-cli 로)', () {
|
||
test('진행을 알리고, 여러 장을 시드와 함께 돌려주고, 옆에 json 을 남긴다', () async {
|
||
if (!Platform.isWindows) return markTestSkipped('Windows 전용 (가짜 sd-cli 가 .cmd)');
|
||
final dir = Directory.systemTemp.createTempSync('jj_ai_local_');
|
||
addTearDown(() => dir.deleteSync(recursive: true));
|
||
final png = File(p.join(dir.path, 'src.png'))..writeAsBytesSync(_png);
|
||
// 받은 인수의 -o (예: ...\ai_X_%d.png) 에 0 · 1 번을 만든다
|
||
final fake = File(p.join(dir.path, 'fake-sd.cmd'))
|
||
..writeAsStringSync('@echo off\r\n'
|
||
'echo ^|=====^> ^| 1/2 - 3.00s/it\r\n'
|
||
'echo ^|==========^| 2/2 - 3.00s/it\r\n'
|
||
'echo [I] decode_first_stage completed\r\n'
|
||
':loop\r\nif "%~1"=="" goto done\r\nif "%~1"=="-o" (set OUT=%~2)\r\nshift\r\ngoto loop\r\n:done\r\n'
|
||
'setlocal enabledelayedexpansion\r\nset A=!OUT:%%d=0!\r\nset B=!OUT:%%d=1!\r\n'
|
||
'copy /y "${png.path}" "%A%" >nul\r\ncopy /y "${png.path}" "%B%" >nul\r\n'
|
||
'echo [I] generate_image completed\r\n');
|
||
final engine = LocalSdEngine(device: SdDevice(fake.path, 'cpu', 'CPU'), models: const SdModelPaths(model: 'm'));
|
||
final steps = <String>[];
|
||
final out = p.join(dir.path, 'out');
|
||
final made = await AiJobs.instance.run(engine, const ImageGenRequest(prompt: 'cat', seed: 100, count: 2), outDir: out);
|
||
expect(made.map((g) => g.seed), [100, 101]);
|
||
for (final g in made) {
|
||
expect(File(g.path).existsSync(), isTrue);
|
||
final j = jsonDecode(File('${p.withoutExtension(g.path)}.json').readAsStringSync()) as Map;
|
||
expect(j['prompt'], 'cat');
|
||
expect(j['model'], 'sd15-lcm');
|
||
}
|
||
expect(made.first.seed, (jsonDecode(File('${p.withoutExtension(made.first.path)}.json').readAsStringSync()) as Map)['seed']);
|
||
expect(AiJobs.instance.busy, isFalse);
|
||
expect(steps, isEmpty);
|
||
});
|
||
|
||
test('엔진이 실패하면 마지막 출력으로 알린다', () async {
|
||
if (!Platform.isWindows) return markTestSkipped('Windows 전용');
|
||
final dir = Directory.systemTemp.createTempSync('jj_ai_fail_');
|
||
addTearDown(() => dir.deleteSync(recursive: true));
|
||
final fake = File(p.join(dir.path, 'bad.cmd'))..writeAsStringSync('@echo off\r\necho [E] out of memory\r\nexit /b 3\r\n');
|
||
final engine = LocalSdEngine(device: SdDevice(fake.path, 'cpu', 'CPU'), models: const SdModelPaths(model: 'm'));
|
||
await expectLater(engine.generate(const ImageGenRequest(prompt: 'x'), outDir: dir.path, baseName: 'a'),
|
||
throwsA(isA<ImageAiException>().having((e) => e.message, 'message', contains('out of memory'))));
|
||
});
|
||
});
|
||
|
||
test('123: AI 설정 처음 값 · 저장 · 예전 설정에 AI 그림 화면을 한 번 넣음', () {
|
||
final s = AppSettings();
|
||
expect((s.aiEngine, s.aiDevice, s.aiTaesd, s.aiWidth, s.aiHeight, s.aiSteps, s.aiCfg, s.aiCount),
|
||
('local', 'auto', true, 512, 512, 4, 1.0, 1));
|
||
expect(s.components, contains('aiimage'));
|
||
final b = AppSettings.fromJson((AppSettings()
|
||
..aiWidth = 768
|
||
..aiSteps = 8
|
||
..aiTaesd = false)
|
||
.toJson());
|
||
expect((b.aiWidth, b.aiSteps, b.aiTaesd), (768, 8, false));
|
||
final old = AppSettings.fromJson({'components': ['mkv', 'browser']});
|
||
expect(old.components, contains('aiimage'));
|
||
expect(old.migrated, contains('aiImage'));
|
||
final off = AppSettings.fromJson((AppSettings()..components = ['mkv']).toJson());
|
||
expect(off.components, isNot(contains('aiimage')), reason: '사용자가 끈 것은 그대로');
|
||
});
|
||
}
|