Features
- A Gemini-style assistant on your PC — chat with any local text model (pick it from the menu at the top), make pictures and videos from the same box, and find every conversation again in the sidebar.
- Videos — text or picture to video with Wan 2.2 (built in, one download) or your own ComfyUI workflow. Music and sound effects are next.
- An AI writer that builds the whole game — it plans the story, writes every scene and choice, defines items and variables, checks for broken links and dead ends, and fixes them.
- Runs on your PC — text models (GGUF) through the built-in llama.cpp engine with automatic GPU detection (Vulkan, CUDA or CPU), or your Ollama / LM Studio server. No account, no cloud.
- Illustrations and animations — scene pictures from stable-diffusion.cpp or ComfyUI, rendered in one batch after writing with the text model unloaded, plus living effects: rain, snow, fog, embers, slow pans.
- Image studio — make pictures without a game: prompts adapted to your model (Danbooru tags for Illustrious/Pony, or written for you by the AI), LoRAs with strength sliders, a detail pass for sharp faces, batches and a gallery. ComfyUI workflows are built automatically.
- Everything built in — the image engine ships with the app (NVIDIA, AMD, Intel or CPU): install Z-Image Turbo, FLUX.3 / FLUX.2 klein, Chroma or Qwen-Image in one click, with no ComfyUI or other program.
- Picture editing & consistent characters — edit any picture with words, or put a game character into a new scene with the same face and clothes (FLUX.2 / FLUX.3 klein, Qwen-Image-Edit); game scenes use the characters’ portraits automatically.
- Models & LoRAs installer — the studio finds the LoRAs your prompt needs (characters, styles, concepts) and installs them automatically; or search Civitai for LoRAs and models made for your model type, install them in one click into ComfyUI or your models folder, plus upscalers and the stable-diffusion.cpp engine itself.
- Your models folder — pick the folder once: text models, image models and the stable-diffusion.cpp executable are found automatically, with a hint on whether each model fits your GPU.
- Full editor — edit any scene, choice, condition or effect by hand, play from any scene, regenerate pictures.
- Share your games — export a single HTML file that plays in any browser, with saved progress.
Getting started
- Put a text model in a folder, e.g. Qwen3-8B-Q4_K_M.gguf (from Hugging Face). For pictures, open Image studio → Models & LoRAs and install a ready-to-use model in one click (or use your own SD/SDXL models, or ComfyUI).
- Open Settings, choose that folder and pick the models.
- Create a game, describe it in the Create tab and watch the agent build it. Then play, edit and export.