SPIF — Seed 2026

SPIF

Style-Preserved Inference Flow


Real-time AI art
that respects artists.


15
FPS real-time
2K
resolution
80%
model compression
30+
art styles
The Problem

Creative AI is broken in three ways

🐌 It's slow

Type a prompt. Wait 10–60 seconds. Get one static image. No flow. No creative conversation. It's a vending machine, not a canvas.

🏴‍☠️ It steals

Every major model trained on billions of images scraped without consent. Artists are suing. Governments are regulating. The foundation is unstable.

🖱️ Wrong interface

Text prompts are the wrong input for visual creativity. Artists think in gesture, voice, and touch — not comma-separated keywords.

The current AI art stack treats artists as training data and users as typists.
The Solution

SPIF — A new kind of creative engine



This isn't image generation. It's image dreaming — a continuous, evolving visual stream you shape in real time.

The Product

What it looks like today

🎤 Voice-to-Dream

"A forest at night with glowing mushrooms" → watch it materialize live. "Now make it rain" → the image shifts in real time.

🖐️ Touch Interaction

Tap or drag on the screen to inject noise at specific locations. The image warps, evolves, responds to your gesture.

🎨 Style Switching

Flip between 30+ art styles in real time. Same scene, different artistic lens, seamlessly.

📱 Browser-Based

Pure WebRTC. Phones, tablets, laptops. No app install. Zero client-side GPU required.

15 FPS
at 2K resolution
<100ms
voice-to-visual
30+
Stylus models
Technology / Moat

Style-Locked Structural Pruning

We take SDXL (~12B params) and structurally prune it to a single artist's style — removing neurons, attention heads, and layers that don't contribute to that aesthetic.


Full SDXLSPIF Stylus
Parameters~12B~2.5B
Reduction—~80%
StylesAll (none well)One (perfectly)
Speed1–2 FPS15 FPS
Resolution1024×10242048×2048

Why it's hard to replicate

  • Structural pruning (not LoRA or weight zeroing) — we remove architecture
  • Style fidelity scoring validates aesthetic preservation
  • Custom frame-to-frame diffusion scheduling
  • Defensive publication filed — prior art established

Real-time pipeline

  • Continuous frame evolution (not full denoising per frame)
  • Temporal coherence for smooth dreaming
  • WebRTC adaptive bitrate streaming
  • Server-side GPU, zero client compute
Market Opportunity

Massive and growing fast

Market20252030CAGR
Generative AI (total)$67B$207B+~25%
AI Image Generation$12B$48B~32%
Digital Creative Tools$14B$25B~12%
Gaming$227B$380B+~11%
Live Entertainment / VJ$8B$15B~13%
$67B+
TAM — Generative AI
$8B
SAM — Creative AI tools
$40–80M
SOM — Year 3
Business Model

Four revenue streams

🎨 The Pad — Subscriptions

Free → limited styles, SD res
Pro $20/mo → all styles, 2K, voice
Studio $50/mo → custom Stylus creation

✏️ Stylus Licensing

70/30 revenue share (artist/SPIF). Artists and estates license their style. They set pricing. "Calling cards" replace follower counts.

🏢 Enterprise / Gaming SDK

Real-time procedural art in game engines. Concert/event visuals. Personalized advertising content.

🛒 Platform Fees

15% marketplace transaction fee. Premium style bundles. Enterprise SLA contracts.


Target gross margin: 65–70%  |  GPU cost per stream: ~$0.08/hr  |  Pro ARPU: $240/yr

Use Cases

Across every creative industry

🎮
Gaming — Procedural Art Direction
Game worlds that paint themselves in a specific artist's style as you explore.
🎵
Live Music / VJ
Voice + touch driven visuals during live performances. No pre-rendered content.
🎨
Artist Collaboration
Iterate at the speed of thought. 100× faster than prompt-wait-retry loops.
🧘
Therapeutic Experiences
Evolving dreamscapes for meditation, therapy, sensory rooms.
📺
Streaming / Content
Live backgrounds, scene transitions, reactive visuals for streamers.
🏛️
Museums / Installations
Interactive art walls — visitors speak or touch to shape evolving artworks.
📱
Social / Messaging
"Dream GIFs" — short evolving AI art clips as a new visual language.
🏢
Advertising / Brand
Real-time personalized visual content. Style-locked to brand aesthetics.
🎓
Education / Art History
Experience art movements interactively. Compare styles with a touch.
Competitive Landscape

