Open tkittich opened 1 year ago
Thank you for the extension. It would be great if this extension could be used with Dynamic Thresholding (CFG Scale Fix). Currently it doesn't seem to work together. I got the following error messages when using both extensions together:
MultiDiffusion hooked into 'Euler a' sampler, Tile size: 96x96, Tile batches: 4, Batch size: 8. (ext: ContrlNet) [Tiled VAE]: input_size: torch.Size([1, 3, 2048, 3648]), tile_size: 3072, padding: 32 [Tiled VAE]: split to 1x2 = 2 tiles. Optimal tile size 1792x1984, original tile size 3072x3072 [Tiled VAE]: Executing Encoder Task Queue: 100%|███████| 182/182 [00:06<00:00, 29.35it/s] [Tiled VAE]: Done in 7.372s, max VRAM alloc 16676.609 MB 166/182 [00:06<00:00, 43.32it/s] 11:02:13-947530 ERROR Running script process batch: D:\theera\aiart\automatic\extensions-builtin\sd-dynamic-threshold ing\scripts\dynamic_thresholding.py: AttributeError ╭────────────────────────── Traceback (most recent call last) ───────────────────────────╮ │ D:\theera\aiart\automatic\modules\scripts.py:413 in process_batch │ │ │ │ 412 │ │ │ │ args = p.per_script_args.get(script.title(), p.script_args[scrip │ │ ❱ 413 │ │ │ │ script.process_batch(p, *args, **kwargs) │ │ 414 │ │ │ │ s.append(f'{script.title()}:{round(time.time()-t0, 2)}s') │ │ │ │ D:\theera\aiart\automatic\extensions-builtin\sd-dynamic-thresholding\scripts\dynamic_t │ │ hresholding.py:127 in process_batch │ │ │ │ 126 │ │ if p.sampler is not None: │ │ ❱ 127 │ │ │ p.sampler = sd_samplers.create_sampler(fixed_sampler_name, p.sd_mode │ │ 128 │ │ │ │ D:\theera\aiart\automatic\extensions-builtin\multidiffusion-upscaler-for-automatic1111 │ │ \scripts\tilediffusion.py:373 in <lambda> │ │ │ │ 372 │ │ sd_samplers.create_sampler_original_md = sd_samplers.create_sampler │ │ ❱ 373 │ │ sd_samplers.create_sampler = lambda name, model: self.create_sampler_hij │ │ 374 │ │ │ name, model, p, Method(method), │ │ │ │ D:\theera\aiart\automatic\extensions-builtin\multidiffusion-upscaler-for-automatic1111 │ │ \scripts\tilediffusion.py:445 in create_sampler_hijack │ │ │ │ 444 │ │ # create a sampler with the original function │ │ ❱ 445 │ │ sampler = sd_samplers.create_sampler_original_md(name, model) │ │ 446 │ │ if method == Method.MULTI_DIFF: delegate_cls = MultiDiffusion │ ╰────────────────────────────────────────────────────────────────────────────────────────╯ AttributeError: module 'modules.sd_samplers' has no attribute 'create_sampler_original_md' 100%|██████████████████████████████████████████████████████| 2/2 [00:05<00:00, 2.85s/it] [Tiled VAE]: input_size: torch.Size([1, 4, 256, 456]), tile_size: 192, padding: 1185s/it] [Tiled VAE]: split to 2x3 = 6 tiles. Optimal tile size 160x128, original tile size 192x192 [Tiled VAE]: Executing Decoder Task Queue: 100%|███████| 738/738 [00:13<00:00, 53.13it/s] [Tiled VAE]: Done in 15.064s, max VRAM alloc 12484.830 MB733/738 [00:13<00:00, 45.90it/s] MultiDiffusion Sampling: 19%|█████▎ | 3/16 [00:29<02:08, 9.86s/it]
it was working fine until I updated it a few days ago weirdly enough, with this same error after multidiffusion tries hooking after the 50% mark.
bump
+1
Thank you for the extension. It would be great if this extension could be used with Dynamic Thresholding (CFG Scale Fix). Currently it doesn't seem to work together. I got the following error messages when using both extensions together: