[docs] Quickstart #12128
[docs] Quickstart #12128
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
this is just my opinion and maybe an unpopular one too, but since we're a fully open source library with an apache 2.0 license and this is the entry point for most users, we probably should also do an example with a similar model, using a non-commercial and gated repo for the beginners is not an ideal UX or a showcase for those that want to use diffusers in a commercial product. |
|
Happy to add an example with a similar model, did you have one in mind? |
|
Qwen-Image seems like a good alternative right now, it has the first trending position for image generation models and it's third overall. Same for HunyuanVideo, WAN 2.2 is a lot more popular with an Apache 2.0 license and while HunyuanVideo is not a gated repo, it has a custom license. |
|
Updated to use Qwen-Image and Wan 2.2! |
docs/source/en/quicktour.md
Outdated
| pipeline = DiffusionPipeline.from_pretrained( | ||
| "Qwen/Qwen-Image", | ||
| torch_dtype=torch.bfloat16 | ||
| ).to("cuda") |
There was a problem hiding this comment.
Nit:
| pipeline = DiffusionPipeline.from_pretrained( | |
| "Qwen/Qwen-Image", | |
| torch_dtype=torch.bfloat16 | |
| ).to("cuda") | |
| pipeline = DiffusionPipeline.from_pretrained( | |
| "Qwen/Qwen-Image", torch_dtype=torch.bfloat16 | |
| ).to("cuda") |
There was a problem hiding this comment.
how about using the new device_map="cuda" when loading the pipelines?
There was a problem hiding this comment.
Updated to use device_map="cuda" in the recently refreshed docs to increase visibility :)
Refactors the Quickstart to provide a more complete snapshot of what Diffusers offers rather than focusing on how the components in a pipeline works. It was a good opportunity to include/link to some of the cool optimization stuff @sayakpaul and @a-r-r-o-w have been working on, which I consider to be a pretty important aspect of the library.
This is a pretty high visibility doc, so let me know if there is anything else you think we should include or is missing. For example, should we include Hybrid inference and Modular Diffusers, or are these more experimental still?