Azaran on 10/10/2024 at 02:51

---

downwinder's comment on another thread got me thinking if it's possible to use a current AI to control Dromed and create AI fan missions, and looks like it's challlenging, but possible according to Chat GPT, using API or SDK's. Yes I know this is controversial, but we already talked about AI generated voices, etc. might as well bring this up too

1. Integration with Dromed
API or Scripting Integration: Dromed would need an accessible API or scripting interface for the AI to interact with. If Dromed lacks this capability, a custom integration layer (like a wrapper or a mod) would need to be developed to interpret AI instructions and translate them into Dromed commands.
Automation Tools: Alternatively, automation tools like AutoHotkey could simulate keyboard and mouse inputs to control Dromed, although this approach is less efficient and harder to make reliable.

2. Natural Language Processing (NLP)
Interpreting Prompts: The AI would use NLP to understand the user's prompt and translate it into specific actions. For example, if a user prompts, "Create a Hammerite cathedral with hidden passages," the AI would break down the request into components (building the structure, adding specific textures, creating passages) and sequence the appropriate Dromed commands or scripts to execute them.
Level Design Rules: The AI would need a set of rules and heuristics about the Thief games’ aesthetics, level structure, and gameplay mechanics to generate levels that feel authentic and functional. It might draw from pre-existing designs and patterns found in Thief levels to inform its decisions.

3. Procedural Generation and AI Model Training
Procedural Generation Algorithms: Implementing procedural generation techniques could help the AI create environments quickly, which can then be fine-tuned based on the user's specifications. These algorithms could generate basic layouts, architecture, and object placements.
Training an AI Model: If enough Dromed levels or assets are available, an AI model could be trained on these examples to understand how different types of levels are constructed, improving its ability to create similar or varied designs.

4. 3D and Object Placement Capabilities
3D Layout Planning: The AI would need to understand 3D space, not just at a coding level but also in terms of game design principles (e.g., ensuring accessibility for player movement, proper lighting, and logical flow).
Placement of Objects and NPCs: It would also need to know where and how to place objects like loot, enemies, and environmental hazards, based on gameplay principles of the Thief series.

5. User Interface and Feedback Loop
Interactive Interface: An interface allowing the user to guide or refine the process, either by updating the prompt or directly interacting with the AI’s current work in Dromed, would be useful. This feedback loop would ensure that the AI’s creations match the user’s vision.
Real-Time Adjustment: The AI could run in tandem with the editor, showing real-time updates as it builds the level, and the user could stop or adjust the process as needed.
Feasibility and Challenges

Technical Complexity: Building this kind of integration would be complex, especially if Dromed does not have a modern scripting API. The use of automation tools could be limited in precision and flexibility.

Development Time: Training an AI model to understand and design levels would require a significant dataset of Thief levels and game design knowledge.
Creativity vs. Authenticity: Balancing the AI’s creativity while maintaining the authentic feel of Thief’s level design would be a challenge, requiring a deep understanding of the game’s mechanics and atmosphere.
While it’s a challenging project, with the right technical foundation and enough training data, an AI could be designed to assist in or even autonomously create levels using prompts. This could open up new possibilities for fan-created content in the Thief community.

<Username> on 10/10/2024 at 18:18

---

The Varyx Obelisk (TG, 2001) is (https://www.ttlg.com/forums/showthread.php?t=29474&p=351416&viewfull=1#post351416) over 90% CowGEN terrain. We already had cases in which scripts were used to generate missions more than 20 years ago. Using generative AI to create the scripts for a CowGEN 2.0 does sound technically feasible.

Kamlorn on 12/10/2024 at 21:30

---

Can you elaborate, please? Decided to check this by actually playing this FM through; those random dirt\gravel inclusions around paving stones, generic street arcs\tunnel spans?

gamophyte on 12/10/2024 at 22:25

---

I think AI can be applied to many areas that could benefit humanity. However story and narrative driven pieces of art, such as FMs, are not one of them.

There is something that happened to me while playing TBP, when I saw a little hut hanging from the roof of a sewer. I feel like that was a human touch, in a way I can't describe. But I felt the emotion of someone who might have to stay there, forgotten by the inhabitants of the city. I could have felt the author in some way. It is something I believe that can only well up in the imagination of a human, with human experiences. AI can't describe what is forgotten, and I believe it cannot recreate a moment I will never forget playing TBP.

Anyone wouldn't be mistaken for coming to realize the vapid nature of AI in its current state, and when you couple that with its threat against industry pioneers (who I hope we admire), it can be quite stomach turning. You can prompt AI to try to create what I said in the paragraph above, and it still won't "connect".

Ever since there was DromEd, and the difficulty it presents, people have been looking for shortcuts, and time savers, I get that. Hell maybe AI can auto room brush for us LOL. But not build a world of thought and emotion deep into our untracked unsaid nature. In my opinion.