generate mech warrior image with diffusers

gen.py

import torch
import random
from diffusers import StableDiffusionKDiffusionPipeline
from diffusers import StableDiffusionPipeline, StableDiffusionLatentUpscalePipeline
from transformers import CLIPProcessor, CLIPTextModel
from diffusers.schedulers import DPMSolverMultistepScheduler



class MyPipe:

    def __init__(self, model_id, dtype=torch.float16):
        self.pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=dtype)

        # we skip 2 layers of the encoder
        self.pipe.text_encoder = CLIPTextModel.from_pretrained(model_id, subfolder="text_encoder", num_hidden_layers=10, torch_dtype=dtype)

#        self.pipe.set_scheduler('sample_dpmpp_2m') # it's like DPM++2M Karras, doesnt work with long prompt weighting
        self.pipe.scheduler = DPMSolverMultistepScheduler.from_config(self.pipe.scheduler.config)
        self.pipe.to("cuda")
        self.pipe.vae.enable_tiling()

    def get_pipline_description(self):
        pass

    def gen(self, inputs, upscale=False):
        if 'seed' in inputs:
            inputs['generator'] = torch.Generator("cuda").manual_seed(inputs.pop('seed'))
        image = self.pipe(**inputs).images[0]
        return image


if __name__ == '__main__':
    path = "/home/imgen/models/icb_diffusers_final/"
    path = "/home/imgen/models/icb_diffusers/"
    p = MyPipe(path, dtype=torch.float32)
    seed = 2001481868 
    inputs = dict()
    inputs['seed'] = seed
    inputs['guidance_scale'] = 4.5
    inputs['height'] = 1024
    inputs['width'] = 512
    inputs['num_inference_steps'] = 25
    inputs['prompt'] = 'close up Portrait photo of muscular bearded guy in a worn mech suit, (snow:1.4) , ((light bokeh)), intricate, (steel metal [rust]), elegant, sharp focus, photo by greg rutkowski, soft lighting, vibrant colors, masterpiece, ((streets)), detailed face'
    inputs['negative_prompt'] = 'nude, cross eyed, tongue, open mouth, inside, 3d, cartoon, anime, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), skin spots, acnes, skin blemishes, bad anatomy, red eyes, muscular'
#    inputs['use_karras_sigmas'] = True
    image = p.gen(inputs)
    image.save('mech_beard.jpeg', comment='snvlfjvnf' * 100, quality=100, subsampling=0)