{"id":34104,"date":"2025-11-11T07:40:46","date_gmt":"2025-11-11T04:40:46","guid":{"rendered":"https:\/\/fatihsoysal.com\/blog\/?p=34104"},"modified":"2025-11-11T07:40:46","modified_gmt":"2025-11-11T04:40:46","slug":"stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma","status":"publish","type":"post","link":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/","title":{"rendered":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma"},"content":{"rendered":"

<\/p>\n

Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma<\/h2>\n

Merhaba ve Stable Diffusion ile ki\u015fiselle\u015ftirilmi\u015f yapay zeka modelleri olu\u015fturma d\u00fcnyas\u0131na ho\u015f geldiniz! Bu makale serisinin ilk b\u00f6l\u00fcm\u00fcnde, pop\u00fcler Dreambooth tekni\u011fini kullanarak kendi Stable Diffusion modelinizi bir notebook ortam\u0131nda nas\u0131l e\u011fitece\u011finizi ad\u0131m ad\u0131m inceleyece\u011fiz. \u0130ster belirli bir nesneyi, ister bir ki\u015fiyi, isterse de sanatsal bir stili Stable Diffusion’a \u00f6\u011fretmek isteyin, Dreambooth bu hedeflere ula\u015fmak i\u00e7in g\u00fc\u00e7l\u00fc ve eri\u015filebilir bir y\u00f6ntem sunar. Notebook ortamlar\u0131, \u00f6zellikle Google Colab gibi platformlar, pahal\u0131 GPU donan\u0131m\u0131na sahip olmayanlar i\u00e7in bu karma\u015f\u0131k s\u00fcre\u00e7leri bulut tabanl\u0131, eri\u015filebilir bir \u015fekilde \u00e7al\u0131\u015ft\u0131rma imkan\u0131 sa\u011flar. Bu kapsaml\u0131 rehberde, kurulumdan e\u011fitime ve sonu\u00e7lar\u0131n yorumlanmas\u0131na kadar her \u015feyi detayl\u0131ca ele alaca\u011f\u0131z.<\/p>\n

Stable Diffusion ve Dreambooth Nedir?<\/h3>\n

Stable Diffusion, metin a\u00e7\u0131klamalar\u0131ndan y\u00fcksek kaliteli g\u00f6rseller \u00fcretebilen a\u00e7\u0131k kaynakl\u0131 bir derin \u00f6\u011frenme modelidir. Geni\u015f bir veri k\u00fcmesi \u00fczerinde e\u011fitilmi\u015f olup, \u00e7e\u015fitli kavramlar\u0131 ve stilleri anlayabilir ve g\u00f6rselle\u015ftirebilir. Ancak, varsay\u0131lan olarak, Stable Diffusion sizin \u00f6zel ilgi alanlar\u0131n\u0131za veya benzersiz nesnelerinize dair derinlemesine bir bilgiye sahip de\u011fildir. \u0130\u015fte bu noktada Dreambooth devreye girer.<\/p>\n

Dreambooth, Google ara\u015ft\u0131rmac\u0131lar\u0131 taraf\u0131ndan geli\u015ftirilen ve \u00f6nceden e\u011fitilmi\u015f bir dif\u00fczyon modelini k\u00fc\u00e7\u00fck bir g\u00f6rsel k\u00fcmesi kullanarak belirli bir konuyu veya stili “\u00f6\u011fretmek” i\u00e7in tasarlanm\u0131\u015f bir ince ayar (fine-tuning) tekni\u011fidir. Bu teknik sayesinde, modelin daha \u00f6nce hi\u00e7 g\u00f6rmedi\u011fi bir nesneyi (\u00f6rne\u011fin, kendi evcil hayvan\u0131n\u0131z\u0131n foto\u011fraflar\u0131n\u0131) veya bir sanatsal stili (\u00f6rne\u011fin, belirli bir ressam\u0131n tarz\u0131n\u0131) tan\u0131mas\u0131n\u0131 ve yeni ba\u011flamlarda \u00fcretmesini sa\u011flayabilirsiniz. Dreambooth, modelin mevcut bilgisini korurken, yeni kavramlar\u0131 etkili bir \u015fekilde \u00f6\u011frenmesine olanak tan\u0131yan benzersiz bir “s\u0131n\u0131f g\u00f6r\u00fcnt\u00fcs\u00fc” (class image) ve “d\u00fczenlile\u015ftirme” (regularization) yakla\u015f\u0131m\u0131 kullan\u0131r.<\/p>\n

Neden Notebook Ortam\u0131nda \u00c7al\u0131\u015fmal\u0131?<\/h3>\n

Dreambooth gibi derin \u00f6\u011frenme modellerini e\u011fitmek, \u00f6nemli miktarda hesaplama g\u00fcc\u00fc, \u00f6zellikle de g\u00fc\u00e7l\u00fc bir GPU gerektirir. Herkesin evinde bu t\u00fcr donan\u0131mlara sahip olmas\u0131 m\u00fcmk\u00fcn de\u011fildir. Notebook ortamlar\u0131, bu sorunu \u00e7\u00f6zmek i\u00e7in m\u00fckemmel bir \u00e7\u00f6z\u00fcm sunar:<\/p>\n

* Eri\u015filebilirlik:<\/strong> Google Colab, Kaggle Notebooks veya Gradient gibi platformlar, bulut tabanl\u0131 GPU’lara eri\u015fim sa\u011flar. Bu, kendi donan\u0131m\u0131n\u0131z olmasa bile Dreambooth gibi yo\u011fun modelleri \u00e7al\u0131\u015ft\u0131rman\u0131za olanak tan\u0131r.
\n* Kolay Kurulum:<\/strong> Ba\u011f\u0131ml\u0131l\u0131klar\u0131 ve k\u00fct\u00fcphaneleri manuel olarak kurmak yerine, notebooklar genellikle \u00f6nceden yap\u0131land\u0131r\u0131lm\u0131\u015f ortamlar sunar veya tek bir kod h\u00fccresiyle gerekli t\u00fcm kurulumlar\u0131 yapman\u0131za izin verir.
\n* Payla\u015f\u0131labilirlik ve Tekrarlanabilirlik:<\/strong> Notebooklar, kodunuzu, a\u00e7\u0131klamalar\u0131n\u0131z\u0131 ve \u00e7\u0131kt\u0131lar\u0131n\u0131z\u0131 tek bir belgede birle\u015ftirir. Bu, ba\u015fkalar\u0131yla payla\u015fmay\u0131 kolayla\u015ft\u0131r\u0131r ve ayn\u0131 sonu\u00e7lar\u0131 elde etmek i\u00e7in ad\u0131mlar\u0131 tekrarlamay\u0131 basitle\u015ftirir.
\n* Ad\u0131m Ad\u0131m Y\u00fcr\u00fctme:<\/strong> Kod h\u00fccrelerini tek tek \u00e7al\u0131\u015ft\u0131rarak, s\u00fcreci ad\u0131m ad\u0131m takip edebilir, hatalar\u0131 ay\u0131klayabilir ve her a\u015faman\u0131n \u00e7\u0131kt\u0131s\u0131n\u0131 an\u0131nda g\u00f6rebilirsiniz.
\n* Maliyet Etkinli\u011fi:<\/strong> Google Colab’\u0131n \u00fccretsiz katman\u0131 veya uygun fiyatl\u0131 Colab Pro aboneli\u011fi, pahal\u0131 GPU yat\u0131r\u0131m\u0131 yapmadan bu teknolojiyi denemenize olanak tan\u0131r.<\/p>\n

