Sqwibbl Box
Generation runs on your desk. Not a server queue.
The Sqwibbl Box is the superfan experience — a Windows desktop agent that runs local image-to-video generation on your own hardware, eliminating server queue wait times and cloud API costs entirely.
Local Generation
Wan 2.2 1.3B Image-to-Video runs on your GPU. No LoRA files to manage. No cloud queue. No per-generation API cost.
How It Works
The desktop generation flow
Select your snippet
Pick a track from your Living Album and choose an 8, 10, 20, or 30-second window — the chorus, the bridge, the moment that hits hardest.
Album Brain responds
Sqwibbl queries the Album Brain PLM and surfaces the artist's guiding visual direction for that specific section — their own words about what that moment looks and feels like.
Artist-authored promptReference image downloaded
The reference image is downloaded from the PLM image library at request time (~1–5MB). It's passed directly into the local ComfyUI I2V workflow — no LoRA files to download or manage.
~1–5MB per imageLocal I2V generation
Wan 2.2 1.3B I2V (GGUF Q6_K, hotswap via GpuRouter) generates your video clip on your local GPU. No server round-trip, no queue, no one else's job ahead of yours.
3–6 minutes per clipAudio muxed
Sqwibbl muxes the artist's master audio snippet onto your generated video — permanently married. FFmpeg handles the mux; the result is a clip that's yours to publish.
Artist audio + your visualsPublish to SqwibblSocial
Push to the CDN and SqwibblSocial via the same provenance pipeline. 40/40/20 attribution is active from the first view — revenue starts flowing immediately.
40% to youHardware
What you need to run the Box
Recommended Hardware
- HP Omen 16" — RTX 5060 16GB VRAM
- Or equivalent: any laptop/desktop with 16GB+ VRAM
- Windows 11
- 512GB+ NVMe SSD for model weights and output staging
What's Running Locally
- Gemma 4B (Q4, ~3GB VRAM) — Sqwibbl conductor LLM
- Wan 2.2 1.3B I2V (GGUF Q6_K, ~2GB loaded)
- ComfyUI with WanVideoWrapper I2V workflow
- No LoRA download — just the PLM endpoint URL
Performance
What you get — and how fast
480p Generation
~3–6 minutes
per clip at 480p on 16GB VRAM, no offloading needed for the 1.3B model. No server queue wait — it's just you and your GPU.
Zero Per-Album Setup
No LoRA files to download. No training files to manage. Just provide the PLM endpoint URL for each album you subscribe to — Sqwibbl handles the rest.
Local vs. Cloud
Cloud generation queues you behind other users and meters every generation. Local generation on the Sqwibbl Box is included in your Living Album subscription — bonus creation allowance for local-gen users.
The Visual Grounding
Same quality anchor as server-side generation
Without Living Album
A fan using a generic AI video tool gets generic output — no connection to the music, no artist's visual world, no approved aesthetic. Anyone can do it.
With Sqwibbl Box
Your reference image comes from the artist's actual approved image library. Wan 2.2 uses it as the visual anchor for your generated video. The output feels like it came from inside the album — because the artist's images are what guided it.
The PLM image library is the differentiator — not a trained LoRA. The curation is the moat. The artist's images are the reason fan generation on the Living Album looks and feels right.
Subscriber Benefit
Why go local?
No Queue Wait
Server-side generation means sharing GPU resources with other users. The Sqwibbl Box is yours alone — your creation runs immediately when you're ready.
No Cloud API Cost
Every server-side generation carries a token meter cost. Local generation on the Box is covered by your subscription — no surprise per-generation charges.
Bonus Creation Allowance
Subscribers who use the Sqwibbl Box get a bonus creation allowance for local-gen users — more clips per month than the web-tier equivalent.
Ready to run generation on your desk?
Download the Sqwibbl Box and connect your Living Album subscription. Your GPU. Your queue. Your clips.