sophrocyne

Time to vibe code the data into "That supa awesome database over there"

For context, I'm Invoke's CEO.

I'm not going to argue that they're not complicated tools - Invoke and Comfy are both actively used to push the boundaries of what can be achieved in professional generative media.

But I'd be curious to hear what issues you ran into with Invoke, and where you found issues with the community -- Or, are you generalizing across SD reddit?

We try to do a pretty good job keeping things welcome, and are nearing 100 videos of educational content to help folks learn how to use GenAI. Open to feedback on where we can promote more accessibility.

Free lunch means it doesn’t require training to achieve the results - it solves for it at run time.

You can use this in local generative media tools like Invoke or Comfy, which are both open source and free.

Forgot to post some context.

I'm the CEO at Invoke (i.e., the creator of this piece) -- we're one of the longest running projects for open-source image generation, originating from the initial explosion around Stable Diffusion's release.

We've always focused more heavily on professionals/artists, and are happy to have (finally) been able to get some clarity on these points as we've pushed the USCO to respect where/how human creativity factors into GenAI usage.

You can learn more about us invoke.com (and download the local studio at invoke.com/downloads)

Take care HN.

Hey all. I'm the CEO of Invoke - appreciate everyone who has mentioned us in the thread.

To OP -- We work with professional artists regularly, and I'm seeing things pick up as more begin to understand the potential for creative control. Artists mainly want to be afforded creative flexibility and control, and need an interface that feels natural for their workflow.

Invoke is OSS, we release continued training/education on a weekly basis (free, on YT) and we'll be releasing a simplified installer soon.

Invoke is model agnostic, and supports Flux, including quantized versions.

The Invoke team released regional guidance using IP Adapter a few months ago, which can use color palettes + style transfer mode, along with text prompts and controlnets.

Would take a look at that for some inspiration -- The UI is Apache 2.0 and used by professional artists. I'd be curious how you think it performs relative to the workflow you've developed.

You're spot on that researchers don't always build the UI that end-users want to use. Always love to see people thinking about the creatives. Good work!

Some perspectives from someone working in the image space.

These tests don't feel practical - That is, they seem intended to collapse the model, not demonstrate "in the wild" performance.

The assumption is that all content is black or white - AI or not AI - and that you treat all content as equally worth retraining on.

It offers no room for assumptions around data augmentation, human-guided quality discrimination, or anything else that might alter the set of outputs to mitigate the "poison"