I’m curious what you think about this concept: a Blender add-on that takes your existing Render Result, enhances it with AI, and then loads the generated image back into Blender’s Image Editor.
The basic workflow would be:
Create your scene, camera and composition in Blender
Render it normally
Describe the materials, lighting or atmosphere you want to explore
Generate an AI-enhanced version
Compare the original and generated result directly inside Blender
The idea is not to replace Cycles or Eevee. Blender would still define the geometry, camera, proportions and composition. AI would mainly help with faster material studies, atmosphere variations, early concepts and presentation images.
Here is the concept and current prototype:
Would something like this be useful in your workflow?
I would also be interested to hear what level of control you would expect from such an add-on. Prompts, strength settings, masks, render passes, depth maps, material IDs, viewport captures?
And do you know of any existing Blender add-ons that already offer a similar workflow?
I’m aware of SketchUp AI, but I’ve also seen quite a few users who are not very happy with it because it tends to hallucinate and change parts of the design too much.
Rendero should offer much better control over specific details. One idea would be to use Blender’s built-in Cryptomatte or material IDs as the default definition of individual objects, materials, and surfaces. That could give the AI add-on almost pixel-level control over what is allowed to change and what should remain untouched.
By the way, do you work in architecture or interior design?
I also do interiors or construction as a private project.
I prefer to work it out on my own rather than relying on AI.
I think it would be good to use AI for simple tasks, but it seems that AI needs as much effort as improving its quality in 3D in order to produce the desired results without hallucinations.
It’s different from what regular users usually lean on AI (feels like using fancy filters), but it doesn’t seem bad.
Time red tape seems to help, but I don’t think we know if we’ll meet the expectations for AI use. (Basic expectations for AI use are likely to be high)
I think it’s a mix of art and science to tease out the very last detail of realism from any renderer one can think of.. so my biggest question would be:
If i’m not yet that expert to know how some rendereing should look and so tweak some setting to achieve this goal.. then how do i know if any AI will give me that last bit of realism or does it “just” copy some general look which in fact would make my rendereings maybe even look more “fake” or unprofesionel..
..if you know what i mean.
But that maybe would derail into some too general topic about AI..
Why would I want something that randomly changes subtle details of my work without my control? When I’m making fried rice, I don’t want my frying pan to randomly decide to add a banana to the mix
I’m not too much into “realistic” renders.. as i said: i’m not enough “expert” to know if any these really make my things better.. i would guess scene composition and also “light design” would be the next thing to tackle first for me..
Sure, that’s a great point. Even still, I like knowing exactly what is in my render and why, and that seems doubly important with client facing work. If a render enhancer adds some tropical trees to a yard that would cost thousands of dollars to actually plant, for example, I could see that being an awkward conversation with the client
Well, that depends on one thing: do you really care about the exact position of every single grain of rice, or about the overall image of the fried rice?
If the goal is to create a compelling render, why be perfectionistic about tiny, unimportant details? Is painting about being pixel-perfect, or is it about capturing the right composition, atmosphere, and overall feeling?
I think you’re missing my point on purpose, but that’s fine, I’m not the target audience for this and I’m not going to argue with you about it. Carry on, I’ll leave
For me personally, a generative step after the render would defeat the purpose of using Blender. This is a hobby for me, I am under no time or budget pressure when working. I would rather work to improve my skills to achieve the result I want using the normal predictable and reproducible tools, even if it potentially takes longer.
I understand if others may have different reasons for using Blender, but it’s not clear to me what would be gained from implementing this directly in Blender. My understanding is that it takes a lot of iteration to get what you want from a generative workflow. You would want to store multiple results along with the prompts for each, and allow the user to compare and adjust. Would it be easier to implement the UI for this inside Blender as opposed to making a separate tool?
on point. Or on grain?