automatic/TODO.md

TODO

Issues

Stuff to be fixed...

• ClipSkip not updated on read gen info
• Usage of sd_vae in quick settings
• Run VAE with hires at 1280

Features

Stuff to be added...

• Add Gradio theme maker
• Create new GitHub hooks/actions for CI/CD
• Move Restart Server from WebUI to Launch and reload modules
• Redo Extensions tab: see https://vladmandic.github.io/sd-extension-manager/pages/extensions.html
• Stream-load models as option for slow storage
• Autodetect nVidia and AMD: nvidia-smi vs rocm-smi

Investigate

Stuff to be investigated...

• Best practices for Apple M1
• Best practices for AMD GPUs
• Torch Compile
• Torch-DirectML
• TensorRT

Merge PRs

Pick & merge PRs from main repo...

• Merge backlog: https://github.com/vladmandic/automatic/pulls

Models

StabilityAI is working on new stuff...

• SD XL
• SD ReImagined

Integration

Tech that can be integrated as part of the core workflow...

• Merge without distortion
• Weighted merges
• Null-text inversion
• Custom diffusion, Custom diffusion
• Dream artist

Random

• Bunch of stuff: https://pharmapsychotic.com/tools.html

Pending Code Updates

• fix VAE dtype
  should fix most issues with NaN or black images
• add built-in Gradio themes
• fix requirements
• more AMD specific work