Replies: 6 comments 1 reply
-
I was just going to suggest this... you might also submit a feature request through the Issues tab. |
Beta Was this translation helpful? Give feedback.
-
Could you implement it as an extension that has its own queue button, rather than having to modify the UI around the current generate button? Or is the architecture unsuitable for this approach? |
Beta Was this translation helpful? Give feedback.
-
Did this ever get anywhere? Seems like an important feature. I see how it could be tricky though. I would also like to be able to enqueue a job after the current one (skip the line). |
Beta Was this translation helpful? Give feedback.
-
Is there currently in base webui such queue system like easydiffusion has implemented it? There you can write your prompt, make settings and press the "Generate" Button. The whole task is stored in a queue. The queue will be started automatically. During generation of the task in the queue you are able to edit prompt (settings) and press again generate button. New task is also stored in the queue. You are able to stop tasks in the queue and you are able to get settings from a task in the queue to adapt them and again send them as new task into the queue. |
Beta Was this translation helpful? Give feedback.
-
check this out, i think this is useful. |
Beta Was this translation helpful? Give feedback.
-
the idea
a system that will buffer all jobs that need to be done and execute them in order without the need for a human
the current solution
use batches, file prompts or just wait for it to finish
current solutions limits
the current solution is limited to one type of job without user input.
if you want 100 images of cats after training over night you'd have to wake up to do that.
if you have a large batch running and have an idea, you'd have to wait or cancel the batch to try it
if you have multiple people, then they are both waiting for compute time while the other runs (with strange feedback currently)
slower computer users would benefit more from a queue as it would let their ideas flow without being limited by wait times (cough cough, like me)
the problem with implementation
the way the project is structured makes it hard to add a queue without a lot of work
for example the image output can no longer directly return from the job, the generate button giving progress makes less sense, special cases for "generate forever"
this small change is a big change
while I am going to code this for myself because I see it as a must have, the community and or auto might not think it right for this project so id like your feedback.
auto, my current implementation is well above 100 lines so consider this me asking if you want a pull request if I manage to get it working
Beta Was this translation helpful? Give feedback.
All reactions