FIGMA Make - Poor quality

Hi ​@Peter Bao - I saw you reply in an already existing post here, and have split it into it’s own topic individually. This was done so I could have a direct feedback conversation with you.

First: we know Figma Make is not perfect, but we are constantly working on improving it’s functionalities. This work does take time, but feedback helps us understand what parts of Figma Make we should be focusing on. 

It sounds like you have very powerful feelings on how Figma Make is today. I’d like to know more of your specific pain points. From what I gather right now, these are your main concerns:

• Quality of results from Figma Make has decreased
• The prompts you’ve created will change randomly as you continue to work on your design

You mentioned ‘clumsy’ -- can you give me more details on what that means for you? Is it that the results are not as precise as they used to be? Could you give me an example of what you request, how you expect it to react, and then the actual result that you receive?

When the ‘random changes’ happen -- do they just seem to happen suddenly? Or do you notice a certain action you take causes them to change? I’d love to get more specifics from you.

I appreciate the passion you have for us, our features, and your work. In that spirit, I’d kindly ask you to review our forum guidelines as well, so we can make sure everyone here can contribute conversations in a safe and respectful place.

Thank you - hope to hear from you soon!!

Hi ​@ksn -- When using prompts in figma make to change the content of certain pages, the AI will often modify other pages on its own. Even if you have told the AI not to change the requirements of other pages, it will still make changes. When MAKE was first released, this situation rarely happened! It has happened more frequently recently! I suspect that you are now using a large language model developed by yourself, rather than the Claude model that was used when MAKE first started. The Claude model is so smart that it can accomplish many things immediately without too many prompts. The model used by MAKE now is basically no different from an idiot. Often, MAKE will misunderstand the prompts of a sentence and go in the wrong direction! It often takes several rounds of back and forth, which is very bad. I think I will observe it again. If it cannot be improved, I will start using other AI tools to assist. Otherwise, my work cannot be completed through MAKE at all. I hope you can restore the Claude model used at the beginning in the short term. Your large language model is simply not good enough.

Thank you for the details -- I did get your feedback (word-for-word) over to the group of engineers and data scientists that are working on Figma Make. I can confirm that there has been discussion regarding some of the points you’ve emphasized, and it’s helping develop a better model.

I can’t speak on when/if/how updates will be implemented at the moment. However, I can provide a couple of recommendations:

• Bookmark out release notes page here, to learn about any new product implementations or updates we push out.
• Keep tabs on our blog here, where we post articles/commentary on subjects that relate to Figma. This is is a combo of design subject articles you may find interest in, as well as news updates.

Once again, thank you for willingness to write out your feedback. 

This is my third time commenting on these forums today. I’m so disappointed in it. It’s gotten worse and worse the more I work with it. Like so many others have reported, it seemingly gets dumber with every prompt. 

As of right now, it’s totally unusable. I am giving up on it.