Not runnable todayEmbeddingVerified 2026-11-05
InternVideo2 Stage2 1B (224p, 4 frames)
OpenGVLab/InternVideo2-Stage2_1B-224p-f4
InternVideo2 Stage2 1B (224p, 4 frames) is a video and text embedding model (Apache-2.0). Not runnable on Mixpeek today.
Something blocks it today, such as the license, remote code or a GPU-only runtime. The reason is below.
No demo for InternVideo2 Stage2 1B (224p, 4 frames) yet
A video, image, slideshow or playground goes here once we record one on our own data.
What it is good and bad at
- Best at
- Video-text retrieval and zero-shot action recognition on short clips; motion-aware where frame CLIP models are not.
- Worst at
- Only 4 frames at 224p per clip: long or slow scenes need splitting; research code base, not a packaged loader.
- Task
- Embedding
- Input
- Video, Text
- Output
- Vector
- Use cases
- Video understanding, Action recognition
Templates that use it
No template uses this model yet. Browse templates.
Input and output
Input
- video
- frames
- 4
- resolution
- 224x224
- sampling
- uniform over the clip
- text
- max_tokens
- 40
Output
- vector
- dims
- projection dim per the repo config (not verified here)
- distance
- cosine
- type
- dense
How Mixpeek wires it
- Extractors
- None today
- Feature URIs
- None today
- Vector indexes
- None
- Retriever stages
- None
- Combines with
- scene splits (transcoding/multimodal extractor) -> clip-level vectors -> feature_search with a text query
How it runs
- Status
- Not runnable today Weights are a single .pt pickle and the model is defined only in the InternVideo GitHub repo (multi_modality), so it needs vendored model code plus a safetensors conversion before it can be a plugin; GPU in practice for 1B.
- Serve apps
- None
- Runtime
- InternVideo repo (multi_modality InternVideo2 Stage2)
- Weights
- Pickle, no remote code
- Parameters
- 1,000 M
- Size on disk
- 2,821 MB
- Preprocessing
- Sample 4 frames uniformly, resize to 224, normalize ImageNet mean/std.
- Postprocessing
- L2-normalize video and text projections.
- Serving
- batch
- Hardware
- Needs a GPU
Scale to zero
- Minimum replicas
- 0
- Maximum replicas
- 1
- Idle timeout
- 600 s
- Memory
- 8 GB
- CPUs
- 2
- GPUs
- 1
- Cold start, node warm
- Not recorded
- Cold start, new node
- Not recorded
Measured cost
Unmeasured
No committed measurement; plan row 3 measures it in the standalone image.
Our evaluations
Unmeasured
We have not evaluated this model on our own data yet. Publisher results are in the sources below.
Questions
- What is InternVideo2 Stage2 1B (224p, 4 frames) best at?
- Video-text retrieval and zero-shot action recognition on short clips; motion-aware where frame CLIP models are not.
- What are the limits of InternVideo2 Stage2 1B (224p, 4 frames)?
- Only 4 frames at 224p per clip: long or slow scenes need splitting; research code base, not a packaged loader.
- Can InternVideo2 Stage2 1B (224p, 4 frames) be used commercially?
- Yes. Apache-2.0 per the hub tag; access gated by an automatic agreement on the hub.
- How does InternVideo2 Stage2 1B (224p, 4 frames) run on Mixpeek?
- Weights are a single .pt pickle and the model is defined only in the InternVideo GitHub repo (multi_modality), so it needs vendored model code plus a safetensors conversion before it can be a plugin; GPU in practice for 1B.