Engineering • Primary Division
Enterprise AI Hardware Implementation
Turnkey local ComfyUI workstations for audiovisual studios and enterprises requiring predictable operating costs, asset privacy, and workflow continuity.
Verified Workstation Profiles (4 Tiers)
Technical sizing without generic public list pricing; hardware is quoted post technical diagnostic.
Estudio base
Entry 16 GB- GPU:
- 1× RTX 5070 Ti (16 GB VRAM)
- RAM:
- 64 GB DDR5
- 3s 1080p speed:
- ~2.3 min (~36 clips/day, 10 attempts)
Single-operator entry workstation for 720p/1080p with LTX-2.5 22B, Kandinsky 6.0 Lite, and Wan 2.2 5B.
Estudio chico
Recommended- GPU:
- 1× RTX 5080 (16 GB VRAM)
- RAM:
- 64 GB DDR5
- 3s 1080p speed:
- ~2.1 min (~38 clips/day, 10 attempts)
Primary daily production profile (~38 clips/day at 10 attempts per final clip).
Estudio doble
2 Stations / 2 GPUs- GPU:
- 2× RTX 5080 (16 GB VRAM each; independent stations, not pooled 32 GB)
- RAM:
- 64 GB DDR5 per independent host (or mandatory 128 GB 2×64 GB kit + >= 1,200 W PSU + isolated --cuda-device 0/1 instances if consolidated)
- 3s 1080p speed:
- ~2.1 min / GPU (~77 clips/day, 10 attempts)
Doubles parallel throughput (~77 clips/day combined); retains 16 GB per-GPU model ceiling.
Profesional
32 GB VRAM- GPU:
- 1× RTX 5090 (32 GB VRAM)
- RAM:
- 64 GB DDR5 (128 GB via 2×64 GB optional)
- 3s 1080p speed:
- ~0.90 min (~79 clips/day, 10 attempts)
Exclusive tier supporting Wan 2.2 14B (28.6 GB) up to 720p and Kandinsky 6.0 Pro without heavy RAM offload.
Mandatory 2×64 GB Memory Topology Rule
A single LTX-2.5 pipeline reaches ~45 GB RAM. Every 128 GB configuration uses 2×64 GB DIMMs exclusively (avoiding 4×32 GB 2DPC clock penalties) and keeps >= 64 GB DDR5 per concurrent ComfyUI operator.
Verified Open-Weights Catalog & Licensing
Only audited open-weights models with verified VRAM footprints and commercial terms.
| Model | VRAM / Res. | License | Tier Fit & Commercial Scope |
|---|---|---|---|
| LTX-2.5 22B (destilado / cuantizado) | 16 GB · 1080p | LTX-2.x Community License | All tiers (16 GB & 32 GB), native 3 s + audio Royalty-free commercial use below US$10M consolidated corporate group annual revenue; license notices delivered with pre-installed weights. |
| Kandinsky 6.0 Lite (3B) | 16 GB · 1080p | MIT License | All tiers (16 GB & 32 GB) Unrestricted commercial use. |
| Kandinsky 6.0 Pro (29B) | 32 GB · 1080p | MIT License | Recommended 32 GB (Profesional); 16 GB requires heavy CPU-RAM offload (~50 GB RAM footprint) Unrestricted commercial use. |
| Wan 2.2 5B | 16 GB · 720p | Apache 2.0 License | All tiers (16 GB & 32 GB, native 720p) Permissive Apache 2.0 commercial use. |
| Wan 2.2 14B (28.6 GB) | 32 GB · 720p | Apache 2.0 License | Exclusive to Profesional (32 GB VRAM) capped at 720p Permissive Apache 2.0 commercial use. |
Included Implementation Deliverables
Hardware integration & thermal/load burn-in
Validated assembly, sustained thermal/power burn-in, and 2×64 GB DIMM topology whenever 128 GB DDR5 is deployed.
16 h on-site LAN & ComfyUI workflow setup
Local studio network deployment with versioned JSON graphs (or client-designated orchestration tool).
8 h operator training
Practical production-team training on two-pass workflows (720p drafts + 1080p final pass).
30-day configuration support & OEM warranty per quote
30 days of workflow configuration support; manufacturer/OEM hardware warranty terms documented per formal sales quote.
Request a tailored technical diagnostic
Share your target monthly clip volume, output resolution, and studio network layout to size the optimal configuration.