H.D.Tài 🇻🇳

Train inpainting model/LoRA/TI (template)

1

Reference: https://github.com/huggingface/diffusers/blob/main/examples/research_projects/dreambooth_inpaint/train_dreambooth_inpaint.py This PR won't work due to aspect ratio bucketing ig, but it may be a good template.

Implement some features that can make TI training 6x faster

2

Cafeai from 東方Project AI used some cool tricks to make TI training 6x faster compared to auto webui. I don't understand them so I don't know if they has been...

Add preprocessors for ControlNet

1

Reference: https://github.com/lllyasviel/ControlNet/tree/main

[Feature Request] Add generating hand depth map tab

5

Quote from [Twitter user @toyxyz3](https://twitter.com/toyxyz3/status/1629620938894643200?s=20): > By separately rendering the hand mesh depth and open pose bones and inputting them to Multi-ControlNet, various poses and character images can be generated...

Thanks for the work you've done on Marigold. This is a revolutionary depth map estimator with extreme details compared to previous methods. Imo, one noticable application of it is [ControlNet](https://arxiv.org/pdf/2302.05543.pdf)...

Video depth map deflickering

3

Depth maps generated from video frames by Marigold is obiviously flickering. While this can be partially fixed by setting high n_repeat and fixed seed, such method is inefficent imo. Is...

Any recommended config for MarigoldPipeline?

1

I tried to compile [MarigoldPipeline](https://github.com/prs-eth/Marigold) on Google Colab, T4. While it worked out-of-the-box, the result isn't noticeably better compared to non-compiled pipeline.

feat: Add option to select ONNXRuntime Execution Providers for fastembed

3

### Related Issues - In some cases, fastembed does not use GPU despite `onnxruntime-gpu` and `fastembed-gpu` are installed correctly ### Proposed Changes: - Add option `onnx_providers` to set ONNXRuntime Execution...