Sana
Sana copied to clipboard
SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer
Hi, I'm planning to evaluate the reproduced Sana-Sprint model alongside the pretrained one. I’ve reviewed the [Inference & Test Metrics guide](https://github.com/NVlabs/Sana/blob/main/asset/docs/metrics_toolkit.md) and the scripts in ./Sana_sprint/scripts/. However, I noticed that...
hi everyone can you help me about this eror. I think my hardware is not enough for this model. I tried 2 diffrent workflow but it doesnt work. specs of...
Add [uv](https://github.com/astral-sh/uv) (package manager) and [ruff](https://github.com/astral-sh/ruff) (linter/formatter), both Rust-based tools. If merged, CI can switch to ruff for linting.  
Thank you for releasing SANA-Sprint code! It seems like SANA-Sprint doesn't support xformer memory efficient attention? https://github.com/NVlabs/Sana/blob/d88cb447bf83c73d72ce084a215db5968c4cf2cf/app/sana_sprint_pipeline.py#L28 I am wondering if this is correct from my understanding, and if so,...
Hi, I have extensively tried both the Sana Versions in COmfy UI and Your Inference. There is a significant difference, and it is **due to the lack of Flow-DPM-Solver support...
I get (sana) U:\Sana-for-Windows>conda install nvidia/label/cuda-12.6.0::cuda-toolkit C:\Users\cleve\miniconda3\lib\site-packages\conda\base\context.py:202: FutureWarning: Adding 'defaults' to channel list implicitly is deprecated and will be removed in 25.3. To remove this warning, please choose a default...
Great work, is there any possibility for GGUF quantizations? Especially for 4.8B.
When I use `SanaControlNetPipeline` in Kaggle T4*2, its response is `TypeError: SanaControlNetPipeline.from_pretrained() got an unexpected keyword argument 'device_map'`. Please add `device_map="balanced"` to support for multi-GPUs inference as possible as `diffusers`.
Repository: https://huggingface.co/Swarmeta-AI/Twig-v0-alpha We are not sure about the specific license requirements, it seems to be Apache2.0, right? This model, Twig v0 alpha, is a multilingual text-to-image (T2I) model based on...
Hi my flow model currently uses/trained with sampling in continuous time space [0,1] with some tricks like time shifting . how do i go about using dpmsolver++