Phosphene launches local video suite

Phosphene AI said on May 22 that it released Phosphene 3.0, a free video-creation suite for Apple Silicon Macs built around local inference rather than cloud rendering. The developer said the update supports text-to-video, audio-to-video and character workflows on-device, and promoted the release in a post on X with demo clips and performance claims centered on Apple hardware. The project’s public GitHub repository describes Phosphene as a local generative video panel for Apple Silicon with one-click installation through Pinokio. ### What exactly did Phosphene say was new in version 3.0? The May 22 X post said Phosphene 3.0 was “fully redesigned” and added image-to-video, characters, LoRA training, FFLF and Extend, along with text and audio to video. The same post said users could “Train unlimited characters now,” framing the release around features that run locally on Apple Silicon rather than through a hosted service. (x.com) GitHub materials for the project show Phosphene had previously been marketed as a free desktop panel for local video and audio generation on Apple Silicon. The repository summary says it wraps LTX-2 MLX and offers a one-click Pinokio install, while project documentation describes the software as Apple-Silicon-only. ### What is the software built on? The public project documentation says Phosphene runs Lightricks’ LTX 2.3 model natively on Apple’s MLX framework. (x.com) The same documentation says the panel itself is MIT-licensed, while model weights carry Lightricks’ own license terms. The GitHub repository also lists recent work on image-to-video pipelines, warm helpers and release flows, offering a technical paper trail behind the local-video claims. (github.com) Commit and file listings visible on the repository include `mlx_ltx_panel.py`, `mlx_warm_helper.py` and install scripts tied to Pinokio deployment. ### Why is Apple Silicon central to this launch? Phosphene’s public materials repeatedly describe the software as being for Apple Silicon, not general desktop GPUs. (github.com) The repository summary calls it a “Local generative video panel for Apple Silicon,” and the project documentation says it is “Apple-Silicon-only.” Startup Fortune, summarizing the earlier release, reported that Phosphene runs LTX locally on Apple Silicon Macs through MLX and positions that setup as a way to avoid cloud API fees. (github.com) That characterization matches the developer’s own emphasis on local processing in the May 22 launch post. ### How does Phosphene fit into the local-AI tool chain? Pinokio’s listing describes Phosphene as “local video and audio generation for Apple Silicon,” and says it is installable through the Pinokio app. The GitHub documentation likewise presents the software as a desktop panel wrapped around upstream model packages and local downloads, not a browser-based service. (startupfortune.com) The project documents also outline large local model files, including quantized model variants and a text encoder, indicating that users are expected to download and run substantial assets on their own machines. The repository notes directories for model weights, generated outputs and uploaded reference images. ### Where can users verify the release and follow what comes next? (beta.pinokio.co) The May 22 announcement was posted on Phosphene AI’s X account, where the developer shared demos tied to the 3.0 release. The codebase and install path are public through the `mrbizarro/phosphene` GitHub repository and Phosphene’s Pinokio page, which the project documentation identifies as the main public distribution points. (x.com) (github.com)