PS IMAGE v1.2 — Diverso Fotorrealista (pessoas + tudo)

Evolução da v1.1 (só pets) para 100% mais variada, concisa, bonita e realista.

  • Base: runwayml/stable-diffusion-v1-5 512px
  • LoRA: rank 12 (~2.4M params, 9.2MB)
  • Dataset público variado (3395 imgs):
    • Caltech101 — 2595 imgs, 101 classes (Faces/pessoas com peso x2, airplanes, car_side, motorbikes, animais, objetos...)
    • Oxford-IIIT Pet — 800 imgs (gato/cachorro HQ)
  • Treino: 3000 steps, batch efetivo 4, lr 1e-4 cosine, fp16, Tesla T4
  • Prompts: ultra realistic photo of ..., detailed skin/fur, sharp focus, natural lighting, high quality

Resultado

grid

6/6 validados: rosto pessoa, mulher sorrindo no parque, avião sobre cidade, carro branco, Beagle, gato Persa.

Uso

from diffusers import StableDiffusionPipeline
from peft import PeftModel
import torch
pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, safety_checker=None).to("cuda")
pipe.unet = PeftModel.from_pretrained(pipe.unet, "Pedro21613/PS-IMAGE-1.2")
pipe.unet = pipe.unet.merge_and_unload()
img = pipe("ultra realistic photo of a person's face, detailed skin texture, sharp focus, natural lighting, portrait, high quality", num_inference_steps=35, guidance_scale=7.5).images[0]

Ver usar_v12.py.

Downloads last month
25
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Pedro21613/PS-IMAGE-1.2

Adapter
(2760)
this model