Bu makalede, en pop\u00fcler ve eri\u015filebilir se\u00e7eneklerden biri olan Google Colab’\u0131 kullanarak Dreambooth’u nas\u0131l \u00e7al\u0131\u015ft\u0131raca\u011f\u0131m\u0131za odaklanaca\u011f\u0131z.<\/p>\n

\u00d6n Ko\u015fullar ve Haz\u0131rl\u0131k<\/h3>\n

Dreambooth e\u011fitimine ba\u015flamadan \u00f6nce baz\u0131 temel gereksinimleri ve haz\u0131rl\u0131klar\u0131 tamamlaman\u0131z gerekmektedir:<\/p>\n

1. Google Hesab\u0131:<\/strong> Google Colab kullanaca\u011f\u0131m\u0131z i\u00e7in bir Google hesab\u0131na sahip olman\u0131z \u015fartt\u0131r.
\n2. Temel Python Bilgisi:<\/strong> Kod h\u00fccrelerini \u00e7al\u0131\u015ft\u0131rmak ve gerekirse k\u00fc\u00e7\u00fck de\u011fi\u015fiklikler yapmak i\u00e7in temel Python s\u00f6zdizimine a\u015fina olmak faydal\u0131d\u0131r.
\n3. Hugging Face Hesab\u0131 ve Eri\u015fim Token’\u0131:<\/strong> Stable Diffusion modellerini indirmek ve kaydetmek i\u00e7in Hugging Face platformunu kullanaca\u011f\u0131z. Bir hesap olu\u015fturman\u0131z ve “Settings -> Access Tokens” b\u00f6l\u00fcm\u00fcnden bir token alman\u0131z gerekecek. Bu token’\u0131 notebook’a yap\u0131\u015ft\u0131rarak kimlik do\u011frulamas\u0131 yapaca\u011f\u0131z.
\n4. Google Drive:<\/strong> E\u011fitilmi\u015f modelinizi ve veri setinizi depolamak i\u00e7in Google Drive’\u0131 kullanaca\u011f\u0131z. Colab, Google Drive’\u0131n\u0131za kolayca ba\u011flanabilir.
\n5. Veri Seti Haz\u0131rl\u0131\u011f\u0131:<\/strong> Bu, en kritik ad\u0131mlardan biridir ve ayr\u0131nt\u0131l\u0131 olarak ele al\u0131nacakt\u0131r.<\/p>\n

Veri Seti Haz\u0131rl\u0131\u011f\u0131: Kalite ve Miktar<\/h3>\n

Dreambooth’un ba\u015far\u0131s\u0131, b\u00fcy\u00fck \u00f6l\u00e7\u00fcde sa\u011flad\u0131\u011f\u0131n\u0131z e\u011fitim verilerinin kalitesine ve \u00e7e\u015fitlili\u011fine ba\u011fl\u0131d\u0131r. \u0130\u015fte veri setinizi haz\u0131rlarken dikkat etmeniz gerekenler:<\/p>\n

Instance G\u00f6r\u00fcnt\u00fcler (Konu G\u00f6r\u00fcnt\u00fcleri)<\/h4>\n

Bunlar, Stable Diffusion’a \u00f6\u011fretmek istedi\u011finiz belirli nesnenin, ki\u015finin veya stilin g\u00f6rselleridir.<\/p>\n

* Miktar:<\/strong> Genellikle 10 ila 20 y\u00fcksek kaliteli g\u00f6r\u00fcnt\u00fc yeterlidir. Daha az g\u00f6r\u00fcnt\u00fc, modelin konuyu yeterince \u00f6\u011frenememesine (underfitting) neden olabilirken, \u00e7ok fazla g\u00f6r\u00fcnt\u00fc modelin a\u015f\u0131r\u0131 genelle\u015fmesine (overfitting) ve daha az yarat\u0131c\u0131 olmas\u0131na yol a\u00e7abilir.
\n* \u00c7e\u015fitlilik:<\/strong> Konunuzu farkl\u0131 a\u00e7\u0131lardan, farkl\u0131 ayd\u0131nlatma ko\u015fullar\u0131nda, farkl\u0131 arka planlarda ve farkl\u0131 pozlarda g\u00f6steren g\u00f6r\u00fcnt\u00fcler se\u00e7in. \u00d6rne\u011fin, bir evcil hayvan\u0131 e\u011fitiyorsan\u0131z, hayvan\u0131n otururken, yatarken, ko\u015farken, farkl\u0131 odalarda veya d\u0131\u015far\u0131da \u00e7ekilmi\u015f foto\u011fraflar\u0131n\u0131 kullan\u0131n. Bu, modelin konuyu \u00e7e\u015fitli ba\u011flamlarda tan\u0131mas\u0131n\u0131 sa\u011flar.
\n* Kalite:<\/strong> G\u00f6r\u00fcnt\u00fcler y\u00fcksek \u00e7\u00f6z\u00fcn\u00fcrl\u00fckl\u00fc, net ve iyi ayd\u0131nlat\u0131lm\u0131\u015f olmal\u0131d\u0131r. Bulan\u0131k, pikselli veya d\u00fc\u015f\u00fck kaliteli g\u00f6r\u00fcnt\u00fclerden ka\u00e7\u0131n\u0131n.
\n* K\u0131rpma ve Boyutland\u0131rma:<\/strong> G\u00f6r\u00fcnt\u00fclerin \u00e7o\u011fu kare (1:1 en boy oran\u0131) ve ortak bir boyutta (\u00f6rne\u011fin, 512×512 veya 768×768 piksel) olmal\u0131d\u0131r. \u00c7o\u011fu Dreambooth notebook’u, e\u011fitim s\u0131ras\u0131nda bu i\u015flemi otomatik olarak yapsa da, \u00f6nceden d\u00fczenlemek daha iyi sonu\u00e7lar verebilir. G\u00f6r\u00fcnt\u00fclerin ana konusunu merkeze alarak k\u0131rpmaya \u00f6zen g\u00f6sterin.
\n* Arka Plan:<\/strong> M\u00fcmk\u00fcnse, konunuzun net bir \u015fekilde \u00f6ne \u00e7\u0131kt\u0131\u011f\u0131 ve dikkat da\u011f\u0131t\u0131c\u0131 arka planlar\u0131n olmad\u0131\u011f\u0131 g\u00f6r\u00fcnt\u00fcler se\u00e7in.
\n* Dosya Yap\u0131s\u0131:<\/strong> Google Drive’\u0131n\u0131zda dreambooth_data\/instance_images<\/code> gibi bir klas\u00f6r olu\u015fturun ve t\u00fcm instance g\u00f6r\u00fcnt\u00fclerinizi buraya y\u00fckleyin.<\/p>\n

