Unfriendly to modifying

[skeskinen]

Hi, this is a great project and there is some excellent work done here to make all of this work at all.

But I must say that the structure of the project and the libraries makes it very painful to see what is actually going on.
Just a simple example: I want to disable the INFO logging -> look through documentation, find nothing -> look through DepthFlow code -> see that logging is done through ShaderFlow -> look through ShaderFlow documentation, nothing -> look at ShaderFlow code -> see that logging in ShaderFlow is done through Broken libraries so I go look at Broken core -> there are no obvious logging utils, so do a code search for logging and finally find that it uses loguru and LOGLEVEL env variable.

This is very simple example and doing anything more complicated is a whole project. I tried to upgrade the ComfyUI nodes to use 0.9 instead of 0.8, but I think I'll give up soon.

This isn't really a question or request or anything. Just an observation. Again, DepthFlow project itself is really amazing and ShaderFlow looks like a very useful framework. But I don't think it's very modifiable for external contributors the way that things are laid out.

[Tremeschin]

Yup, fair points! Some important context, thoughts, where to improve:

The main library you found is in the base monorepo, which is a ongoing experimental design for python projects, and not always obvious where things are without knowing the structure or intentions of the codebase, no surprises there's zero contribs (im fine). It allows me to work/start multiple projects easily, any improvement is global, I have lots of private stuff - one with a soft contract.

Most stuff is only code-documented as I take too long writing texts, organizing them, making it cohesive, and lay out all important information others do not know (I often lack Theory of Mind). It's also more of a personal collection of solutions over the years that happen to be splittable into well-defined projects. Honestly, this is the main problem, I'm only good at raw code currently.

I'll be taking some time to improve writing eventually, but the monorepo situation is unlikely to change. That isn't to say I couldn't have a simple document about its structure, idea, better examples, etc!

Anyhow, yes, LOGLEVEL=error should silence loguru, you might want TQDM_DISABLE=1 too for the progress bars. And v0.9 renamed and moved some stuff, it's better to wait akatz/me to port the nodes (middle of the priority list)

Hope it clarifies where the confusion came from, I'm aware many parts can be improved 🙂

[skeskinen]

Thanks for the response and I agree completely. Don't let anybody tell you how to structure your project. But if you ever feel like lifting DepthFlow from the monorepo, I think it could improve the contributing experience.

My suggestion for low hanging fruit to make the experience better for integrating (comfyui, custom python pipeline or otherwise) would be to add more code examples where numpy/PIL images go in and numpy/PIL images come out. Also with different camera motions, effects, inpainting, etc.

[Tremeschin]

Aye, I was working on most of these suggestions but side tracked to make own solution to executables, which is taking some development time away. Some recent/ongoing work after I backtrack on improving depthflow were/are/will be:

• Examples of Screenshot.py for getting a numpy image, and SideBySide.py special camera projection
• Unpublished documentation improvements like splitting each project into their own website ("it's possible" demo), making it less cross-referenced, render all presets videos to the website, post processing examples, etc
• Installing outside the monorepo now works, just git clone, pip install ., it grabs latest git+ of shaderflow and the main library, a build hook script adds the version when publishing / installed within the monorepo. Still some cross usage of packages, as the base shader engine needs to be its own package that depthflow depends on

I'm still thinking on how to rework the animation system in a better way which is the main problem. But anyways, thanks a lot for the friendly feedback! Definitely things I'll be improving when some side infrastructure work is done!