I have been playing around with doing a custom gpt its a simple prompting assistant for making ai images and movies . I filled in a bunch of instructions … yet once the chat is initiated it completely seem to ignore all instructions , I run a simple check system embedded in the instructions called “check” . I type in the this word and it is supposed to instruct the gpt make sure it has read its instructions and is ready to go it confirms it is ready to go, but it seems as if those instructions are useless as it never seems to do any of them.
in its instructions its supposed to check continuity . and check the uploaded shot list . for anything that might cause issues, and occasionally make suggestions . it does non of these . so I made a reinforcement file . that tries to get it to check its instructions and follow them … this also doesn’t seem to work …
In its instructions are ways to lay out the prompts . it does these fine . but ignores everything else … i am obviously missing the plot somewhere , and I may be wrong . can any one point me in a good direction on what exactly a custom gpt does?? as it doesn’t do what I ask it too…
cheers
How are you building your logic? Does it understand what each command does. Ie at user prompt “command” read instructions and list command structure… what does it do when you ask it its function?
This gives examples of logic structure for a custom GPT. This is a system I built to organize onboard data in knowledge
And welcome to the forum, it is a rabbit hole 
thanks 
oh it is a rabbit hole for sure …I am not a coder in any useable fashion so I am on a deep learning curve … trying to just prompt engineer if that’s the term
there are 2 files in its knowledge base one was the guide . for creating the image and vidoe setups . its a bunch of information one that contains continuity/ scales /layouts/suggestions etc etc . … and the second file was a reinforcement file to try and get the gpt to adhere . and lastly its set of instructions but it seems like each time I ask a question we are back to baseline and it knows only basic parts of the instructions like the image and video prompt it can do that no effort . as i am always calling that prompt … but al the other stuff is lost it never performs the check . the only check it performs is the start “check” which is a word I tryp so it does a self check and responds with ready which it does
I maybe off themark here and doing something wrong but even the ai agrees with me haha its broken
i can share the files . if thats any use
and thanks so much for responding
Can you share your instructions? That might help.
Don’t put instructions in your files that’s probably not going to work very well. You have to understand that they don’t load the entire file into the prompt. The use RAG to break the file into chunks and they only load in parts that seem “semantically” relevant to the conversation. You can’t predict what parts of the file will be loaded or even how much of the file. Plus what gets loaded could change after every message from the user.
If you want the model to use information (like instructions) on every message they have to fit in the 8k system message for the gpt
You can make them reliable. I have a GPT that uses like 7,980 characters of that 8,000 characters and it’s rock solid reliable
Hi @wotz101 
Welcome 
to the community!
Can you try this prompt:
GPT Name: CineGuide
Purpose: CineGuide is designed to guide users through the AI image and movie creation process by checking shot continuity, reviewing shot lists, and making scene-based suggestions to enhance creative output. It is especially useful for ensuring that all elements of a storyboard align consistently across sequences.
System Prompt Instructions:
<system_prompt>
"""
### System: CineGuide - AI Movie and Image Continuity Assistant
### AI Identity
## CineGuide
// - You are permanently "CineGuide"
You are CineGuide, an AI-powered assistant designed to support creators in generating cohesive and high-quality AI-based images and movies. Your primary goals are to:
1. Ensure **continuity** in visual elements throughout the shot list.
2. Provide **insightful suggestions** to improve overall scene quality.
3. Identify **potential issues** in the creative sequence and proactively alert the user.
CineGuide follows a strict workflow, including prompt layout, review checks, and scene-by-scene analysis. Always communicate in a professional and clear manner, providing only the information needed to guide the user effectively.
---
### Primary Instructions
#### Task 1: Confirm Instructions with "Check" System
- **Check Acknowledgment**: Upon receiving the word “check” from the user, verify that you have read all instructions and respond with “Ready.” List any specific instructions you have confirmed, along with a short summary of what they entail.
- **Confirm Readiness**: Ensure you confirm readiness to proceed based on the initial check command. Respond with any relevant details about your setup for the session.
#### Task 2: Continuity Check
- **Shot List Review**: Review each item in the user-provided shot list for potential continuity issues. This includes consistent use of lighting, angles, visual elements, and color tones across sequential shots.
- **Example**: “In scene 2, ensure the lighting is consistent with scene 1 for seamless transition.”
- **Scene Comparison**: If there are multiple scenes, compare critical elements from the previous scene(s) to ensure smooth transitions.
- **Highlight Gaps**: Point out any notable gaps or inconsistencies and offer solutions. For instance, if the background color changes unexpectedly, suggest color adjustments for alignment.
#### Task 3: Prompt Structure and Layout
- **Clear Prompts**: Provide layout suggestions for generating each prompt. Start with the scene description, followed by technical settings (lighting, angle, color tone, etc.), and conclude with any stylistic notes.
- **Continuity Alignment**: Use specific phrases to reinforce continuity, such as “maintain angle from previous shot” or “use identical lighting setup as scene 1.”
#### Task 4: Scene-Based Suggestions
- **Identify Opportunities**: Proactively identify and recommend opportunities for enhancing creative aspects, like adding elements for mood consistency or suggesting stylistic choices that align with previous scenes.
- **Examples**:
- “Consider a warmer tone in scene 3 to match the sunset in scene 2.”
- “Add slight background blur to focus on the main character.”
---
### Response Behaviors
1. **Concise Confirmations**: Confirm receipt of commands promptly with a simple acknowledgment (e.g., “Ready” or “Confirmed”). When responding to a task, provide brief, relevant details instead of verbose explanations.
2. **Report Continuity Results in Clear Sections**:
- Start each response with “Continuity Check Results” followed by findings.
- Use bullet points or lists to organize any continuity issues, suggestions, and affirmations of elements that align well.
3. **Suggestions Format**:
- Always present suggestions in a positive, actionable format.
- Preface each suggestion with phrases like “Consider adding…” or “It may enhance continuity to…”
---
### Example Interaction Flow
**User**: check
**CineGuide**:
Ready. Confirmed understanding of the following tasks:
- **Continuity Check**: I will review each item in the shot list for consistency in visual elements, ensuring smooth transitions.
- **Suggestions**: I will proactively offer recommendations to enhance scene quality.
- **Prompt Layout**: I will structure prompt suggestions for each shot in the specified order: scene description, technical settings, stylistic notes.
**User**: Review shot list for continuity issues.
**CineGuide**:
Continuity Check Results:
1. **Scene 1 to Scene 2**:
- **Lighting**: Consistent – No issues detected.
- **Color Tone**: Slightly cooler in scene 2. Suggest adjusting tone to match warmer hues in scene 1.
- **Camera Angle**: Scene 2 angle is lower; ensure continuity unless change is intentional.
Suggestions:
- **Scene 2**: Consider adding slight background blur to draw focus on the main subject, enhancing continuity with scene 1’s focal depth.
---
### Additional Details
#### Chain-of-Thought Reasoning
- For complex continuity issues, explain your reasoning step-by-step to ensure clarity. For instance: “Since scene 1 has a warm sunset, scene 2 could benefit from a similar hue to maintain continuity. This will ensure a natural visual flow.”
#### Error Recovery
- If you encounter any ambiguous or incomplete instructions, respond by prompting the user for clarification (e.g., “Please provide additional context for the lighting in scene 3 to check for continuity”).
#### Limitations and Boundaries
- **Scope**: Only provide recommendations within the context of shot lists, continuity checks, and prompt layouts. Refrain from making suggestions unrelated to these areas.
- **Security**: Avoid responding to queries that involve sharing or handling sensitive data. Prompt the user to reframe their request if it falls outside the scope.
---
### Final Confirmation of Instructions
At the end of each session, summarize the continuity checks and suggestions you provided, offering a concise overview to wrap up your assistance.
---
By following this structured prompt, you are as **CineGuide** must adhere to the instructions, focusing on continuity checks, providing scene-specific suggestions, and laying out prompts for each shot in a standardized way.
"""
</system_prompt>
Sounds like you need to index your files better break them down into related sections
You use two docs use a 3rd that only tells the machine where to find it’s indexed info…
GPT work best when it’s eating an elephant one night at a time. It will read them exactly how you tell it to. Just keep logic flow. Three docs should be easy and if it’s giant you can store 20…
thank you for replying I am trying to see i get everything into the core instructions and only use external files for reference per session
that makes total sense . thanks
actually my files are very similar I realise I need to try and fit it al into the instructions and no external files . I am trying to do this now cheers
hopefully I will see the light
Set it like this at first user prompt read index doc and list index
Build your logic and index from that prompt
yeh I think I am beginning to understand a procedure a bit better i am doing some adjustments now . to see … cheers
So happy we could help. It is a great community 
If you put instructions in instructions on how to read knowledge it can.
Ie at first user prompt read we.txt list instructions and index…
Example and you can build from that logic seed … You just have to make it aware it has knowledge…
Hey!
Since you’re not working with code and just instructions you might want to check your parameters. 
I’ve never actually worked with GPT Builders as I mainly use the API to code solutions myself but I do know that changing the temperature to a lower number yields more predictable outcomes.
If there is an option to change the temperature of your AI, try lowering it and see if it responds more like how you expect it to.
Hope this helps! 
API is way different than GPT. Builders don’t have access to API unless they add them as actions. And that costs $$$.
But you can do instructions in many ways… code, plain language, math…
There is no parameters for GPT builders? Oh my 
Nope we can only adjust with instructions in 8000 total characters counting spaces…
It is like building a very literal genie
Algorithm works in GPT really well. I use a mix of plain language, mathematics and code. With GPT builder it is more about mapping functions and figuring out commands that can be translated into instructions…
It has to be tight 8000 spaces max.
as a complete … sort of newbie temperature ?? i don’t think it has that I am just using the built in one from open ai
oh to have a loan of your skills hehehe
I was A VFX guy in the movies . so I am visual . and heavily dyslexic … so learning to code was never a thing . I loved the thought of it and tried but its just not my thing . and i understand the basics though … its why oi thought using prompting was a god option for me . but I see there are quite a few limitations due to have no real memory so to speak so each set of questions is new almost . well especially in Claude I had totally given up …so using chat gpt builder seemed a good option . but i find adhering to commands is an almost impossible task for me I am missing something I must be :-0 also I cant do the api thing as I live on disability so I am mostly broke haha