Welcome @vinceja,
it´s not that easy to say do this and this and it works. It highly depends on several factors, what the output is:
-
Prompting (How good is your requirement, etc.)
=> try using experimental prompt in Settings
=> Use “Enhance Prompt” Button before sending
=> See the FAQ: Frequently Asked Questions (FAQ) - bolt.diy Docs -
Model
=> Depending on the Model you get other output. Also for Claude 3.5 Sonnet is an old and a new Model available. I´ve read some user topics, where is mentioned that the new Model not performing as good with bolt, as the old one
=> The System Prompt not working perfect/good with each model. In the future bolt.diy will provide multiple prompts / prompt library to cover this and maybe also allow to add own system instructions in the settings -
Starter Template
=> When starting with a good base, you get way better results, then starting from scratch. Consider to use the Starter Template Buttons at the startpage and you can also directly clone git-projects you find online with appending the gitrepo to the bolt.diy url:
http://localhost:5173/git?url=https://github.com/YZarytskyi/react-voice-visualizer.git
-
Images
=> As today images not working within bolt, expect you are using SVGs. (@aliasfox / @thecodacus correct me please if this is maybe already working or working soon)
Instead of burning much credits, I personally would recommend just switching to Google Gemini 2.0 Flash, as it is free to use and maybe switch to another Model, if you come to a point where the current model cannot fix it, despite you asking again and again.
Also what I do then is, go to ChatGPT o1 (I´ve got PLUS abo) and ask it for specific solutions to the error => paste the provided solution to bolt and often it can fix it then based on the provided solution.
Hope this helps a bit. Let me know if I missed something