I faced this problem on and not off but couldn’t really find any patterns. Sometimes, figma make is hallucinating with my prompts and giving weird screens (replacing existing screens) and giving wild outputs. After some time or the next day, the problem is getting fixed. For sure, the problem is not with the prompt, the problem starts with the way it is reasoning it. If this can be fixed, it will be really good because the flow is getting cut quite often.
Question
Prompts are hallucinating sometimes
This is an issue because they switched from Claude Sonnet 4 model to 3.5 …. not even 3.7. They got out of beta and will be limiting prompts as well … so basically limited prompts with sonnet 3.5 is like just waste of time .. you wont be able to finish anything with quality. I’m very very disapointed of this decision, and how they upped the model from 3.7 to 4 and then switched to 3.5 in just 2 weeks … very very bad.
Ohh, that’s bad. Where to find which model is being used?
It was written here at the bottom… now its not, so you just ask the model and it will answer you … but its very noticable and obvious even if you don’t … It just breaks things.
This is what i got as answer ln the morning when i noticed something is not right with the model after the update.
Can you please give an explanation regarding this change. If this is the backend scene, then the ongoing projects are really getting a big hit. Understanding that we’re in beta testing, but the model couldn’t make anything useful
Still no explanation.
Also I think that they may even instructed the agent not to answer you if you ask it directly what model it is… you have to work around it to find out.
Maybe I’m just imagining things, but I tried couple of times, and it never gives me an answer, it just starts reasoning nad stops without providing the answer.
But with a bith of prompt manipulation you can still find out …
In the end they can do whatever they want, but this was not announced nor explained anywhere so we are aware and don’t start breaking our projects.
This is bad. My ongoing project is stopped because of this. I can no longer give prompts for me to get a valid answer/reply from the model. Hope this is getting resolved soon
Please check your email when you have a chance -- I directed it to the email you have associated with your Figma Community account. If there’s another email it should be directed to, let me know in a PM.
Came across this thread via Google. I’m not using Make for proper projects yet (everything I work on is too complex) but have been trying to prototype with it, and it just straight up lies. What is saying it has done is what I want, but the result on screen is not aligned at all, and is basically unusable. At the moment this thing is trash.
I ran into hallucinations as well working on trying to build a 4 screen order flow POC in Make.
It sees once you get past 40 prompts or so, the likelihood increases.
Side note: the 80-100 estimated prompt limit to be put in place for Enterprise users is going to be woefully low as I broke 120 just with one version of the aforementioned file, and I need >1 for workflow comparison testing.
Only being able to make 1 POC workflow a month would be extremely problematic.
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.