Behind the scenes at TW Farms
TW Farms
AI Workshop
One creator. A virtual team. Fewer barriers.
I have plenty of ideas for TW Farms. Getting them from “that could be good” to something that actually works is the harder part.
As a disabled independent creator, I’m using ChatGPT, local AI and familiar creative tools to make that work more manageable—with my own direction, judgement and a limited budget.
See what exists todayA working experiment in making things. Still very much a human at the controls.
01 / The practical challenge
Plenty to build.
One person to build it.
Running TW Farms means moving between website development, video production, FS25 modding, mapping, testing, research and technical troubleshooting. There is creative work in there too—once the configuration has stopped arguing.
Disability can make that mix of detailed tasks and constant context-switching difficult to manage. A project can be worth doing and still be a lot to hold together on your own.
The aim is practical: break the work into manageable steps, get help when something gets stuck, and keep enough context to pick it up again. I still choose what matters and decide when it is ready.
02 / How the workshop works
Direction stays human.
“Virtual team” describes a way of organising assistance. The tools do not run TW Farms on their own.
Creator / TW Farms
I set the brief, priorities and boundaries: what we are making, why it matters and what must stay untouched.
ChatGPT
Helps turn the brief into steps, investigate problems, develop code and coordinate which specialist tool is useful for each job.
Specialist tools
Different jobs use different tools, with scoped instructions and checks.
Development
Code, website changes and debugging, reviewed in VS Code.
Local AI
LM Studio and Riley/Qwen for selected local tasks.
Visuals
ComfyUI for image and video experiments.
Voice
An isolated Chatterbox bench for authorised reference auditions.
Video
Editing, audio, captions and conventional production tools.
Mapping & modding
FS25 files, inspection tools and development builds.
Back to TW Farms: review & decide
I inspect the result, test it where needed, and decide whether to keep it, revise it or stop. That decision shapes the next brief.
TW Farms directs → tools assist → TW Farms reviews → the next brief.
03 / What exists today
Real work. Honest labels.
Completed means the output or milestone described here exists. It does not mean every part of the wider project is finished.
The TW Farms website
A working home for map tours, videos, community links and project updates. The site is live; improvements continue.
Public website
Explore TW FarmsFS25 map tours & videos
Published tours and gameplay give viewers a proper look around Farming Simulator maps. The map makers and contributors keep their credit.
Published output · Central Pennsylvania map tour
Watch the map tourRoute Forge & FS25 mods
Local mod builds and routing tools are taking shape. Route Forge still depends on AutoDrive and Courseplay; it is a development build, not a finished replacement.
Local source and packaged prototypes · not a verified public release
Map Modder
Tools for inspecting map files, analysing farmland and reviewing changes. Local reports and diagnostic images exist; testing and development continue.
Local code, inspection reports and map diagnostics
The local AI workshop
LM Studio and ComfyUI are installed alongside the development tools. Separate GPU workloads have been demonstrated; individual creative workflows still need review.
Working local setup · a completed setup milestone
Chatterbox voice bench
An isolated local speech test produced a technical baseline. An audition harness records settings and protects existing files. This is not an approved SADIE voice.
Baseline WAV and harness checks · technical milestone
SADIE
A digital presenter and narrator in development. Visual references and technical previews exist; the final voice and consistent long-form delivery are still to be established.
Presenter experiments · final narration not approved
These examples show the wider workshop, not a claim that AI made every output. Map makers, modders and other contributors remain part of the story. Meet the people behind the work.
04 / What AI makes more manageable
Less to hold
in my head at once.
The useful part is often a clear next step, not a spectacular demo.
- From a large problem to a small next step
- A website or modding task can be broken into an inspection, a proposed change and a test—rather than trying to solve the whole thing at once.
- Context I can return to
- Written plans, change notes and handoffs help me track what was tried, what worked and what still needs a decision. I can check them rather than relying on memory alone.
- A second look at a technical problem
- AI can help read an error, compare a configuration and suggest a diagnosis. The answer still has to survive an actual test.
- A workable route from idea to production
- Rough ideas can become shot lists, development stages or a voice-test brief. Specialist local jobs can be kept separate so one experiment does not disrupt another.
05 / Local + cloud
The right tool for the job.
I’m investing my own resources in a workshop that can keep developing, using local hardware where it makes practical sense.
Cloud: thinking the work through
ChatGPT is the central planning and problem-solving layer: reasoning through options, helping with research and development, and coordinating the next steps for me to review.
Local: specialist workloads
LM Studio, ComfyUI and the voice bench handle particular local jobs. The RTX 5060 Ti and RTX 5070 Ti setup has allowed separate AI workloads to run on separate cards.
Conventional development, editing and testing tools still do essential work. Local and cloud AI are parts of that toolkit, not substitutes for checking the result.
06 / The experiment
How much can one independent creator accomplish when AI reduces some of the barriers between an idea and actually building it?
The website and published videos are real outputs. Other parts are development builds, technical tests or early experiments. SADIE’s final voice, for example, still needs an authorised reference and careful listening tests.
That is the experiment I want to keep documenting: what helps, what fails, and what becomes possible when a useful idea has a manageable path to being built.
07 / Continued access
Keep building.
Keep learning.
This proposal seeks product / access support, not financial sponsorship.
Advanced AI access is a meaningful ongoing expense on a limited personal budget. I’m already investing in local tools and hardware, but continued access to ChatGPT remains an important part of how this workshop comes together.
I would welcome a conversation with OpenAI about continued or subsidised access to ChatGPT Pro, or appropriate product and usage support, to help me keep developing and documenting the experiment.
TW Farms is not seeking financial sponsorship from OpenAI through this proposal. This is an invitation to discuss a practical creator workflow and whether there is a suitable way to support its continuation.
OpenAI does not currently sponsor or endorse TW Farms. No official partnership exists.