Class G\u00f6r\u00fcnt\u00fcler (S\u0131n\u0131f G\u00f6r\u00fcnt\u00fcleri)<\/h4>\n

Class g\u00f6r\u00fcnt\u00fcler, Dreambooth tekni\u011finin \u00f6nemli bir par\u00e7as\u0131d\u0131r. Modelin, \u00f6\u011fretti\u011finiz yeni kavram\u0131 a\u015f\u0131r\u0131 \u00f6\u011frenmesini (overfitting) engellemeye ve genel Stable Diffusion bilgisini korumas\u0131na yard\u0131mc\u0131 olurlar.<\/p>\n

* Amac\u0131:<\/strong> Instance g\u00f6r\u00fcnt\u00fclerinizdeki konunun genel kategorisine ait g\u00f6rsellerdir. \u00d6rne\u011fin, bir “sks dog” (sks, instance token’\u0131n\u0131z olacak) \u00f6\u011fretiyorsan\u0131z, class g\u00f6r\u00fcnt\u00fcleriniz “dog” (k\u00f6pek) kategorisindeki genel g\u00f6rseller olmal\u0131d\u0131r. Bu, modelin “sks dog” ile “dog” aras\u0131ndaki fark\u0131 \u00f6\u011frenmesini ve sadece “sks dog” g\u00f6rd\u00fc\u011f\u00fcn\u00fczde belirli \u00f6zellikleri \u00fcretmesini sa\u011flar.
\n* Miktar:<\/strong> Instance g\u00f6r\u00fcnt\u00fclerinizden \u00e7ok daha fazla class g\u00f6r\u00fcnt\u00fcye ihtiyac\u0131n\u0131z olacakt\u0131r, genellikle 200 ila 500 aras\u0131nda.
\n* Elde Edilmesi:<\/strong> Bu g\u00f6r\u00fcnt\u00fcleri genellikle Stable Diffusion’\u0131n kendisiyle olu\u015fturursunuz. \u00d6rne\u011fin, “a photo of a dog” gibi bir prompt kullanarak \u00e7ok say\u0131da k\u00f6pek g\u00f6r\u00fcnt\u00fcs\u00fc \u00fcretebilirsiniz. Bu, modelin kendi “k\u00f6pek” bilgisini peki\u015ftirmesine yard\u0131mc\u0131 olur.
\n* Dosya Yap\u0131s\u0131:<\/strong> Google Drive’\u0131n\u0131zda dreambooth_data\/class_images<\/code> gibi bir klas\u00f6r olu\u015fturun ve t\u00fcm class g\u00f6r\u00fcnt\u00fclerinizi buraya y\u00fckleyin.<\/p>\n

