Salom AIStable Diffusion — bepul AI rasm yaratishni o'rganish
Stable Diffusion — bepul, ochiq manbali AI rasm yaratuvchi. Kompyuteringizda ishlatasiz, internet kerak emas. Lekin GPU kerak.
1. Talablar
- Windows 10/11 yoki Linux yoki Mac (M1/M2/M3)
- Nvidia GPU 8GB+ VRAM (RTX 3060, 3070, 4060 va h.k.)
- 20 GB diskda joy
- 16 GB RAM
2. O'rnatish (Windows)
- Python 3.10 o'rnating
- Git o'rnating
- AUTOMATIC1111 WebUI'ni github.com/AUTOMATIC1111/stable-diffusion-webui'dan klonlang
- webui-user.bat'ni ishga tushiring
- Browser'da http://127.0.0.1:7860 oching
3. Asosiy modellar
- SD 1.5 — klassik, eng ko'p loRA bor
- SDXL 1.0 — yuqori sifat, ko'p resurs
- DreamShaper — fotorealistik universal
- Realistic Vision — odam yuzlari uchun
- Anime models — Counterfeit, Anything V5
4. Asosiy parametrlar
- Sampling steps — 20-40 (sifat)
- CFG Scale — 7-12 (promptga sodiqlik)
- Resolution — 512x512 (SD 1.5), 1024x1024 (SDXL)
- Sampler — DPM++ 2M Karras (eng yaxshi)
5. Yaxshi prompt
"masterpiece, best quality, ultra detailed, photorealistic portrait of a young man, soft lighting, 8k, sharp focus, professional photography"
Negative prompt ham muhim:
"low quality, blurry, distorted, ugly, deformed hands, watermark, text"
6. LoRA va embeddings
LoRA — qo'shimcha treninglar. Civitai.com'dan yuklab olib, /models/Lora papkasiga qo'ying. Mashhurlar: Detail Tweaker, Add More Details, EasyNegative.
7. ControlNet
Eng kuchli kengaytma. Eskiz, OpenPose, Canny, Depth — ma'lumotni boshqarish. Tijoriy ish uchun majburiy.
8. Cloud variant
GPU yo'qmi? RunDiffusion, Replicate, Hugging Face Spaces — soatlik to'lov bilan.
9. Salom AI bilan farqi
Stable Diffusion — texnik, o'rganish vaqt oladi, lekin to'liq erkin. Salom AI — bir click, o'zbek prompt, oson natija. Yangi boshlovchi uchun Salom AI, professional dizayner uchun SD.
💡 Maslahat: O'rganishni xohlamasangiz Salom AI, chuqur o'rganmoqchi bo'lsangiz Stable Diffusion.