Feature Request: Add a node with the Stable Diffusion image, so compositor effects can be applied afterward

[biroMann]

Hello, and thank you for the great job. I will tag and share an animation as soon as I complete it.
...
I saw in previous versions that an AiR-Node Group was automatically generated in Compositor, for what I see it has been removed.
I think that it was useful because with it you could switch on/off the AiR effect just disconnecting it from the node tree or muting it(in this case also deselecting "Run Automatically on Render"), in this way it wasn't necessary a new workspace section(AI Render), it was enough the common Compositor section plus the AiR-Node Group.
The second function that I found interesting(and that now I don't know how to make) about AiR-Node Group was that in the node tree I was able to add other effects AFTER AiR-Node Group, for instance to make some adjustments on colors because some SD checkpoints modify slightly them.

So, in my opinion, just adding AiR-Node Group in Compositor without new AI Render section, was working better, but maybe it's just my experience.

Thank again for this amazing add-on.

Additional information

No response

[benrugg]

Yeah, I did like the node group for the second reason you mentioned - the ability to add effects or color modifications after Stable Diffusion. I ended up removing it when I added upscaling, because Blender doesn't have a way of allowing a larger image in the compositor than whatever was rendered. (So as far as I can tell, there's no way to keep the workflow of a node group now that upscaling is a feature).

[biroMann]

Ok, thank you very much for the answer.
Not to reopen the issue, I see it's closed, but just to understand if it possible to find a solution.
It could work if the upscaled rendered image is read through a RenderResultNode inside an AirNodeGroup. Maybe the Group can have different resolutions?

[benrugg]

You can read it back into an image node, but the final image will still be the original size (so an upscaled 2048x2048 for example, would still be rendered as 512x512).

If you want to do something similar to the workflow you had before, you can manually load in the upscaled image through an image node, add any nodes after that, and connect it to the output. And then you'll just need to render something quickly at the final output size, and the compositor will kick in.

[biroMann]

But after that I have to render another time the sequence. Consider that like in my case, if you are working not on a single image but on a clip with multiple scenes and sequences in which you must intervene often, you have to find and change the image in ViewerNode every time you change frame just to see the result on the particular bunch of frames, and then rerender, in the previous workflow you just hit F12 and see result, if it's ok you keyframe adjustments and lunch the final render. In future advances, in the case should be possible, I think integrate the previous workflow it would be much better for animators. Anyway maybe I find a work around.
Thank you again.

[benrugg]

Yes, that is a good point. The new way isn't a good workflow for animations, if you want to apply any compositing effects after Stable Diffusion.

I will reopen this issue as a feature request, and maybe someone can provide insight on how this would be possible with an upscaled image.

(One thing I will suggest as a possibility is that you could render out all the Stable Diffusion images, and then bring the entire image strip into Blender as a node, to do the compositing afterward. I think this is doable).

[biroMann]

Thank you very much. Yes, the suggestion is correct and it's what I am doing, anyway it's cumbersome, animation projects become complex and direct workflow it is necessary, I deal with image sequences of 150/500 frames, that have to be visually coordinated, so Air and AdjustmentNode are continuously tweaked and often keyframed; a direct output it's important to have an harmonious visual baseline.
Upscale upgrade it's wonderful in obtaining from an input image/a 3d object artistic and coherent results. I have already seen this feature in SD, and it is an "evolutionary leap", I would use previous version of the add-on, but the difference with Upscale it is consistent.

So to summarize the issue from the prospective of animator, what in my opinion would be the best is:
An AirNode that could be mute or unmute to could use the CompositorWS with an output path and AirWS with another output path; in this way, as happen to me, when I want to overlap a basic image sequence and an Air image sequence I can send two render with same adjustments.
It could also work just 1 CompositorWS with an AirNode, for me this is the leaner and most direct solution.

I attach an image with a CompositorTree scheme.

This method allows to verify all the necessary adjustments just lunching the RenderImage command.

Anyway these are just suggestions to improve the add-on, that is already a great tool. Thanks again.

[benrugg]

Thank you for writing this up. And for anyone coming across this in the future who may want to work on this feature, the summary of the core issue is that Blender's compositor only works with the image resolution that was rendered, so in this case it would be before it has been upscaled.