\u00d6rnek Veri Seti Yap\u0131s\u0131:<\/strong><\/p>\n

Google Drive\/\n\u2514\u2500\u2500 dreambooth_data\/\n    \u251c\u2500\u2500 instance_images\/\n    \u2502   \u251c\u2500\u2500 sks_dog_1.jpg\n    \u2502   \u251c\u2500\u2500 sks_dog_2.jpg\n    \u2502   \u2514\u2500\u2500 ... (10-20 adet)\n    \u2514\u2500\u2500 class_images\/\n        \u251c\u2500\u2500 dog_1.jpg\n        \u251c\u2500\u2500 dog_2.jpg\n        \u2514\u2500\u2500 ... (200-500 adet)<\/code><\/pre>\n
Google Colab Ortam\u0131n\u0131 Kurma<\/h3>\n
\u015eimdi Google Colab’\u0131 a\u00e7\u0131p e\u011fitim ortam\u0131m\u0131z\u0131 haz\u0131rlayal\u0131m:<\/p>\n
1.  Yeni Bir Notebook Olu\u015fturma:<\/strong> Google Drive’\u0131n\u0131za gidin, sa\u011f t\u0131klay\u0131n -> Di\u011fer -> Google Colaboratory.
\n2.  GPU \u00c7al\u0131\u015fma Zaman\u0131 Se\u00e7me:<\/strong>
\n    *   Notebook a\u00e7\u0131ld\u0131ktan sonra, \u00fcst men\u00fcden \u00c7al\u0131\u015fma Zaman\u0131<\/code> (Runtime) -> \u00c7al\u0131\u015fma Zaman\u0131 T\u00fcr\u00fcn\u00fc De\u011fi\u015ftir<\/code> (Change runtime type) se\u00e7ene\u011fine t\u0131klay\u0131n.
\n    *   Donan\u0131m h\u0131zland\u0131r\u0131c\u0131<\/code> (Hardware accelerator) olarak GPU<\/code>‘yu se\u00e7in. Genellikle T4<\/code> varsay\u0131lan olarak gelir ve Dreambooth i\u00e7in yeterlidir. Daha h\u0131zl\u0131 e\u011fitim i\u00e7in Colab Pro aboneli\u011finiz varsa V100<\/code> veya A100<\/code> se\u00e7ebilirsiniz.
\n    *   Kaydedin.
\n3.  Google Drive’\u0131 Ba\u011flama:<\/strong>
\n    *   \u0130lk kod h\u00fccresine a\u015fa\u011f\u0131daki kodu yap\u0131\u015ft\u0131r\u0131n ve \u00e7al\u0131\u015ft\u0131r\u0131n:<\/p>\n
from google.colab import drive\n        drive.mount('\/content\/drive')<\/code><\/pre>\n
    *   Bir ba\u011flant\u0131ya t\u0131klaman\u0131z, Google hesab\u0131n\u0131zla oturum a\u00e7man\u0131z ve Colab’a Drive’\u0131n\u0131za eri\u015fim izni vermeniz istenecektir. Bu ad\u0131m, veri setinize ve e\u011fitilmi\u015f modelinizi kaydetmek i\u00e7in gereklidir.<\/p>\n
Dreambooth E\u011fitim Noteboook’u Se\u00e7imi ve Ad\u0131mlar<\/h3>\n
Topluluk taraf\u0131ndan geli\u015ftirilmi\u015f bir\u00e7ok Dreambooth notebook’u bulunmaktad\u0131r. Bunlardan baz\u0131lar\u0131 TheLastBen’in, ShivamShrirao’nun veya Joe Penna’n\u0131n notebooklar\u0131d\u0131r. Genellikle, bu notebooklar benzer bir ak\u0131\u015f\u0131 takip eder. Biz burada genel bir ak\u0131\u015f\u0131 sim\u00fcle edece\u011fiz:<\/p>\n
1. Gerekli K\u00fct\u00fcphaneleri ve Ba\u011f\u0131ml\u0131l\u0131klar\u0131 Kurma<\/h4>\n
\u0130lk ad\u0131m, Dreambooth’u \u00e7al\u0131\u015ft\u0131rmak i\u00e7in gerekli t\u00fcm Python k\u00fct\u00fcphanelerini ve ara\u00e7lar\u0131n\u0131 y\u00fcklemektir. Bu, genellikle diffusers<\/code>, transformers<\/code>, accelerate<\/code>, xformers<\/code> gibi k\u00fct\u00fcphaneleri ve bitsandbytes<\/code> gibi optimizasyon ara\u00e7lar\u0131n\u0131 i\u00e7erir.<\/p>\n
!pip install -qqq diffusers transformers accelerate ftfy bitsandbytes --upgrade\n!pip install -qqq xformers\n!pip install -qqq gradio<\/code><\/pre>\n
Bu komutlar, sessizce (-qqq<\/code>) ve h\u0131zl\u0131 bir \u015fekilde gerekli paketleri kurar. xformers<\/code> \u00f6zellikle daha az GPU belle\u011fi kullanarak daha h\u0131zl\u0131 e\u011fitim sa\u011flar.<\/p>\n
2. Hugging Face Kimlik Do\u011frulamas\u0131<\/h4>\n
Stable Diffusion modelini indirmek ve model a\u011f\u0131rl\u0131klar\u0131n\u0131 Hugging Face’e y\u00fcklemek i\u00e7in kimlik do\u011frulaman\u0131z gerekir.<\/p>\n
from huggingface_hub import notebook_login\nnotebook_login()<\/code><\/pre>\n
Bu komutu \u00e7al\u0131\u015ft\u0131rd\u0131\u011f\u0131n\u0131zda, sizden Hugging Face token’\u0131n\u0131z\u0131 yap\u0131\u015ft\u0131rman\u0131z istenecektir. Token’\u0131n\u0131z\u0131 yap\u0131\u015ft\u0131r\u0131n ve Enter tu\u015funa bas\u0131n.<\/p>\n
3. E\u011fitim Parametrelerini Tan\u0131mlama<\/h4>\n
Burada, e\u011fitim s\u00fcrecini y\u00f6nlendirecek temel parametreleri belirleyece\u011fiz.<\/p>\n
import os\n\n
Temel Stable Diffusion modeli (\u00f6rn: v1.5)<\/h2>\nMODEL_NAME = \"runwayml\/stable-diffusion-v1-5\"\n\n
E\u011fitilmi\u015f modelinizin kaydedilece\u011fi Google Drive yolu<\/h2>\nOUTPUT_DIR = \"\/content\/drive\/MyDrive\/dreambooth_models\/my_sks_model\"\n\n
Instance g\u00f6rsellerinizin yolu<\/h2>\nINSTANCE_DIR = \"\/content\/drive\/MyDrive\/dreambooth_data\/instance_images\"\n\n
Class g\u00f6rsellerinizin yolu<\/h2>\nCLASS_DIR = \"\/content\/drive\/MyDrive\/dreambooth_data\/class_images\"\n\n
\u00d6\u011fretmek istedi\u011finiz konu i\u00e7in benzersiz token (\u00f6rn: sks)<\/h2>\nINSTANCE_PROMPT = \"a photo of sks dog\" # \"sks\" burada benzersiz token\nCLASS_PROMPT = \"a photo of a dog\"\n\n
Modelin kaydedilece\u011fi dizini olu\u015fturma<\/h2>\nos.makedirs(OUTPUT_DIR, exist_ok=True)\nos.makedirs(INSTANCE_DIR, exist_ok=True)\nos.makedirs(CLASS_DIR, exist_ok=True) # Class g\u00f6rselleri manuel olarak olu\u015fturulacaksa<\/code><\/pre>\n
*   MODEL_NAME<\/code>: Hangi \u00f6nceden e\u011fitilmi\u015f Stable Diffusion modelini temel alaca\u011f\u0131n\u0131z\u0131 belirtir. runwayml\/stable-diffusion-v1-5<\/code> genellikle iyi bir ba\u015flang\u0131\u00e7 noktas\u0131d\u0131r.
\n*   OUTPUT_DIR<\/code>: E\u011fitilmi\u015f modelinizin kaydedilece\u011fi yer.
\n*   INSTANCE_DIR<\/code>: Instance g\u00f6rsellerinizin bulundu\u011fu klas\u00f6r.
\n*   CLASS_DIR<\/code>: Class g\u00f6rsellerinizin bulundu\u011fu klas\u00f6r.
\n*   INSTANCE_PROMPT<\/code>: Modelin yeni konuyu tan\u0131mas\u0131 i\u00e7in kullanaca\u011f\u0131n\u0131z prompt. Genellikle [benzersiz_token] [konu_tan\u0131m\u0131]<\/code> \u015feklinde olur. sks<\/code> gibi rastgele bir token se\u00e7mek, modelin bu kelimeyi mevcut kelime da\u011farc\u0131\u011f\u0131yla kar\u0131\u015ft\u0131rmamas\u0131n\u0131 sa\u011flar.
\n*   CLASS_PROMPT<\/code>: Class g\u00f6rsellerinizi a\u00e7\u0131klayan prompt.<\/p>\n
4. Class G\u00f6r\u00fcnt\u00fclerini Olu\u015fturma (Gerekliyse)<\/h4>\n
E\u011fer class g\u00f6rsellerinizi hen\u00fcz olu\u015fturmad\u0131ysan\u0131z, bu ad\u0131mda Stable Diffusion’\u0131 kullanarak bunlar\u0131 otomatik olarak \u00fcretebilirsiniz. Bu, modelin genel “dog” kavram\u0131n\u0131 korumas\u0131na yard\u0131mc\u0131 olurken, “sks dog” kavram\u0131n\u0131 \u00f6\u011frenmesini sa\u011flar.<\/p>\n
# Sadece class g\u00f6rselleri yoksa \u00e7al\u0131\u015ft\u0131r\u0131n\nif not os.listdir(CLASS_DIR):\n    print(f\"Generating class images for {CLASS_PROMPT} in {CLASS_DIR}...\")\n    !accelerate launch train_dreambooth.py \\\n        --pretrained_model_name_or_path=$MODEL_NAME \\\n        --output_dir=$CLASS_DIR \\\n        --class_data_dir=$CLASS_DIR \\\n        --class_prompt=\"$CLASS_PROMPT\" \\\n        --num_class_images=200 \\ # Ka\u00e7 adet class g\u00f6rseli olu\u015fturulacak\n        --with_prior_preservation \\\n        --prior_loss_weight=1.0 \\\n        --seed=1234 \\\n        --resolution=512 \\\n        --train_batch_size=1 \\\n        --sample_batch_size=1 \\\n        --gradient_accumulation_steps=1 \\\n        --learning_rate=5e-7 \\\n        --lr_scheduler=\"constant\" \\\n        --lr_warmup_steps=0 \\\n        --num_train_epochs=1 \\\n        --save_interval=10000 \\\n        --max_train_steps=10000 \\\n        --mixed_precision=\"fp16\" \\\n        --train_text_encoder\n    print(\"Class images generation complete.\")\nelse:\n    print(\"Class images already exist. Skipping generation.\")<\/code><\/pre>\n
Bu ad\u0131m, accelerate<\/code> k\u00fct\u00fcphanesini kullanarak train_dreambooth.py<\/code> scriptini \u00e7al\u0131\u015ft\u0131r\u0131r ve belirtilen CLASS_PROMPT<\/code> ile 200 adet class g\u00f6r\u00fcnt\u00fcs\u00fc olu\u015fturur.<\/p>\n
5. Dreambooth E\u011fitimini Ba\u015flatma<\/h4>\n
\u015eimdi as\u0131l Dreambooth e\u011fitimini ba\u015flatma zaman\u0131. Bu, en uzun s\u00fcren ad\u0131md\u0131r.<\/p>\n
!accelerate launch train_dreambooth.py \\\n    --pretrained_model_name_or_path=$MODEL_NAME \\\n    --instance_data_dir=$INSTANCE_DIR \\\n    --class_data_dir=$CLASS_DIR \\\n    --instance_prompt=\"$INSTANCE_PROMPT\" \\\n    --class_prompt=\"$CLASS_PROMPT\" \\\n    --with_prior_preservation \\\n    --prior_loss_weight=1.0 \\\n    --output_dir=$OUTPUT_DIR \\\n    --seed=1234 \\\n    --resolution=512 \\\n    --train_batch_size=1 \\\n    --sample_batch_size=1 \\\n    --gradient_accumulation_steps=1 \\\n    --gradient_checkpointing \\\n    --learning_rate=5e-6 \\ # Dikkat! Bu \u00e7ok \u00f6nemli bir parametre\n    --lr_scheduler=\"constant\" \\\n    --lr_warmup_steps=0 \\\n    --num_train_epochs=1 \\\n    --max_train_steps=800 \\ # Instance g\u00f6rsel say\u0131s\u0131na g\u00f6re ayarlanmal\u0131 (\u00f6rn: 10 resim i\u00e7in 800-1000 ad\u0131m)\n    --checkpointing_steps=500 \\ # Checkpoint kaydetme s\u0131kl\u0131\u011f\u0131\n    --mixed_precision=\"fp16\" \\\n    --use_8bit_adam \\ # Bellek kullan\u0131m\u0131n\u0131 optimize eder\n    --train_text_encoder<\/code><\/pre>\n
Bu komut blo\u011fu, Dreambooth e\u011fitimini ba\u015flat\u0131r. \u0130\u015fte baz\u0131 \u00f6nemli parametreler:<\/p>\n
*   --instance_data_dir<\/code> ve --class_data_dir<\/code>: Veri setlerinizin yollar\u0131.
\n*   --instance_prompt<\/code> ve --class_prompt<\/code>: Daha \u00f6nce tan\u0131mlad\u0131\u011f\u0131m\u0131z prompt’lar.
\n*   --with_prior_preservation<\/code> ve --prior_loss_weight<\/code>: Class g\u00f6rselleriyle d\u00fczenlile\u015ftirmeyi etkinle\u015ftirir.
\n*   --resolution<\/code>: E\u011fitim s\u0131ras\u0131nda g\u00f6rsellerin yeniden boyutland\u0131r\u0131laca\u011f\u0131 \u00e7\u00f6z\u00fcn\u00fcrl\u00fck. Genellikle 512 veya 768.
\n*   --train_batch_size<\/code>: Her e\u011fitim ad\u0131m\u0131nda i\u015flenecek g\u00f6rsel say\u0131s\u0131. Bellek k\u0131s\u0131tlamalar\u0131 nedeniyle genellikle 1 olarak ayarlan\u0131r.
\n*   --gradient_accumulation_steps<\/code>: Bellek yetersizli\u011fi durumunda, birden fazla k\u00fc\u00e7\u00fck ad\u0131m\u0131 birle\u015ftirerek daha b\u00fcy\u00fck bir sanal batch boyutu elde etmenizi sa\u011flar.
\n*   --learning_rate<\/code>: Bu, en kritik parametrelerden biridir. \u00c7ok y\u00fcksek bir de\u011fer modelin h\u0131zl\u0131ca bozulmas\u0131na neden olurken, \u00e7ok d\u00fc\u015f\u00fck bir de\u011fer e\u011fitimin \u00e7ok yava\u015f ilerlemesine yol a\u00e7ar. Genellikle 1e-6 ile 5e-6 aras\u0131nda bir de\u011ferle ba\u015flan\u0131r.
\n*   --max_train_steps<\/code>: Toplam e\u011fitim ad\u0131m\u0131 say\u0131s\u0131. Instance g\u00f6rsel say\u0131n\u0131za ba\u011fl\u0131d\u0131r. 10-20 g\u00f6rsel i\u00e7in 800-1200 ad\u0131m genellikle iyi sonu\u00e7lar verir. \u00c7ok fazla ad\u0131m overfitting’e yol a\u00e7abilir.
\n*   --mixed_precision=\"fp16\"<\/code>: Yar\u0131 duyarl\u0131kl\u0131 kayan nokta aritmeti\u011fi kullanarak bellek kullan\u0131m\u0131n\u0131 azalt\u0131r ve e\u011fitimi h\u0131zland\u0131r\u0131r.
\n*   --use_8bit_adam<\/code>: Adam optimizat\u00f6r\u00fcn\u00fcn 8-bit versiyonunu kullanarak GPU belle\u011finden tasarruf eder.
\n*   --train_text_encoder<\/code>: Metin kodlay\u0131c\u0131s\u0131n\u0131 da e\u011fitir, bu genellikle daha iyi sonu\u00e7lar verir ancak daha fazla bellek gerektirir.<\/p>\n
