Instructions to use circlestone-labs/Anima with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use circlestone-labs/Anima with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Problem with upscale artifacts
Can anyone tell me how to avoid artifacts during second-pass resampling upscaling (hires.fix like)? I use euler_sde + beta, cfg 4.5, 30 steps. The upscale parameters don't really matter, artifacts appear in any case, even when upscaling with a model. Maybe it's because the sampler adds too much extra noise... Anyway i brightened the image a bit to make them easier to spot:
I understand there's an option to use tile upscale, but the result obtained this way doesn't match my goals. Meanwhile, the result I get using regular latent upscale without a model satisfies me 100%, but these artifacts ruin everything :(
So I would be very glad if anyone has a solution to this problem...
Excessive upscaling factors can cause artifacts. You can try a 1.5x upscale for 1024×1024 resolution, and a 1.3x upscale for 1536×1536 or resolutions between 1024 and 1536.
Excessive upscaling factors can cause artifacts. You can try a 1.5x upscale for 1024×1024 resolution, and a 1.3x upscale for 1536×1536 or resolutions between 1024 and 1536.
I understand that, but this is the default sdxl resolution of ~1mp 832x1264 with a 1.5x upscale to 1248x1872. Logically, it shouldn't cause any problems at all
Excessive upscaling factors can cause artifacts. You can try a 1.5x upscale for 1024×1024 resolution, and a 1.3x upscale for 1536×1536 or resolutions between 1024 and 1536.
I understand that, but this is the default sdxl resolution of ~1mp 832x1264 with a 1.5x upscale to 1248x1872. Logically, it shouldn't cause any problems at all
You can try adjusting the sampling steps to 30, or setting the denoising strength to 0.7–0.8.
You can try adjusting the sampling steps to 30, or setting the denoising strength to 0.7–0.8.
I've already tried that, the artifacts are still there :( Guess I should opt for tiled upscale after all, even though it means losing global context...
even though it means losing global context...
You should use multi diffusion instead of ultimate sd upscale https://github.com/Mooshieblob1/ComfyUI-MooshieTiledDiffusion
I'm having issues with the upscalers I used for Illustrious; for some reason, they produce a strange noise effect, especially on light elements. If anyone else is having the same problem, the best upscaler I've found is 4x_IllustrationJaNai_V1_ESRGAN_135k


