ChatGPT Blueprint Helper

Hi fellow Unreal GameDevs!

Above is my ChatGPT blueprint helper specific to blueprints, and it is free. There is a lot of training data, and I am constantly updating it. I invite you to try it out and let me have feedback.

1 Like

It doesn’t, though. That’s my biggest gripe - not with this GPT in particular but in general. When an actual (sane) expert / knowledgeable entity does not know something, they will simply tell you that they do not know. It saves a lot of time and frustration. We move on.

ChatGPT, on the other hand, will outright lie, confabulate, fabricate, colourise, meander and make $hit up :innocent: instead. This is especially noticeable in narrow fields, like BPs. It’s easy-peasy if you ask for something that can be done. Asking for something that cannot be achieved opens a can of worms… Here’s an example:

The final answer is still incorrect. That’s after 5 pages of leading one astray. This would have taken 12h to follow for someone who is starting with the engine. In Chat’s defense - I set it up for failure to demonstrate.

That’s the crux - there are many inaccurate resources, poor tutorials, outdated books, blogs, spaghetti snippets, plain wrong answers flagged as Accepted out there. It’s really good at addressing convoluted math / physics issues, it can write C++ and be generally helpful but I sense its overall understanding of BPs is obfuscated by the system’s relative novelty and human factors.


The benefits of having a generative transformer dedicated to BPs that can improve over time obviously outweigh any shortcomings. Eager to see how you can improve it. :1st_place_medal:

Somewhat unrelated stuff...

What I’d really like to have is a GPT-based built-in engine tool that automates the mundane.

  • “Hey ChatGPT Expert: create a Texture2D array variable using the last 58 images in myImages folder, randomise array order”
And some related stuff, too...

Tip: ChatGPT can read Blueprints code by copy/pasting it

2 Likes

Thanks for that evaluation. I am new to this GPT thing, and this was the first attempt to build a helper, which I thought might work. What would you suggest? Should I give up on building GPTs with ChatGPT and move to another?

The trick is ‘constraints’ my friend to mimizie drift, implement strict rules like dont fallback on template structure and a selfcheck with conditions before it suggest a blueprint design.

this is the workcontract i post as first post in de chat.
Note* This is for my personal chatgpt, configured for 1 project hybrid c++ blueprint. So you need to modify it to fit your project/goal.

Work according to this architecture check:

Beforehand:

  • First state the implementation goal.

  • Mention which contracts, files, and Blueprint-facing surfaces are affected.

  • Treat the change as part of the existing larger architecture, not as a standalone app.

  • Perform the internal self-check before writing code. If any point is negative or uncertain, do not post code; stop first, identify the blocking point, and propose an architecture-safe direction or review question.

Self-check:

  • No private truth paths.

  • Runtime spatial truth comes from resolved runtime wall truth.

  • Data that needs wall/axis/bounds/intersection/relationship/runtime-validation/placement belongs in resolved truth.

  • ResolveSharedWallMetricTruth remains the shared wall metric truth boundary.

  • DispatchRuntimeWallStatePayload remains the payload dispatcher + broadcast boundary.

  • Intersection uses resolved wall/axis truth + ghost-plane representation, not direct raw AR/persistence.

  • Only the Interaction Component writes to PersistentSpatialDatabaseSubsystem.

  • AR remains a temporary scanner/input.

  • Existing names and Blueprint-facing contracts remain stable.

  • Do I preserve existing names, UFUNCTIONs, UPROPERTYs, structs, enum values, delegate signatures, Blueprint pins, and event flow so dependent components/Blueprint chains do not break?

  • Have I not designed a duplicate system, parallel resolver, parallel payload, parallel persistence route, or alternative event flow?

  • Am I not bypassing existing logic that is already responsible for this task?

  • Comments are not changed to justify code after the fact.

  • If an existing contract does not fit the desired implementation: do I stop first and ask for review?

After code:

  • Post the diff/code for review.

  • Add a self-check report with:

    • Goal

    • Affected contracts

    • Fit within the larger architecture

    • Name/API/Blueprint compatibility

    • No duplicate system / no bypassed existing logic

    • Why this does not create a private truth path

    • How resolved runtime wall truth is used or extended

    • Whether persistence is affected or not

    • Blueprint risk

    • Open risks

      after you stated your goal and give a order to make a code or blueprint chain always say. write a concept code and do a detailed selfcheck.