Skip to content

Cam2V

Cam2V is an interactive FlashDreams v2 camera-to-video application. It turns keyboard camera controls, a prompt, and model-specific starting inputs into a long-running generated view.

Four-second Cam2V demo preview generated by FlashDreams.

Quick Start:

LingBot-World SANA-WM

Controls

Keys Action
W / S, Up / Down Move forward / backward.
A / D, J / L Yaw left / right.
Q / E Strafe left / right.
I / K Pitch up / down.

Losing browser focus clears held keys. Presentation pacing includes optional post-processing; model-only timing remains separate.

Demo Arguments

Argument Description
--prompt TEXT, --prompt-path PATH Set the text prompt directly or from a file.
--image-path PATH Set the first frame.
--pose-path PATH Set a pose trace used by the model adapter.
--intrinsic-path PATH Set camera calibration.
--world-scale FLOAT Set the camera-motion scale.
--example-data, --no-example-data Enable or disable packaged/example inputs.
--example-idx INT Select an example input.
--device DEVICE Select the model device.
--total-blocks INT Stop after this many autoregressive chunks.
--warmup-blocks INT Set chunks excluded from steady-state FPS.
--ui, --no-ui Enable or disable the controls/status overlay.
--compile, --no-compile Enable or disable model compilation.
--postprocess-preset NAME Select an installed generated-video post-processing preset.
--postprocess-chunk-size {8,16} Set the steady post-processing window. Default: 8.
--postprocess-compile, --no-postprocess-compile Enable or disable postprocessor compilation.
--postprocess-comparison-ui, --no-postprocess-comparison-ui Show synchronized original and post-processed panes.
--seed INT Override the diffusion seed.

Post-processing comparison

With a postprocessor configured, --postprocess-comparison-ui shows the original video on the left and the post-processed video on the right:

uv run --no-sync flashdreams-run-v2 DEMO_MODEL_SLUG \
  --mode webrtc --host 0.0.0.0 --port 8089 -- \
  --postprocess-preset PRESET --postprocess-comparison-ui

The mode requires the UI and a frame-preserving postprocessor. It doubles the delivered video width and may increase encoding and transfer cost.

Model-free demo

Exercise the Cam2V UI without loading a model:

uv run --no-sync flashdreams-run-v2 cam2v-dummy --mode webrtc \
  --host 0.0.0.0 --port 8089 -- \
  --step-wait-seconds 0.9 --frames-per-chunk 12

Tests

uv run --no-sync pytest apps/cam2v/tests -m ci_cpu