E\u011fitim devam ederken, Colab \u00e7\u0131kt\u0131s\u0131nda kay\u0131p (loss) de\u011ferlerini ve bazen \u00f6rnek olu\u015fturmalar\u0131 g\u00f6receksiniz. Kay\u0131p de\u011feri d\u00fc\u015fmeye ba\u015flamal\u0131d\u0131r.<\/p>\n
6. E\u011fitilmi\u015f Modeli Stable Diffusion Web UI Uyumlu Formata D\u00f6n\u00fc\u015ft\u00fcrme<\/h4>\n
E\u011fitim tamamland\u0131ktan sonra, modeliniz Hugging Face Diffusers format\u0131nda kaydedilecektir. Bunu AUTOMATIC1111 Web UI gibi yerel Stable Diffusion aray\u00fczlerinde kullanabilmek i\u00e7in .ckpt<\/code> (Checkpoint) format\u0131na d\u00f6n\u00fc\u015ft\u00fcrmeniz gerekir.<\/p>\n
import os\n\n
D\u00f6n\u00fc\u015ft\u00fcrme scriptinin yolu (genellikle ayn\u0131 repo i\u00e7inde bulunur)<\/h2>\nCONVERT_SCRIPT = \"\/content\/diffusers\/scripts\/convert_diffusers_to_original_stable_diffusion.py\" # Bu yol notebook'a g\u00f6re de\u011fi\u015febilir\n\n
E\u011fitilmi\u015f modelinizin kaydedildi\u011fi klas\u00f6r<\/h2>\nOUTPUT_DIR = \"\/content\/drive\/MyDrive\/dreambooth_models\/my_sks_model\"\n\n
D\u00f6n\u00fc\u015ft\u00fcr\u00fclen modelin kaydedilece\u011fi yol ve dosya ad\u0131<\/h2>\nCHECKPOINT_PATH = os.path.join(OUTPUT_DIR, \"sks_model.ckpt\")\n\n
D\u00f6n\u00fc\u015ft\u00fcrme komutu<\/h2>\n!python $CONVERT_SCRIPT --model_path $OUTPUT_DIR --checkpoint_path $CHECKPOINT_PATH --half<\/code><\/pre>\n
*   --model_path<\/code>: E\u011fitilmi\u015f Diffusers modelinin klas\u00f6r\u00fc.
\n*   --checkpoint_path<\/code>: D\u00f6n\u00fc\u015ft\u00fcr\u00fclen .ckpt<\/code> dosyas\u0131n\u0131n kaydedilece\u011fi yol.
\n*   --half<\/code>: Modelin yar\u0131 duyarl\u0131kl\u0131 (fp16) olarak kaydedilmesini sa\u011flar, bu da dosya boyutunu k\u00fc\u00e7\u00fclt\u00fcr ve \u00e7o\u011fu GPU’da performans art\u0131\u015f\u0131 sa\u011flar.<\/p>\n
Bu ad\u0131m\u0131 tamamlad\u0131ktan sonra, Google Drive’\u0131n\u0131zdaki OUTPUT_DIR<\/code> i\u00e7inde sks_model.ckpt<\/code> ad\u0131nda bir dosya bulacaks\u0131n\u0131z. Bu dosyay\u0131 indirip yerel Stable Diffusion Web UI kurulumunuza (\u00f6rne\u011fin, stable-diffusion-webui\/models\/Stable-diffusion<\/code> klas\u00f6r\u00fcne) ta\u015f\u0131yarak kullanabilirsiniz.<\/p>\n
E\u011fitilmi\u015f Model ile G\u00f6r\u00fcnt\u00fc Olu\u015fturma<\/h3>\n
.ckpt<\/code> dosyan\u0131z\u0131 yerel Stable Diffusion Web UI’n\u0131za y\u00fckledikten sonra, art\u0131k kendi ki\u015fiselle\u015ftirilmi\u015f modelinizle g\u00f6r\u00fcnt\u00fc olu\u015fturmaya ba\u015flayabilirsiniz.<\/p>\n
1.  Modeli Se\u00e7in:<\/strong> Web UI’da model se\u00e7icisinden yeni e\u011fitilmi\u015f sks_model.ckpt<\/code> dosyan\u0131z\u0131 se\u00e7in.
\n2.  Prompt Kullan\u0131m\u0131:<\/strong> G\u00f6r\u00fcnt\u00fc olu\u015fturmak i\u00e7in INSTANCE_PROMPT<\/code>‘unuzu kullan\u0131n. \u00d6rne\u011fin:
\n    *   a photo of sks dog in a spaceship<\/code>
\n    *   sks dog wearing a hat, digital art<\/code>
\n    *   sks dog as a cyberpunk character<\/code>
\n3.  Deneyler:<\/strong> Farkl\u0131 prompt’lar, negatif prompt’lar, CFG \u00f6l\u00e7e\u011fi (genellikle 7-10 aras\u0131), sampler (\u00f6rne\u011fin DPM++ 2M Karras, Euler a) ve ad\u0131m say\u0131s\u0131 (genellikle 20-30 aras\u0131) ile deneyler yaparak en iyi sonu\u00e7lar\u0131 elde edin.
\n4.  \u0130pucu:<\/strong> Modelin, instance token’\u0131n\u0131z\u0131 (\u00f6rne\u011fin sks<\/code>) kullanarak do\u011fru konuyu tan\u0131d\u0131\u011f\u0131ndan emin olun. E\u011fer model konuyu tan\u0131m\u0131yorsa veya yanl\u0131\u015f yorumluyorsa, e\u011fitim parametrelerinizi veya veri setinizi g\u00f6zden ge\u00e7irmeniz gerekebilir.<\/p>\n
S\u0131k Kar\u015f\u0131la\u015f\u0131lan Sorunlar ve \u00c7\u00f6z\u00fcmleri<\/h3>\n
*   CUDA Out of Memory (Bellek Yetmezli\u011fi):<\/strong> Bu, en yayg\u0131n sorunlardan biridir.
\n    *   --train_batch_size<\/code> de\u011ferini 1’e d\u00fc\u015f\u00fcr\u00fcn.
\n    *   --gradient_accumulation_steps<\/code> de\u011ferini art\u0131r\u0131n (\u00f6rne\u011fin 2 veya 4).
\n    *   --use_8bit_adam<\/code> ve --gradient_checkpointing<\/code> parametrelerini etkinle\u015ftirin.
\n    *   --mixed_precision=\"fp16\"<\/code> kulland\u0131\u011f\u0131n\u0131zdan emin olun.
\n    *   Daha d\u00fc\u015f\u00fck \u00e7\u00f6z\u00fcn\u00fcrl\u00fck (--resolution<\/code>) kullanmay\u0131 deneyin (\u00f6rn: 512).
\n    *   Colab Pro kullan\u0131yorsan\u0131z daha g\u00fc\u00e7l\u00fc bir GPU (V100 veya A100) se\u00e7in.
\n*   K\u00f6t\u00fc Kaliteli veya Bozuk G\u00f6r\u00fcnt\u00fcler:<\/strong>
\n    *   Overfitting (A\u015f\u0131r\u0131 \u00d6\u011frenme):<\/strong> Model, instance g\u00f6rsellerini ezberlemi\u015f ve yeni \u015feyler \u00fcretemiyordur.
\n        *   --max_train_steps<\/code> de\u011ferini azalt\u0131n.
\n        *   --learning_rate<\/code> de\u011ferini d\u00fc\u015f\u00fcr\u00fcn.
\n        *   Class g\u00f6rsellerinizin yeterli ve \u00e7e\u015fitli oldu\u011fundan emin olun.
\n    *   Underfitting (Yetersiz \u00d6\u011frenme):<\/strong> Model konuyu yeterince \u00f6\u011frenememi\u015ftir.
\n        *   --max_train_steps<\/code> de\u011ferini art\u0131r\u0131n.
\n        *   --learning_rate<\/code> de\u011ferini hafif\u00e7e art\u0131r\u0131n.
\n        *   Instance g\u00f6rsellerinizin kalitesini ve miktar\u0131n\u0131 kontrol edin.
\n        *   Daha fazla ve daha \u00e7e\u015fitli instance g\u00f6rsel ekleyin.
\n    *   K\u00f6t\u00fc Veri Seti:<\/strong> Bulan\u0131k, d\u00fc\u015f\u00fck kaliteli veya konuyla alakas\u0131z g\u00f6rseller kullanmak k\u00f6t\u00fc sonu\u00e7lara yol a\u00e7ar. Veri setinizi dikkatlice g\u00f6zden ge\u00e7irin.
\n*   E\u011fitim \u00c7ok Uzun S\u00fcr\u00fcyor:<\/strong>
\n    *   Daha g\u00fc\u00e7l\u00fc bir GPU’ya ge\u00e7in (Colab Pro).
\n    *   --mixed_precision=\"fp16\"<\/code> ve use_8bit_adam<\/code> kulland\u0131\u011f\u0131n\u0131zdan emin olun.
\n    *   xformers<\/code> k\u00fct\u00fcphanesinin kurulu ve etkin oldu\u011fundan emin olun.
\n*   Hugging Face Kimlik Do\u011frulama Hatas\u0131:<\/strong> Token’\u0131n\u0131z\u0131 do\u011fru girdi\u011finizden emin olun. Token’\u0131n s\u00fcresinin dolmad\u0131\u011f\u0131n\u0131 kontrol edin.<\/p>\n
En \u0130yi Uygulamalar ve \u0130pu\u00e7lar\u0131<\/h3>\n
*   Veri Seti Her \u015eeydir:<\/strong> Kaliteli, \u00e7e\u015fitli ve iyi haz\u0131rlanm\u0131\u015f bir veri seti, ba\u015far\u0131l\u0131 bir Dreambooth modelinin temelidir. Bu ad\u0131ma en \u00e7ok zaman\u0131 ay\u0131r\u0131n.
\n*   K\u00fc\u00e7\u00fck Ad\u0131mlarla Ba\u015flay\u0131n:<\/strong> \u00d6zellikle learning_rate<\/code> ve max_train_steps<\/code> gibi parametreler i\u00e7in k\u00fc\u00e7\u00fck de\u011ferlerle ba\u015flay\u0131n ve sonu\u00e7lara g\u00f6re ayarlamalar yap\u0131n.
\n*   Kay\u0131p De\u011ferini \u0130zleyin:<\/strong> E\u011fitim s\u0131ras\u0131nda kay\u0131p de\u011ferinin d\u00fc\u015ft\u00fc\u011f\u00fcnden emin olun. E\u011fer y\u00fckseliyorsa veya durmuyorsa, bir sorun var demektir.
\n*   D\u00fczenli Kay\u0131t Noktalar\u0131 (Checkpoints):<\/strong> E\u011fitim s\u0131ras\u0131nda d\u00fczenli aral\u0131klarla modelinizi kaydedin (--checkpointing_steps<\/code>). Bu, e\u011fitimin yar\u0131da kesilmesi durumunda ba\u015ftan ba\u015flamak zorunda kalmaman\u0131z\u0131 sa\u011flar ve farkl\u0131 e\u011fitim a\u015famalar\u0131ndaki model performans\u0131n\u0131 kar\u015f\u0131la\u015ft\u0131rman\u0131za olanak tan\u0131r.
\n*   Deneysel Olun:<\/strong> Dreambooth e\u011fitimi sanattan \u00e7ok bir bilimdir. Farkl\u0131 parametre kombinasyonlar\u0131, prompt’lar ve veri setleri ile deney yapmaktan \u00e7ekinmeyin.
\n*   Token Se\u00e7imi:<\/strong> Benzersiz ve daha \u00f6nce Stable Diffusion’\u0131n bilmedi\u011fi k\u0131sa, anlams\u0131z bir token (\u00f6rne\u011fin “sks”, “zxc”, “jdo”) kullanmak, modelin bu token’\u0131 yeni bir kavramla ili\u015fkilendirmesini kolayla\u015ft\u0131r\u0131r.<\/p>\n
Sonu\u00e7<\/h3>\n
Bu kapsaml\u0131 rehberde, Stable Diffusion’\u0131n g\u00fcc\u00fcn\u00fc Dreambooth tekni\u011fiyle nas\u0131l ki\u015fiselle\u015ftirebilece\u011finizi ve bunu Google Colab gibi eri\u015filebilir bir notebook ortam\u0131nda nas\u0131l yapaca\u011f\u0131n\u0131z\u0131 ad\u0131m ad\u0131m \u00f6\u011frendiniz. Veri setinizi haz\u0131rlamaktan, e\u011fitim parametrelerini ayarlamaya ve sonu\u00e7lar\u0131 yorumlamaya kadar t\u00fcm s\u00fcre\u00e7leri ele ald\u0131k. Art\u0131k kendi benzersiz kavramlar\u0131n\u0131z\u0131 veya stillerinizi Stable Diffusion’a \u00f6\u011fretme yetene\u011fine sahipsiniz.<\/p>\n
Unutmay\u0131n, bu sadece ba\u015flang\u0131\u00e7! Yapay zeka modelleriyle \u00e7al\u0131\u015fmak s\u00fcrekli \u00f6\u011frenme ve deney yapma gerektirir. Modelinizi daha da geli\u015ftirmek i\u00e7in farkl\u0131 veri setleri, e\u011fitim parametreleri ve prompt’lar deneyerek s\u0131n\u0131rlar\u0131 zorlamaktan \u00e7ekinmeyin. Bir sonraki b\u00f6l\u00fcmde, Dreambooth e\u011fitimini daha da optimize etmek ve daha geli\u015fmi\u015f teknikler kullanmak i\u00e7in ipu\u00e7lar\u0131 ve p\u00fcf noktalar\u0131 \u00fczerine odaklanaca\u011f\u0131z. Ba\u015far\u0131lar dileriz!<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma\nMerhaba ve Stable Diffusion ile ki\u015fiselle\u015ftirilmi\u015f yapay zeka","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"csco_page_header_type":"","csco_page_load_nextpost":"","csco_page_subscribe_form":"","csco_page_contact_form":"","footnotes":""},"categories":[1],"tags":[],"class_list":{"0":"post-34104","1":"post","2":"type-post","3":"status-publish","4":"format-standard","6":"category-genel","7":"cs-entry","8":"cs-video-wrap"},"yoast_head":"\nStable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma - Kodlar\u0131n Gizemli D\u00fcnyas\u0131<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/\" \/>\n<meta property=\"og:locale\" content=\"tr_TR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma\" \/>\n<meta property=\"og:description\" content=\"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma Merhaba ve Stable Diffusion ile ki\u015fiselle\u015ftirilmi\u015f yapay zeka\" \/>\n<meta property=\"og:url\" content=\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/\" \/>\n<meta property=\"og:site_name\" content=\"Kodlar\u0131n Gizemli D\u00fcnyas\u0131\" \/>\n<meta property=\"article:published_time\" content=\"2025-11-11T04:40:46+00:00\" \/>\n<meta name=\"author\" content=\"Fatih Soysal\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Yazan:\" \/>\n\t<meta name=\"twitter:data1\" content=\"Fatih Soysal\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tahmini okuma s\u00fcresi\" \/>\n\t<meta name=\"twitter:data2\" content=\"18 dakika\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/\"},\"author\":{\"name\":\"Fatih Soysal\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"headline\":\"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma\",\"datePublished\":\"2025-11-11T04:40:46+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/\"},\"wordCount\":2958,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"inLanguage\":\"tr\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#respond\"]}],\"copyrightYear\":\"2025\",\"copyrightHolder\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#organization\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/\",\"name\":\"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma - Kodlar\u0131n