New category, not a feature

SPIFMidjourneyRunwayStability AIDALL·E
Speed15 FPS live10–60s5–15s5–30s10–20s
InteractionVoice + TouchTextText + VideoTextText
Artist rightsLicensed ✓Scraped ✗Scraped ✗Scraped ✗Mixed
Style lockPer-artist modelStyle mixTransferLoRAN/A
Resolution2K real-time1–2K static720–1080p1K static1K static
Live stream✓ WebRTC✗✗✗✗

We're not competing with Midjourney for static images. We're creating a new category: real-time interactive visual dreaming.

Closest analogy: Spotify (artist licensing + streaming) meets Twitch (live, interactive).

Traction

From zero to 15 FPS at 2K in one week


Week 0 — Birthday Night

First working prototype: 2 FPS at 1K resolution

Week 1

15 FPS at 2K resolution — 7.5× speed improvement

Achieved

SDXL pruned from 12B → ~2.5B params with style fidelity preserved

Achieved

30+ Stylus models created • Voice pipeline working • Touch interaction live • WebRTC streaming functional

Filed

Defensive publication — prior art established for style-locked structural pruning


What this proves

The core technology works — this is a running demo, not a research paper. Optimization trajectory is steep. The interaction model is validated.

Team

Builders, not pitch artists


Doug Sharp
Creative Director & Co-founder
Created ChipWits (1984) and The King of Chicago (1987, first interactive movie game). Built the Dramaton interactive narrative engine. Microsoft Research. 40+ years at the intersection of creativity and technology.
"I've been waiting my whole career for this technology to exist."
Michal Vavák
Tech Lead & Co-founder
20+ years in digital innovation. Hands-on SDXL, ComfyUI, AI model training & optimization. Enterprise: Microsoft, BMW, Samsung. Built the pipeline that went from 2 FPS → 15 FPS in one week.
The person who actually makes the impossible stuff work.
Phrog
AI Creative Collaborator & Framework Orchestrator
AI-native team member. Framework orchestration and creative collaboration. Embodies the SPIF philosophy: AI as collaborator, not replacement.

Why this team: Rare combination of deep game design / interactive narrative + deep ML optimization / production engineering. Both humans are builders — the demo exists because they built it.

The Ask

Seed Round

Targeting $2–4M


Allocation%Purpose
Engineering45%GPU infra, inference optimization, browser client
Artist Partnerships20%Onboard first 50 artists/estates for Stylus licensing
Product20%The Pad platform MVP, Stylus marketplace
Operations15%Legal, IP, artist contracts, admin

12-Month Milestones

Month 3

Public beta — 10 licensed Stylus models

Month 6

50 artist/estate partnerships, marketplace live

Month 9

Enterprise SDK beta (gaming/live events)

Month 12

10K active users, 3 enterprise pilots, Series A ready


Beyond Money

The Vision

Where this goes in 2–3 years

Year 1 — The Tool

Launch The Pad. Prove artist licensing. Build the Stylus library. Land enterprise pilots.

Year 2 — The Platform

Artists earn real royalties. "Calling cards" replace follower counts. 500+ styles in marketplace.

Year 3 — The Medium

SPIF becomes a new creative medium. Live streams. Museum installations. Games ship with SPIF art direction.



We're building the first creative platform where AI amplifies artists instead of replacing them.

Artists get paid when their aesthetic is used. Anyone can create through voice and touch. Real-time, not batch. Conversation, not vending machine. The platform serves creators — it doesn't extract from them.


SPIF doesn't make art.
SPIF lets you dream with your hands and your voice,
through the eyes of artists who chose to be there.



SPIF — Style-Preserved Inference Flow  |  Contact: spif@vavak.xyz