Imagination-Studio-by-keys-Wasserman: video generator — what it does and what you need to run it

Imagination-Studio-by-keys-Wasserman is an open-source video generator that the authors describe as an "AI film studio" for Mac on Apple Silicon and NVIDIA DGX Spark. The tool is built on TensorFold and is designed around a combination of scene planning, image generation, and video creation with sound. The fact sheet separately lists the Fish Audio Drama 3 text-to-speech model, which handles voiceover.

What it does

As described by the author, the studio lets you plan a film like a film crew — via the Slate workflow by Sam Wasserman. Within this process, images are generated with the Qwen-Image-2.1 and FLUX3-K models, and then video is created using MiniMax H3 or FastH3. The video can include sound and voices from Fish Audio Drama 3.

Running is claimed on TensorFold for two platforms: Mac on Apple Silicon (Metal) and NVIDIA DGX Spark. The description mentions LoRA and MLX, which points to working with model adapters and using the Apple machine learning stack.

The Drama 3 model handles text-to-speech. The weights take up 38.6 GB, and the largest file is 3.5 GB.

What you need to run it

The code is written in Python and released as open source under the Apache-2.0 license. Repository: github.com/drowzeys/Imagination-Studio-by-keys-Wasserman. The last code change was on October 30, 2026, and the latest release is v4.3.1.

The Drama 3 model weights are distributed under the Fish Audio Community License. It permits research, non-commercial use, and commercial use with annual revenue under 500,000 USD. Hosted speech generation services require a separate license. Imitating voices without consent, removing watermarks, and training other generative speech models are prohibited. When distributing, you must include the license and provide attribution.

As described by the author, running requires 8.3 GB of free VRAM — the GPU driver takes memory from the free pool. For NVIDIA DGX Spark, 128 GB of unified memory, DGX OS with CUDA 13, and Docker are listed; according to the author, every feature works on this configuration.

Who it suits

The tool is aimed at those who assemble video from generated images and want to get sound and speech right away. The combination of Slate planning with frame and video generation will suit small production teams and independent creators who work on a Mac with Apple Silicon or have access to a DGX Spark.

For users without suitable hardware, running will be limited: the memory requirements are substantial, and support is claimed only for two platforms. Commercial services with revenue above 500,000 USD per year or speech generation hosts will not be covered by the weights license without a separate agreement.

Imagination-Studio-by-keys-Wasserman fills the niche of local video generation with sound on Apple Silicon and DGX Spark. The open source code and specific licensing framework make it clear for independent creators and small studios that are prepared to work within the stated platforms.