Comments (18)
Let's call the image I upload "base image"
If I set random seed and batch 5 images
Does each of the batch images generate from the original "base image" or does
batch image #2 get generated from batch image #1 THEN
batch image #3 get generated from batch image #2
and so on
from fooocus-mre.
@barepixels Good points, discussions enabled.
As for your question - both modes you described make sense (single image, and cycling over different images when more of them is available), but currently all images generated with img2img are using first image from the gallery as an input (so you can see different variants of it, generated using different seeds).
from fooocus-mre.
@barepixels I've just implemented cyclic over input images as the default behaviour for img2img (available on moonride-main). So for example if you have image A, and image B as inputs, and batch size 4, then you will get 2 variants of image A, and 2 variants of image B.
And if you would like to have 4 variants of image A, then you simply load only image A as the source image.
Let me know if it works for you.
from fooocus-mre.
Loading more than one images (A & B) works
Not sure if I will ever load more than 1 because I want to focus on 1 task at a time.
What I would like to see is to have the original uploaded image "A" in the gallery. This will serve 2 purpose
a) easy to compare before and after
b) Ensure all images generated with img2img are using the first image from the gallery (the uploaded image "A") as input
from fooocus-mre.
@barepixels That could complicate UI, and I'd like to keep initial view as simple as possible - I will need to think a bit about how those function could be added to the UI, without making it cluttered.
from fooocus-mre.
Maybe I have not said it clearly. Let me try to illustrate it with Photoshop. Hang on
from fooocus-mre.
This is what I am asking for
After generating 1 using img2img, to include the original (left image) with the generated version (right). If more than 1 generated, the original (left image) is always at the top of the list
This will serve 2 purpose
a) easy to compare before and after
b) Ensure all images generated by img2img are using the first image from the gallery (the uploaded image on the left) as input
This does not affect the existing elegant interface. No extra button or tab
from fooocus-mre.
@barepixels Okay, that makes sense - I will try to figure out something like this.
from fooocus-mre.
Thank you.
from fooocus-mre.
@barepixels I have initial idea for for the layout (and initial buggy implementation - it's currently messed up when preview is displayed, etc.) - I'd like to use two tabs for separate inputs and outputs galleries, it would look like that:
from fooocus-mre.
that would work. plus the preview will be bigger :-)
from fooocus-mre.
@barepixels Implemented, available on moonride-main :).
from fooocus-mre.
Excellent my friend. I can compare let's say #3 vs the original just by clicking on the two tabs back and forth. :-)
Not exactly sure how the Output to Input feature works. I assume:
- Click the Output to Input button
- Highlight (orange border) which I want to use for the "Output to Input" feature
- Choose how many
- Press "Generate"
Can you confirm this is correct?
from fooocus-mre.
@barepixels Selection doesn't matter. If you have 1 image loaded as input, then it will use it for every generated output. If you will load 2, then it will cycle over these 2. So with imported images A and B, and batch size 8 you would get 4 variants of A, and 4 variants of B. If you loaded just A - you would get 8 variants of A.
You can also put inputs of completely different resolutions and aspect ratios - it will automatically upscale (or downscale) input image to roughly 1 megapixel each. And if you didn't set any input, but generated already something to Output - it will automatically import that for further processing :).
Default img2img settings allow it to change input image quite a lot, and use it mostly as a rough guidence - if you want to verify how it works and see similarity between images more clearly, then you can lower Denoise (in Advanced tab) to let's say 0.7 or 0.8 - it will preserve much more from the input images, then.
from fooocus-mre.
Here is a potential problem I might have (I think)
- I upload an image for img2img
- I set a style I want
- I set 32 to generate
- It will give me 32 versions to compare :-)
- NOW if I press output to input
- It will send ALL 32 to "input tab" (that is a lot, maybe too much, especially if 30 of them I do not like) I will have to wait for a long time OR quit CMD and relaunch since there is no abort button
As for Denoise (in Advanced tab), default minimum is at 0.5? What if I want 0.2
from fooocus-mre.
@barepixels For now in situations like that simplest solution would be simply using "Load Image(s)", and picking those you want from the output folder.
I guess it would be nice to be able to send selected to the input gallery, but I am not sure how to proceed about that - I will need to take a deeper look into Gradio and Gallery in there, to see if and how it could be done.
Theoretically you could use very low denoise like that, but it wouldn't change input much - but if you find it useful / needed, then sure - I'll change it.
from fooocus-mre.
Thank you for changing default Denoise.
The "simplest solution would be simply using "Load Image(s)", and picking those you want from the output folder." Agree, this is good enough. Sometimes too much is bad
I think you should stop/drop working on the "Output to Input" and focus on other modifications... such as Abort button. haha
I tried to just reload the browser hoping to abort the queue but it didn't work ... because there was no signal to tell Fooocus to stop. I felt stupid. haha
from fooocus-mre.
Next thing worth doing would be probably preparing easy to use (like download + click on .bat) package for the less experienced users. But both this and aborting the processing (also worth implementing - I agree on that) are separate issues / discussions, so let's stop it here. I consider img2img in current state as good enough for typical use cases, so I am closing this one :).
from fooocus-mre.
Related Issues (20)
- JugernautXL not recognized HOT 2
- I get an error when I launch the tool - 'Lora file not found!' HOT 2
- Generates without any problem first time, then runs out of ram.
- The installation of V2.0.78.5 failed many times and the following error message appeared. What is the reason? HOT 1
- weird preview generating problem here and in the main repo
- My boss, MRE version 2.1 ready? HOT 12
- load_settings, e: 'utf-8' codec can't decode byte 0xd7 in position 1206: invalid continuation byte HOT 1
- Extreme RAM usage on colab
- fooocus-MRE loads model on every startup HOT 1
- All of a sudden
- How to use SDXL CHECKPOINTS that are in subfolders?
- [FIXed] Colab Error WARNING[XFORMERS] HOT 1
- running an error message
- Error while deserializing header: HeaderTooSmall
- ERROR: Failed building wheel for pygit2sudo
- paths.json not recognized
- Colab Error – xFormers wasn't build with CUDA support
- Ordering of Models and Loras
- Running different models similar to running different styles
- complaints
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from fooocus-mre.