varunshenoy / opendream Goto Github PK

I am beginner in this AI world, trying to get this UI running in ec2 instance but having some difficulties getting it to work. I have managed to launch the UI using sh run_opendream.sh and get the UI loading in port 3000. However, I cannot get anything else working other than loading the UI. Here is the error shot from the inspect element, it seems to point something at port 8000. How do I get this working in ec2?

how do we create a mask? follow user mouse? etc

Should be written this requires npm.

Might be missing other requirements? I'm still trying to get this running and hitting issues where it seems like no extensions are loaded.

Once a task is given, the modal should disappear and push the job onto some queue on the backend. This way you can launch a bunch of diffuse tasks at once.

How can I load local models/extensions easily like webuiautomatic1111, the workflow is great, but when I start and run, I can use very little, is there any tutorial, thank you！

generating mask for (758, 876, 4)
set_torch_image input must be BCHW with long side 1024.

with body.png. test.png works fine.

should allow for options to download in various formats/resolutions or maybe simple stub/default for now

Just use openCV or something to perform ops such as (increase saturation, gray out, zoom, sharpen, etc.). Show a good mixture of stable diffusion and basic image processing operations.

In case you create a mask layer that you don't like, for instance

There's a lot of places in the code where we implicitly/explicitly support only 512 x 512 images. If we want to add tools like "generative fill", we need to drop this dependency.

Mikubill/sd-webui-controlnet#1464

Use case: network produces several images. Code currently places all images in one layer. We don't want this.

easier for users. we can just grep all of the files

Attempted to add an instruct_pix2pix layer using the default install.

Nothing in the logfile other than the above error.

INFO:     127.0.0.1:61305 - "GET /schema/instruct_pix2pix HTTP/1.1" 200 OK
INFO:     127.0.0.1:61305 - "POST /operation/instruct_pix2pix/ HTTP/1.1" 307 Temporary Redirect
'str' object has no attribute 'get_image'
INFO:     127.0.0.1:61305 - "POST /operation/instruct_pix2pix HTTP/1.1" 500 Internal Server Error

Mac M2, latest main commit.

workflow.json: https://gist.github.com/mmastrac/15d1f94100a8b312fb96d0dae6ca063e

Similar to .ipynb, we want extensions to be baked into the workflow that is exported.

Easy way to get setup

operators should return multiple layers by default (not just one)

Hello. I just came across this project and it seems interesting.

Perhaps a simple "hello world" style workflow demo would be helpful. You know showing something like a screencast of running the app, installing a new model, a new extension, and showing how to update if possible. Also it would be helpful to also show what the app is capable of. Reading around the repo, I'm not sure how the app works, or how to install models, if that is at all possible to begin with.

I realize most projects have limited manpower to do multiple things, but demos and plenty of docs are crucial to encourage people to try things out.

Anyway,
Best of luck

not that hard to do

In the logs: "POST /operation/mask_and_inpaint HTTP/1.1" 500 Internal Server Error

I attempted to use mask and inpaint with an image that I uploaded myself (attached) and a simple layer I painted with pen.

What more logs or other details can I attach to help debug it?

and SAM allow for semantic masks

Should be able to see prompt etc and experiment with slight changes.

Options should be to "create new layer with changes" or "update layer"

Create a simple Python extension file that can handle Stable Diffusion with LoRAs. It might make sense to create a separate command (not dream).

add a more comprehensive README (outlining design decisions)

Should not be sending empty strings

Hello,

Is it possible to add SD XL support?

I'd love a colab notebook for this as well.

Thanks!

more generally fix these

for SAM/masking, Kayvon mentioned this