Skip to main content
LizP904
Participating Frequently
July 4, 2025
Question

Generative AI producing awful results at the same time they are limiting usage via credit cap

  • July 4, 2025
  • 2 replies
  • 444 views

With the conversion to limited credits, I have been selective about my usage but each time  I have used it, the system has produced options with zero change to the image. Example, "remove shadow". All three options did nothing with the shadow. I reported it and deleted the layer. Prior to the change, I used it several times in many images and it was amazing. Something is wrong. I worry that the standard plans have been assigned to the less trained machines. Regardless, I think we should not be docked the required 3 credits on results that do nothing. There is a way to show the image and the results. Zero result options should be credited back to the account. A standard account gets 240 credits but it is really only 80 uses per month because since we are charged for 3 options per request. That's not much at all and hurts worse when results are sub par. 

2 replies

davescm
Community Expert
Community Expert
July 8, 2025

To add to the reply from  @creative explorer 
If you envisage heavier use of AI, several plans allow unlimited use of standard generations which include generative fill, including the CC Pro plan (North America only at the time of posting this reply) and the additional Firefly plan.
https://helpx.adobe.com/uk/firefly/get-set-up/learn-the-basics/generative-credits-faq.html
https://www.adobe.com/uk/products/firefly/plans.html

 

Dave

creative explorer
Community Expert
Community Expert
July 8, 2025

@LizP904 Adobe's system currently charges credits for each generation, regardless of the quality or whether it meets your expectations. This is because, from a technical standpoint, the AI model still performs a significant amount of computational work to generate those three options, even if the output is flawed. Each attempt, whether successful or not, consumes server resources, GPU time, and energy. While this is the technical reality, it clearly doesn't align with the user's experience of value when the results are useless. 

Instead of 'remove shadow' it's how you 'prompt' the system.  Sometimes adding more context can help. For example, "remove harsh shadow from foreground object" or "make shadow disappear behind subject." This isn't a solution for the core problem but might increase your success rate for now. 

 

m