Programmiererforum

#!/usr/bin/env python3.10

import torch
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler

pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float32)
pipe.load_lora_weights("jennlaw_v1.1.safetensors", adapter_name="faceTest", weight_name="faceTest")
pipe.set_adapters(['faceTest'], adapter_weights=[0.75])
pipe.to('mps')
pipe.fuse_lora(lora_scale=0.75)
pipe.enable_attention_slicing()
pipe.unet.config.clip_skip = 2
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.config)

prompt = "cinematic photo of jennlaw, studio quality, woman standing in a room, blur the background, smiling, having a positive face, wearing a v-neck black sweater, jennlaw, detailed lips, blonde, "
negative_prompt = "asian, chinese, busty, teeth, cartoon, painting, illustration, (worst quality, low quality, normal quality:2), (moles:1.1),"

images = pipe(prompt,
negative_prompt=negative_prompt,
guidance_scale=5,
num_inference_steps=31,
width=768,
height=1024,
generator=torch.manual_seed(2721872758),
num_images_per_prompt=2).images

for idx, img in enumerate(images):
img.save(f"outputs/{idx}.jpg")

Ich versuche, DiffusionPipeline zu verwenden, um ein Bild zu generieren
[list]
[*]Basismodell ist SD 1.5
[*]Lora-Gewichte geladen von https://civitai.com/models/111828/jennifer-lawrence
[*]Läuft auf Mac M3 Max
[*]Eingabeaufforderung, negative Eingabeaufforderung, Startwert, Clip-Überspringen und andere Dinge werden von https://civitai.com/images/1619242 kopiert
[/list]
Aber das Ergebnis ist seltsam und unterscheidet sich von der gleichen Konfiguration auf civitai,
[img]https://i.sstatic.net/F0LU872V.jpg[/img]

[img]https://i.sstatic.net/Hxw8n4Oy.jpg[/img]

Ich glaube, das Problem muss auf meiner Seite liegen, aber ich konnte nicht sagen, welcher Teil falsch ist . Weiß es jemand?
Habe mein Skript angehängt.
[code]#!/usr/bin/env python3.10

import torch
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler

pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float32)
pipe.load_lora_weights("jennlaw_v1.1.safetensors", adapter_name="faceTest", weight_name="faceTest")
pipe.set_adapters(['faceTest'], adapter_weights=[0.75])
pipe.to('mps')
pipe.fuse_lora(lora_scale=0.75)
pipe.enable_attention_slicing()
pipe.unet.config.clip_skip = 2
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.config)

prompt = "cinematic photo of jennlaw, studio quality, woman standing in a room, blur the background, smiling, having a positive face, wearing a v-neck black sweater, jennlaw, detailed lips, blonde, "
negative_prompt = "asian, chinese, busty, teeth, cartoon, painting, illustration, (worst quality, low quality, normal quality:2), (moles:1.1),"

images = pipe(prompt,
negative_prompt=negative_prompt,
guidance_scale=5,
num_inference_steps=31,
width=768,
height=1024,
generator=torch.manual_seed(2721872758),
num_images_per_prompt=2).images

for idx, img in enumerate(images):
img.save(f"outputs/{idx}.jpg")
[/code]