Skip to content

Feat/vistadream package#2

Open
pablo-rerun wants to merge 2 commits intomainfrom
feat/vistadream-package
Open

Feat/vistadream package#2
pablo-rerun wants to merge 2 commits intomainfrom
feat/vistadream-package

Conversation

@pablo-rerun
Copy link
Collaborator

This pull request introduces the initial project structure and core functionality for the vistadream package, focusing on the integration and sampling of the Flux generative model. It includes project configuration, core API and CLI for Flux-based image generation, and utility modules for model versioning and initialization. The most important changes are grouped below.

Project setup and configuration:

  • Added a pyproject.toml for project metadata, dependencies, build system configuration, and development tool settings. This includes dependencies on einops and monopriors, and configures tools like ruff for linting.

Flux model integration and API:

  • Implemented the flux subpackage with the following modules:
    • api.py: Provides an ImageRequest class for interacting with a remote Flux API, including input validation, request/response handling, and image download utilities.
    • cli.py: Implements a CLI for local sampling from the Flux model, supporting interactive and batch modes, prompt parsing, and NSFW filtering.
    • __main__.py: Entrypoint for CLI execution.
    • _version.py: Auto-generated version information for the Flux subpackage.
    • __init__.py: Handles version import and package root path setup.

VistaDream core and outpainting:

  • Added api/flux_outpainting.py, which defines a FluxOutpaintingConfig dataclass and a main function to perform outpainting using the Flux model, including depth prediction, camera parameter logging, and visualization via Rerun.

Development and runtime utilities:

  • Updated the package-level __init__.py to enable runtime type checking with beartype when the PIXI_DEV_MODE environment variable is set.

pablovela5620 and others added 2 commits March 15, 2026 11:04
Migrate VistaDream (single-image 3D scene reconstruction) from standalone
repo into the monorepo as packages/vistadream. Includes pixi.toml feature
with CUDA/gsplat deps, source code, CLI tools, example data, and lint fixes
for ops/ code. Vendored Flux code gets per-file ruff ignores.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gsplat's runtime CUDA extension build needs cuda_runtime.h and cicc,
which live in conda-forge's targets/x86_64-linux/ and nvvm/bin/ dirs
respectively. Set CUDA_HOME, CPATH, and PATH in vistadream activation
so the host compiler and nvcc can find them.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants