1、stable diffusion git下載代碼

https://github.com/CompVis/stable-diffusion

安裝環(huán)境ldm

2、下載模型

從網(wǎng)站中 CompVis (CompVis)?下載compvis的模型，有如下，下載模型如果有問題，運(yùn)行中會(huì)提示錯(cuò)誤。需要重新下載

?3、運(yùn)行

到1的主目錄下，運(yùn)行：其中init-img是自己選的圖片，prompt是生成的提示詞

python scripts/img2img.py --init-img demoimg/man3.png --n_samples 4 --outdir outputs/img2img/0 --ckpt models/ldm/stable-diffusion-v1-4/sd-v1-4.ckpt --prompt “a men in black, tied hair and waist, tall and fit, deep eyes and gnarled beard”

4、bug1:

img2img.py 53:? w, h = image.size

print(f"loaded input image of size ({w}, {h}) from {path}") # 158*162

最終resize到128*160，為32的倍數(shù)，然而有bug，size 問題，提示如下。

解決：w, h = map(lambda x: x - x % 64, (w, h))

改成64的倍數(shù)即可——這里是因?yàn)閡net里有3次downsample，即unet的輸入需要是8的倍數(shù)。若32則僅保證4倍

Traceback (most recent call last):
? File "/cfs/cfs_dataV/workdir/Research_stage2/stable-diffusion/scripts/img2img.py", line 308, in <module>
? ? main()
? File "/cfs/cfs_dataV/workdir/Research_stage2/stable-diffusion/scripts/img2img.py", line 271, in main
? ? samples = sampler.decode(z_enc, c, t_enc, unconditional_guidance_scale=opt.scale,
? File "/data/miniconda3/envs/env-novelai/lib/python3.10/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
? ? return func(*args, **kwargs)
? File "/cfs/cfs_dataV/workdir/Research_stage2/stable-diffusion/ldm/models/diffusion/ddim.py", line 238, in decode
? ? x_dec, _ = self.p_sample_ddim(x_dec, cond, ts, index=index, use_original_steps=use_original_steps,
? File "/data/miniconda3/envs/env-novelai/lib/python3.10/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
? ? return func(*args, **kwargs)
? File "/cfs/cfs_dataV/workdir/Research_stage2/stable-diffusion/ldm/models/diffusion/ddim.py", line 177, in p_sample_ddim
? ? e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in).chunk(2)
? File "/cfs/cfs_dataV/workdir/Research_stage2/stable-diffusion/ldm/models/diffusion/ddpm.py", line 987, in apply_model
? ? x_recon = self.model(x_noisy, t, **cond)
? File "/data/miniconda3/envs/env-novelai/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
? ? return forward_call(*input, **kwargs)
? File "/cfs/cfs_dataV/workdir/Research_stage2/stable-diffusion/ldm/models/diffusion/ddpm.py", line 1410, in forward
? ? out = self.diffusion_model(x, t, context=cc)
? File "/data/miniconda3/envs/env-novelai/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
? ? return forward_call(*input, **kwargs)
? File "/cfs/cfs_dataV/workdir/Research_stage2/stable-diffusion/ldm/modules/diffusionmodules/openaimodel.py", line 751, in forward
? ? h = th.cat([h, hs.pop()], dim=1)
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 3 but got size 5 for tensor number 1 in the list.文章來源地址http://www.zghlxwxcb.cn/news/detail-819786.html

到了這里，關(guān)于運(yùn)行stable diffusion的文章就介紹完了。如果您還想了解更多內(nèi)容，請(qǐng)?jiān)谟疑辖撬阉鱐OY模板網(wǎng)以前的文章或繼續(xù)瀏覽下面的相關(guān)文章，希望大家以后多多支持TOY模板網(wǎng)！