Skip to main content
Known Participant
June 30, 2026
Question

Don't charge generative credits for failed or regenerated results.

  • June 30, 2026
  • 5 replies
  • 0 views

Photoshop's generative features are billed per generation. When a result is usable, it saves real time; however, they aren't consistent. I can use the same engine, same prompt, and same selection, and the result varies wildly run to run. When it misses, I regenerate. The second attempt costs another credit. So does the third.
 

Potential solutions:

1. Don't charge for regenerations of the same prompt and selection in a session. I paid for the intent. Let me iterate to a usable result.
2. Auto-credit when the system returns nothing usable.
3. Deduct the credit only when I accept a result into my file, not on every attempt.
4. Give pros a way to lock and reproduce a generation that worked, plus an honest published success rate.

    5 replies

    Trevor.Dennis
    Community Expert
    Community Expert
    June 30, 2026

    MY advice is to use long prompts with repetition. Think of various ways to describe what you want to happen. Also say what you don’t want to happen.  Get to know the models.  Firefly will get it wrong a lot more than Gemini Nano Banana.  Write your prompts outside of Photoshop so you can a)_see them better, and b) have the prompt available to edit for your next try.  

    The Ai models are literal.  i.e. they don’t have common sense.  They don’t recognise inflection, body language, sarcasm etc.  In fact they can be quite stupid.  Check out the FatherPhi YouTube channel to see what I mean

    I have attached a guide from Colin Smith’s Photoshop Cafe YouTube channel.

    Known Participant
    June 30, 2026

    Thanks, Trevor. This is good advice and reflects my process well. I give a clear prompt, and I tell it what not to do. I also favor Gemini over Firefly. At this point, I’ve had quite a bit of experience with the product and the engines  

    I was recently working on a project where I used a prompt successfully on one image, then when I tried to use it again on a different but ostensibly the same image (same engine, same selection), I could not get a usable result and had to move on.

     

     

    Stephen Marsh
    Community Expert
    Community Expert
    June 30, 2026

    This would seem to be a natural limitation of the diffusion model (slightly mitigated by using a reference image) and that it still has a compute cost, regardless of whether you find the result usable or not.

     

    https://computers.sciencearray.com/how-diffusion-models-generate-images-from-noise

     

    https://www.howtogeek.com/ai-image-generators-explained-how-they-work-and-what-they-can-do/

    Known Participant
    June 30, 2026

    They have a choice in how they charge for it. 

    Stephen Marsh
    Community Expert
    Community Expert
    June 30, 2026

    This is true.

     

    Profitable pricing is generally accepted as "cost plus". There are raw, base costs and a markup applied for margin/profit. For any business (not just Adobe), the debate is on what is a reasonable markup.

     

    So, Adobe could indeed decide to price "unsuccessful" generations at a lower credit that covers costs, and then price "successful" generations at a higher credit rate for profit.

     

    What would you suggest as the criteria for a "successful" generation? How would this be flagged for an "unsuccessful" generation? How would this be deployed (the mechanics)?

    Trevor.Dennis
    Community Expert
    Community Expert
    June 30, 2026
    D Fosse
    Community Expert
    Community Expert
    June 30, 2026

    AI is by its very nature a throw of the dice. You have absolutely no control over what it returns. The concept of “consistency” doesn’t apply here.

     

    If I may - it sounds like perhaps you should join our small, but growing, party of people who reject AI and stick to workflows we can actually control? :-) ;-) AI is just an option, it’s not mandatory.

     

    Known Participant
    June 30, 2026

    I am not asking for the output to be predictable; I know it's a throw of the dice. My point is that the billing isn't a throw of the dice. The model's inconsistency gets priced as my problem. Every miss and regeneration costs another credit. The roll being random is fine, but charging full price on every roll, win or lose, is what needs to be discussed.

    This is a feature request open for voting, not a referendum on whether AI should exist. AI is a tool, and I am happy to use it selectively. The proposal is one line: don't charge for failed and regenerated results. Please either engage with that or let people vote on it. Talking posters out of the request they came to make isn't what this board is for.

    D Fosse
    Community Expert
    Community Expert
    June 30, 2026

    Duplicate thread, a moderator should lock this.

     

     

    Known Participant
    June 30, 2026

    Can you link me to the dupe so I can weigh in?