What is going on with the image generator right now? Prompts seem to be hit and miss with the AI following only following the main prompt and completely ignoring anything in the negative prompt. It's even messing up with the main prompt. Another thing is that it is generating pictures with a plain white background despite me specifying the environment and surroundings in detail in the main prompt, on top of that the images are no where near as good as they were a few weeks ago and it keeps generating anime or bad cartoon images which I do not want and did not specify in my prompts or art styles. I mainly use oil painting realism or professional photo so seeing the anime and cartoon styles is really frustrating, is this going to be fixed?, is something being updated right now that's causing all this? Is anyone else experiencing the above with the text to image generator?

you are viewing a single comment's thread
view the rest of the comments

The "Rev.4" baseline is out, but that was just the warm-up. I’m currently finalizing a deep-dive into the raw JSON prompting syntax and Standardized Technical Rhetoric that will form the backbone of the upcoming Revision 5 rebuild.

What’s being refined in the "Sandbox" right now:

Architectural Standardizing: Moving beyond basic strings into formal JSON wrapping. I’m laying out exactly how the model's internal "weighting" handles bracketing and sub-prompting—no more guessing.

Raw Logic Control: I’m leveraging the model’s native technical language to build templates that are "future-proof." We’re talking about the high-level syntax you need for AdLib prompting and complex template design.

Formalized Nuance: Explaining the function behind the structure—why certain brackets trigger specific rendering behaviors and how to manipulate them for perfect "Material Interdependence."

📝 Note: Details on the final Rev.5 feature-set are being kept close to the chest for now. Just know that we’re scrubbing the "noise" and focusing on the raw technical knowledge required to master these models.

Conclusion: Expect a shift from "following the prompt" to "driving the engine." Rev.5 is about providing the community with the leading technical info needed to build high-level templates. We’re moving the goalposts.

  • source
  • parent