Gizemli D\u00fcnyas\u0131\",\"isPartOf\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#website\"},\"datePublished\":\"2025-11-11T04:40:46+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#breadcrumb\"},\"inLanguage\":\"tr\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Anasayfa\",\"item\":\"https:\/\/fatihsoysal.com\/blog\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#website\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/\",\"name\":\"Fatihsoysal.com\",\"description\":\"Blog - Yaz\u0131l\u0131m D\u00fcnyas\u0131 Tecr\u00fcbelerim\",\"publisher\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/fatihsoysal.com\/blog\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"tr\"},{\"@type\":[\"Person\",\"Organization\"],\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\",\"name\":\"Fatih Soysal\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"tr\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png\",\"contentUrl\":\"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png\",\"width\":512,\"height\":512,\"caption\":\"Fatih Soysal\"},\"logo\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/\"},\"description\":\"Kullan\u0131m ve kodlama m\u00fckemmeliyetini odak alan uygulamalar olu\u015fturma deneyimine sahip, profesyonel olarak 15+ y\u0131l \u00fczeri deneyime sahip bir yaz\u0131l\u0131m m\u00fchendisi.\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/author\/fatihsoysal\/\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma - Kodlar\u0131n Gizemli D\u00fcnyas\u0131","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/","og_locale":"tr_TR","og_type":"article","og_title":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma","og_description":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma Merhaba ve Stable Diffusion ile ki\u015fiselle\u015ftirilmi\u015f yapay zeka","og_url":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/","og_site_name":"Kodlar\u0131n Gizemli D\u00fcnyas\u0131","article_published_time":"2025-11-11T04:40:46+00:00","author":"Fatih Soysal","twitter_card":"summary_large_image","twitter_misc":{"Yazan:":"Fatih Soysal","Tahmini okuma s\u00fcresi":"18 dakika"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#article","isPartOf":{"@id":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/"},"author":{"name":"Fatih Soysal","@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"headline":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma","datePublished":"2025-11-11T04:40:46+00:00","mainEntityOfPage":{"@id":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/"},"wordCount":2958,"commentCount":0,"publisher":{"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"inLanguage":"tr","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#respond"]}],"copyrightYear":"2025","copyrightHolder":{"@id":"https:\/\/fatihsoysal.com\/blog\/#organization"}},{"@type":"WebPage","@id":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/","url":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/","name":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma - Kodlar\u0131n Gizemli D\u00fcnyas\u0131","isPartOf":{"@id":"https:\/\/fatihsoysal.com\/blog\/#website"},"datePublished":"2025-11-11T04:40:46+00:00","breadcrumb":{"@id":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#breadcrumb"},"inLanguage":"tr","potentialAction":[{"@type":"ReadAction","target":["https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/fatihsoysal.com\/blog\/stable-diffusion-ogretici-bolum-1-notebook-ortaminda-dreambooth-calistirma\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Anasayfa","item":"https:\/\/fatihsoysal.com\/blog\/"},{"@type":"ListItem","position":2,"name":"Stable Diffusion \u00d6\u011fretici B\u00f6l\u00fcm 1: Notebook Ortam\u0131nda Dreambooth \u00c7al\u0131\u015ft\u0131rma"}]},{"@type":"WebSite","@id":"https:\/\/fatihsoysal.com\/blog\/#website","url":"https:\/\/fatihsoysal.com\/blog\/","name":"Fatihsoysal.com","description":"Blog - Yaz\u0131l\u0131m D\u00fcnyas\u0131 Tecr\u00fcbelerim","publisher":{"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/fatihsoysal.com\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"tr"},{"@type":["Person","Organization"],"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1","name":"Fatih Soysal","image":{"@type":"ImageObject","inLanguage":"tr","@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/","url":"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png","contentUrl":"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png","width":512,"height":512,"caption":"Fatih Soysal"},"logo":{"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/"},"description":"Kullan\u0131m ve kodlama m\u00fckemmeliyetini odak alan uygulamalar olu\u015fturma deneyimine sahip, profesyonel olarak 15+ y\u0131l \u00fczeri deneyime sahip bir yaz\u0131l\u0131m m\u00fchendisi.","url":"https:\/\/fatihsoysal.com\/blog\/author\/fatihsoysal\/"}]}},"yoast_meta":{"yoast_wpseo_title":"","yoast_wpseo_metadesc":"","yoast_wpseo_canonical":""},"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/34104","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/comments?post=34104"}],"version-history":[{"count":1,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/34104\/revisions"}],"predecessor-version":[{"id":34105,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/34104\/revisions\/34105"}],"wp:attachment":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/media?parent=34104"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/categories?post=34104"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/tags?post=34104"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}