Interfaze-1-lite · Launch Video Breakdown: Hook, Pacing & Motion Design

Scene-by-scene timeline & spoken transcript

1. 0:00 – 0:15

   The Hook

   The Problem with Ambiguity

   “Some tasks can't be almost right. OCR, speech-to-text, extraction, classification and more. It returns confidence scores and metadata like bound.”

   On screen

   Some tasks can't be almost right. <task>ocr</task> statement.png Account Summary Customer Service Account Number NB-2468-1357 (888) 555-6789 Statement Period April 1, 2024 - June 30, 2024 www.northridgeinv.com Investment Type Brokerage service@northridgeinv.com Primary Account Holder MICHAEL T. ANDERSON PO Box 98765 Account Type Individual - Brokerage Boston, MA 02199-7854 Tax Identification XXX-XX-9821 Holdings Summary Security Description Quantity Market Value % of Account Est. Annual Income Domestic Equities Vanguard Total Stock Mkt ETF (VTI) 1,200.000 $688,452.47 95.70% Vanguard S&P 500 ETF (VOO) 500.000 $324,750.00 38.07% $4,128.00 $254,915.00 35.47% $2,152.00 <task>speech_to_text</task> call.mp3 2 speakers 00:00 00:20 00:40 01:00 SPEAKER_01 Who interviewed you at YC? SPEAKER_00 We actually had two interview... SPEAKER_01 Two separate 10-minute interv... SPEAKER_00 I think they just want to und... *Statement Date July 8, 2024* *MICHAEL T. ANDERSON* *Tax Identification XXX-XX-9821* *Individual - Brokerage* 0.99 0.98 0.99 0.99 0.99

   Camera

   Static, centered view on a clean, minimalist canvas. Elements animate in and out around the central text.

   Motion

   Text reveal with subtle scale and opacity, UI elements sliding in from edges, progress bars filling, and confidence scores animating up. Data points highlight with color changes.

2. 0:15 – 0:45

   Product Reveal

   Introducing Interfaze-1-lite: Mixture-of-Architecture

   “Introducing interfaze-1-lite, the first general open-weight model for deterministic work. It runs on a Mixture-of-Architecture (MoA). Unlike a traditional model with a fixed set of weights, MoA is different. It routes requests to specialized models, each with its own architecture.”

   On screen

   model : "interfaze-1-lite" One open-weight model. an open-weight model a new architecture It runs on a Mixture-of-Architecture (MoA) gpu-03 gpu-04 gpu-05 gpu-06 gpu-07 gpu-08 gpu-15 gpu-16 gpu-17 gpu-18 gpu-19 gpu-20 gpu-27 gpu-28 gpu-29 gpu-30 gpu-31 gpu-32 gpu-39 gpu-40 gpu-41 gpu-42 gpu-43 gpu-44 0 requests then scale it out Traditional model model.safetensors fixed weights layer_01 layer_02 layer_03 layer_04 layer_05 layer_06 layer_07 layer_08 layer_09 layer_10 and only as good config.json { "architecture": "monolithic", "weights": "fixed", "config": "locked" } output ok Mixture-of-Architecture is different. statement.png call.mp3 cam_04.mp4 Document model own architecture - own weights Speech model own architecture - own weights Vision model own architecture - own weights Transformer routes + reasons architecture.

   Camera

   Primarily static, with elements appearing and connecting. Zoom-ins on specific UI components.

   Motion

   Node-based diagrams animate with connecting lines and data flow. Text highlights and color changes emphasize key terms. GPU bars animate to show scaling.

3. 0:45 – 1:10

   Feature Teaser

   Deterministic Outputs and Performance

   “This means interfaze-1-lite can transcribe every holding line item, classify each one by asset class, and return a structured JSON. Their results come with confidence scores, so it won't hallucinate. Every output is grounded in facts. It leads general-purpose models in document OCR, speech recognition, visual grounding, and structured output. You can self-host it under Apache 2.0. High reliability, low cost.”

   On screen

   Mixture-of-Architecture 1 request "transcribe every holding line item, then classify each one by asset class." statement.png response.json : holdings description market value class Vanguard Total Stock Mkt ETF (VTI) $324,687.50 Transformer routes + reasons Document model transcribes every line Classifier labels each line item request, transcribe Document *NB-2468-1357* 0.99 Speech SPEAKER_01 0.00-1.08s Vision scaffolding : [419, 0, 113,... Their results come Structured merge typed, validated { document: (...), speakers: [...], objects: [...] } Transformer reads the facts grounded hallucinate. Every response.json { "account_number": { "value": "NB-2468-1357", "confidence": 0.99, "bounds": [472, 41, 231, 12] }, "speaker": { "id": "SPEAKER_01", "start": 0.0, "end": 1.08 }, "objects": [[ "label": "scaffolding", "bounds": [419, 0, 113, 251] ]] } if (account_number.confidence >= 0.95) approve() auto Leads general-purpose models vs. best of Gemini-3.7-Flash, Claude-Sonnet-5, GPT-5.4-Mini, Grok-4.3 interfaze-1-lite best other model Document OCR olmOCR 83.7% 83.5% Claude-Sonnet-5 Speech recognition VoxPopuli WER lower is better 3.0% 4.0% Gemini-3.7-Flash Visual grounding RefCOCO 21.7% 80.9% Gemini-3.7-Flash Structured output SOB value accuracy 80.2% Gemini-3.7-Flash visual grounding self-host terminal $ hf download interfaze-ai/interfaze-1-lite fetching weights 100% $ pip in License: Apache 2.0 1x 80 GB GPU offline under Apache 2.0. api index.ts const res = await interfaze.chat.completions.create({ model: "interfaze-1-lite", messages, }); High reliability Low cost low cost.

   Camera

   Dynamic camera movements, including pans and zooms to follow data flow and highlight specific metrics. Split-screen layouts for code and output.

   Motion

   Complex data flow animations, bar chart races, terminal command line typing effects, and code highlighting. Checkmark animations for benefits.

4. 1:10 – 1:15

   Call to Action

   Unleash Your Pipelines

   “We can't wait to see what you build. Side effects may include fewer retries, fewer regex parsers, and fewer humans in the loop. Your pipelines may begin reading every document, transcribing every call, and watching every camera. Ask your developer if interfaze-1-lite is right for you.”

   On screen

   We can't wait to see statement.pdf <task>ocr</task> classify route support_call.mp3 <task>speech_to_text</task> summarize site_cam.jpg IMPORTANT SAFETY INFORMATION interfaze-1-lite Side effects may include fewer retries, fewer regex parsers, and fewer humans in the loop. Your pipelines may begin reading every document, transcribing every call, and watching every camera. Do not use interfaze-1-lite if you enjoy fixing broken JSON by hand. Consult your GPU before self-hosting. Confidence scores may cause sudden trust in your automations. Bounding boxes may appear around things you forgot were there. Manual review queues may shrink without warning. Results depend on input quality; blurry photos remain blurry. Keep out of reach of hand-written regex. Your pipelines may Ask your developer if interfaze-1-lite is right for you.

   Camera

   Wide shot revealing a network of data flows, then a static shot for the humorous disclaimer.

   Motion

   Node-based data flow animation, text highlights, and a final bold call to action with a playful disclaimer.