# 10 — Video Gallery > **View with playback:** [neuramatedreamgen.pages.dev](https://neuramatedreamgen.pages.dev/) > > GitHub does not render inline video in markdown. Open the link above to play all clips in your browser. Reference index of the videos referenced throughout this findings document. Two groups: the **human-demo inputs** (LAPA / source-video material) and the **GR1-rendered outputs** (what cosmos-predict2 produced from various inference configurations). ← [Back to README](../README.md) | [← 09 Next Session](09-next-session.md) --- ## Factory LAPA videos Real factory recordings of a human (Shaun) performing the paper-into-feeder task from an ego-POV. These are the human-demo source material — used in our v2v_sdedit experiments (sessions 2-3) as the motion source to be repainted as GR1, and likely useful as source for retargeting / IDM if the project pursues the human-demo training path (see [05 Workflow → cross-embodiment shortcut](05-workflow.md#cross-embodiment-shortcut-human-videos--gr1-output-doesnt-work)). ### `ego_shaun.mp4` [Link to Video](https://neuramatedreamgen.pages.dev/assets/ego_shaun.mp4) | [Download](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_shaun.mp4) - **Resolution**: 480 × 640 (3:4 portrait) - **FPS**: 30 - **Duration**: 29.97 s (888 frames) - **Description**: Original human-ego recording. Full-length demonstration of the cart→feeder paper-handling action at native capture rate. Source-of-truth for the human-demo data. ### `ego_shaun_16fps.mp4` [Link to Video](https://neuramatedreamgen.pages.dev/assets/ego_shaun_16fps.mp4) | [Download](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_shaun_16fps.mp4) - **Resolution**: 480 × 640 (3:4 portrait) - **FPS**: **16** (resampled from 30) - **Duration**: 29.94 s (479 frames) - **Description**: Same content as `ego_shaun.mp4`, downsampled to cosmos's native 16 fps so the frame timing matches the model's expected input cadence. This is the file v2v_sdedit consumed in the later sessions — feeding 30 fps in caused the model to silently truncate to ~93 source frames, dropping the back ~half of the action. --- ## Factory Robot videos GR1-rendered outputs from various experiments. Use these as visual reference when reading the experiment log and findings. ### `robot_ego_v2.mp4` [Link to Video](https://neuramatedreamgen.pages.dev/assets/robot_ego_v2.mp4) | [Download](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/robot_ego_v2.mp4) - **Resolution**: **768 × 432 (16:9 landscape)** ← matches cosmos training distribution - **FPS**: 16 - **Duration**: 5.81 s (93 frames) - **Description**: A GR1 ego-POV clip in cosmos's native training spec (16:9 landscape, 432 H × 768 W, 93 frames at 16 fps). The **only** video on disk that already matches the on-distribution geometry from [02 Findings §7](02-findings.md#7-training-distribution-is-169-landscape-at-432768-wed-been-running-34-portrait). Likely sim-rendered. If sim infrastructure already produces clips in this format, that materially de-risks Phase 1 of the [recommended path forward](06-path-forward.md#phase-1--sim-data-collection-days-no-gpu-cost-beyond-local-sim). ### `ego_fourier_prox_feeder_s42.mp4` [Link to Video](https://neuramatedreamgen.pages.dev/assets/ego_fourier_prox_feeder_s42.mp4) | [Download](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_fourier_prox_feeder_s42.mp4) - **FPS**: 16 - **Duration**: 5.81 s (93 frames) - **Description**: `video2world_gr00t` output, seed 42, guidance 8, on `fourier_prox_feeder.png` (close-up of GR1 holding paper near the press's angled feed deck). The "best in proximity" experiment from session 5 — the GR1 was already positioned at the feeder. Result: pivot + slight angle but never placed. See [01 Experiments §5](01-experiments.md#5-video2world_gr00tpy-with-fourier_prox_feederpng-session-5). ### `ego_fourier_worker_v2w_s42.mp4` [Link to Video](https://neuramatedreamgen.pages.dev/assets/ego_fourier_worker_v2w_s42.mp4) | [Download](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_fourier_worker_v2w_s42.mp4) - **Resolution**: 480 × 640 (3:4 portrait — off-distribution) - **FPS**: 16 - **Duration**: 5.81 s (93 frames) - **Description**: `video2world_gr00t` output, seed 42, on `fourier_worker.png` (wide-shot showing GR1, paper, and the press from a distance). From the 3-seed sweep in session 4. Result: GR1 pivots toward the feeder, paper occludes the feeder during the pivot, model approaches but never lowers paper onto it. See [01 Experiments §3](01-experiments.md#3-video2world_gr00tpy-seed-sweep-with-fourier_workerpng-session-4). ### `ego_fourier_over_place_s42.mp4` [Link to Video](https://neuramatedreamgen.pages.dev/assets/ego_fourier_over_place_s42.mp4) | [Download](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_fourier_over_place_s42.mp4) - **Resolution**: 480 × 640 (3:4 portrait — off-distribution) - **FPS**: 16 - **Duration**: 5.81 s (93 frames) - **Description**: `video2world_gr00t` output, seed 42, on `fourier_over.png` (closer-proximity image of GR1 near the press's front cabinet, no visible feeder deck). From session 5. Result: paper placed on the flat cabinet surface — the wrong target — because the actual angled feed deck wasn't visible from this POV. See [01 Experiments §4](01-experiments.md#4-video2world_gr00tpy-with-fourier_overpng-session-5). --- ## At-a-glance comparison | File | Group | Res | FPS | Frames | Distribution | Notes | |---|---|---|---|---|---|---| | [ego_shaun.mp4](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_shaun.mp4) | LAPA | 480×640 | 30 | 888 | n/a (source) | 30 s human demo, native fps | | [ego_shaun_16fps.mp4](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_shaun_16fps.mp4) | LAPA | 480×640 | 16 | 479 | n/a (source) | 30 s human demo, fps-matched to cosmos | | [robot_ego_v2.mp4](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/robot_ego_v2.mp4) | Robot | **768×432** | 16 | 93 | **on-distribution** | Only on-distribution video on disk | | [ego_fourier_prox_feeder_s42.mp4](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_fourier_prox_feeder_s42.mp4) | Robot | 480×640 | 16 | 93 | off-distribution | session 5 GR1 output | | [ego_fourier_worker_v2w_s42.mp4](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_fourier_worker_v2w_s42.mp4) | Robot | 480×640 | 16 | 93 | off-distribution | session 4 GR1 output | | [ego_fourier_over_place_s42.mp4](https://github.com/Neuramate/NeuramateDreamgen/releases/download/v0/ego_fourier_over_place_s42.mp4) | Robot | 480×640 | 16 | 93 | off-distribution | session 5 GR1 output | The distribution column flags the aspect-ratio mismatch documented in [02 Findings §7](02-findings.md#7-training-distribution-is-169-landscape-at-432768-wed-been-running-34-portrait). `robot_ego_v2.mp4` is the format target for Phase 1 sim data; everything else in this gallery is either input source (LAPA) or off-distribution legacy output (Robot, 3:4). --- ← [Back to README](../README.md)