kevinqz/LingBot-Video-Dense-1.3B-DiT-CoreAI

Canonical: kevinqz/LingBot-Video-Dense-1.3B-DiT-CoreAI — source of truth.

LingBot-Video Dense 1.3B DiT (fabric)

An Apple Core AI conversion of the denoising DiT (one flow-matching denoise step) of the LingBot-Video video diffusion model from robbyant/lingbot-video-dense-1.3b. Produced by coreai-fabric and indexed by coreai-catalog.

One component, not the full video pipeline. A video diffusion model is separable blocks — text encoder, VAE encoder, denoising DiT, and VAE decoder. This asset is ONLY the DiT denoise-step: (noisy latent, timestep, text embeddings) → predicted velocity. The host owns the text encoder, the flow-UniPC sampler loop that calls this step, and the VAE decode. It does not, by itself, generate video from a prompt.

Model facts

Field	Value
Parameters (upstream)	1.3B
Architecture	diffusion
Capabilities	text-to-video
Deployable core	DiT denoise-step (latent + timestep + text → velocity)
Velocity shape	—
Quantization / precision	none / float32
On-disk size	5.1 GB
Asset kind	single-graph diffusion transformer (one denoise step)
assetVersion	2.0

Use it — this needs host code you supply

The bundle is a single static-size graph: hidden_states (noisy latent), timestep, encoder_hidden_states (text) in → velocity out. You supply the sampler loop (flow-UniPC), the text encoder, and the VAE decode in your host code (Swift or Python).

pip install coreai-catalog && coreai-catalog install lingbot-video-dense-1.3b-dit

Requirements

• Deployment: macOS 27.0+ / iOS 27.0+, Xcode 27+. The asset serializes with minimum_os v27, so the on-device Swift runtime requires macOS/iOS 27+. A Mac on macOS 26 can convert and inspect it but not run it on-device.
• Apple Silicon.

Verification (output parity)

• Gate A (structure): passed — the bundle's layout + metadata were validated; the graph loads.
• Gate B — graph_output_cosine: 1.000000 min output cosine (median 1.000000) vs the fp32 torch DiT denoise-step over 8 seeded (latent, timestep, text) inputs, measured on apple_silicon. Certifies the export computes the SAME output as the source — a conversion-fidelity metric, not end video quality.
• This certifies the export is numerically faithful to the source VAE decoder — it does NOT certify end video quality. Reproduce with coreai-fabric verify.

Provenance

Field	Value
Base model	robbyant/lingbot-video-dense-1.3b @ f9789a7d9b4772a47aba62d4eb5282ddefd1da21
Converted by	models/lingbotvideo/dit_export.py (version not reported)
Recipe	lingbot-video-dense-1.3b-dit (recipe_source: fabric)
Precision / quantization	float32 / none
Conversion date	2026-07-09

Machine-readable, in this repo: parity-report.json · reproduce-manifest.json · LICENSE.

License and attribution

Weights licensed apache-2.0 — see the bundled LICENSE. This artifact is a converted derivative of the base VAE: its weights were converted to Apple Core AI format. The conversion itself is community work.

Links

• Base model: robbyant/lingbot-video-dense-1.3b
• Reproduce: recipe lingbot-video-dense-1.3b-dit
• Index: coreai-catalog
• HF Collection

The on-device Core AI ecosystem

• coreai-fabric — the reproducible recipe → .aimodel pipeline that produced this asset.
• coreai-catalog — the index of Core AI models with provenance and integration snippets.
• apple/coreai-models — Apple's official exporters and runtimes.

Not affiliated with Apple

Community conversion. Not produced, hosted, or endorsed by Apple. Apple and Core AI are trademarks of Apple Inc., used here only to describe the target runtime/format.