About DreamForge
A self-funded AI lab building
real-time world engines.
DreamForge is exploring a new foundation for interactive virtual environments: worlds generated directly by AI models, controlled as they unfold, and shaped without traditional asset pipelines.
What DreamForge is building
DreamForge is building a real-time AI world engine for controllable, generated virtual environments. Instead of assembling prebuilt assets, the system generates the world itself frame by frame and conditions future output on prompts, visual input, and user actions.
The current prototype delivers multimodal world initialization, live reprompting, first- and third-person real-time control, and long-horizon generation — an early research preview with a working interactive loop, not a finished consumer product.
Why generated worlds
Digital worlds are still mostly built through manual asset creation, scripting, level design, and engine-specific pipelines. That process gives creators control, but it also makes world creation expensive, slow, and inaccessible to many people who can imagine interactive spaces but cannot build them directly.
DreamForge is based on a different premise: if video generation can become controllable, persistent, and responsive, then generated environments can become a new substrate for creative software, simulation, prototyping, and digital world-building.
Current stage
DreamForge-World 0.1 validates the core thesis: generated worlds can run as real-time, interactive systems — not passive video output. The model responds to control, accepts multimodal input, and supports reprompting mid-stream.
Active frontiers include spatial memory, tighter control precision, and richer world consistency — but the real-time controllable loop is already in place.
The current version has been built under strict self-funded constraints, with total training spend around $2K so far. It runs in real time on consumer GPU hardware — roughly 10–12 FPS on an RTX 4090 in fp8, with a quantized build from about 4 GB VRAM, including RTX 2060-class laptops.
Direction
The near-term direction is to make generated environments more controllable, more coherent, and more useful for creative prototyping. The long-term direction is a platform where generated worlds can be created, edited, shared, experienced, and eventually monetized.
DreamForge is deliberately not a conventional game engine, asset generator, or metaverse product. Our differentiator is efficient real-time runtime on consumer hardware, multimodal control with first- and third-person view and mid-stream reprompting, and the path toward a creator-facing platform layer.
Conversations
DreamForge is open to selective conversations with technical advisors, early believers, compute partners, operators, and investors who understand deep technical risk in frontier AI systems.