lshqqytiger / stable-diffusion-webui-amdgpu

Stable Diffusion web UI
GNU Affero General Public License v3.0
1.8k stars 186 forks source link

[Bug]: 使用tiled vae分块放大工具时,左下角出现灰色图块 #258

Open XingtaoZhou opened 1 year ago

XingtaoZhou commented 1 year ago

Is there an existing issue for this?

What happened?

使用分块算法放大图片时左下角会出现灰色图块 00002-2255691027-2

Steps to reproduce the problem

  1. 打开图生图,选择需要放大的图片 1693714711092

2.调整参数,放大2倍,点击生成 1693714839896

  1. 生成图片左下角出现坏块 00002-2255691027-2

What should have happened?

期望能生成一张完整的图片

Sysinfo

win10, AMD Radeon RX 6900 XT

What browsers do you use to access the UI ?

Google Chrome

Console logs

[Tiled Diffusion] upscaling image with R-ESRGAN 4x+ Anime6B...                          | 7/14 [00:36<00:36,  5.17s/it]
        Tile 1/9
        Tile 2/9
        Tile 3/9
        Tile 4/9
        Tile 5/9
        Tile 6/9
        Tile 7/9
        Tile 8/9
        Tile 9/9
[Tiled Diffusion] ControlNet found, support is enabled.
warn: noise inversion only supports the Euler sampler, switch to it sliently...
MultiDiffusion Sampling:   0%|                                                                  | 0/28 [00:00<?, ?it/s]MultiDiffusion hooked into 'Euler' sampler, Tile size: 96x96, Tile batches: 4, Batch size: 1. (ext: NoiseInv, ContrlNet)
[Tiled VAE]: input_size: torch.Size([1, 3, 1024, 1024]), tile_size: 512, padding: 32
[Tiled VAE]: split to 2x2 = 4 tiles. Optimal tile size 480x480, original tile size 512x512
[Tiled VAE]: Fast mode enabled, estimating group norm parameters on 512 x 512 image
[Tiled VAE]: Executing Encoder Task Queue: 100%|███████████████████████████████████| 364/364 [00:00<00:00, 3081.94it/s]
[Tiled VAE]: Done in 1.669sder Task Queue:  83%|█████████████████████████████▏     | 303/364 [00:00<00:00, 3027.26it/s]
MultiDiffusion Sampling: 100%|█████████████████████████████████████████████████████████| 28/28 [00:30<00:00,  1.10s/it]
Noise Inversion: 100%|█████████████████████████████████████████████████████████████████| 10/10 [00:29<00:00,  2.99s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 7/7 [00:50<00:00,  7.25s/it]
[Tiled VAE]: input_size: torch.Size([1, 4, 128, 128]), tile_size: 64, padding: 11       | 7/14 [00:46<01:11, 10.23s/it]
[Tiled VAE]: split to 2x2 = 4 tiles. Optimal tile size 64x64, original tile size 64x64
[Tiled VAE]: Fast mode enabled, estimating group norm parameters on 64 x 64 image
[Tiled VAE]: Executing Decoder Task Queue: 100%|████████████████████████████████████| 492/492 [00:00<00:00, 537.12it/s]
[Tiled VAE]: Done in 3.328s
Total progress:  50%|█████████████████████████████████▌                                 | 7/14 [00:49<00:49,  7.10s/it]
Total progress:  50%|█████████████████████████████████▌                                 | 7/14 [00:49<01:11, 10.23s/it]

Additional information

使用ControlNet里面的tiled也会出现这种情况

jaxinoman commented 12 months ago

我用RX590GME 8GB显卡也有同样的情况,左下角缺失,控制台显示MultiDiffusion到87%就停止了。修改潜变量尺寸也没有改变,而且我用的真实系模型,放大之后遗失了很多细节,产出的画面像是photoshop开了像素化滤镜的效果。改用N卡就一切正常。