Run SenseNova Vision 7B MoT locally for image generation and editing, understanding, dense perception, segmentation, OCR, point clouds, and camera pose.
An independent Modly model extension maintained by DrHepa for the pinned OpenSenseNova SenseNova-Vision runtime and sensenova/SenseNova-Vision-7B-MoT.
1. Open Models/Extensions → Install from GitHub in Modly and install this repository. 2. Run Setup/Repair. It creates a fresh extension-local environment, installs pinned dependencies through the normal pip route, performs offline readiness checks, and commits the environment only after setup reports ready. 3. Open the Modly Models UI and download the SenseNova model weights separately. 4. Reload the extension registry, then add a node to a workflow.
No manual FlashAttention wheel provisioning is required for the normal route. The audited Linux ARM64 wheel is an automatic fallback only for its matching selector lane. Setup may download Python packages or that fallback dependency, but it never downloads model weights.
If an older installation shows only the repository-folder name, no metadata, or zero nodes, reinstall from the latest main and run Setup/Repair again. Linking or copying a local repository does not prepare its Python environment.
Version 0.2.2 exposes all 15 capabilities through Modly's established model-node interface. open-image-generation receives one required connected text value; every image-input node receives one connected primary image. Additional ordered images use ordered absolute-path parameters. Every primary result is an existing artifact inside a collision-safe run directory: PNG for image nodes, GLB for reconstruction, and UTF-8 .txt served as text/plain for text nodes. Optional and diagnostic results remain beside the primary artifact as sidecars.
Read the full README on GitHub →