How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

intel-arc-pro-b70-videogen-cookbook - Skill

Name: intel-arc-pro-b70-videogen-cookbook
Author: SergiioB

Imported from SergiioB/intel-arc-pro-b70-videogen-cookbook (AGENTS.md). Install upstream with npx skills add SergiioB/intel-arc-pro-b70-videogen-cookbook. Copyright stays with the author.

AGENTS.md -- Intel Arc Pro B70 Video Cookbook

For coding agents working on this repo. Last updated: 2026-08-26.

1. Purpose

Portable recipes to run MiniMax H3 T2V on one or two Intel Arc Pro B70 32 GB cards via ComfyUI + PyTorch XPU.

2. Stack

ComfyUI 0.33.0 + PyTorch 2.13.0+xpu. T2V only.
Dual split: CLIP xpu:1, UNet+VAE xpu:0. Not tensor parallel.
OmniXPU ESIMD SDP + oneDNN FP8 GEMM (optional, SDPA is the default).
Metered queues via scripts/queue_metered.py. Power via xe hwmon energy1_input.
Concat + lanczos upscale via ffmpeg.

3. Hardware assumptions

Two Intel Arc Pro B70 32 GB (Xe2 BMG). No GPU P2P.
One B70 works: CLIP on CPU, worse wall time.
scripts/h3 detect and h3 start auto-select dual vs single. Do not hand-edit affinity for dual.
Ubuntu x86_64, xe driver.
Discover xe hwmon by name, not by index. Never hard-code hwmon4 or PCI addresses.

4. Rules for scripts and docs

All paths must use $COMFY_DIR. Never hard-code host-specific paths.
$H3_PYTHON overrides the Python binary. Default: $COMFY_DIR/.venv/bin/python.
Do not source setvars.sh. Do not put oneAPI compiler/lib on LD_LIBRARY_PATH.
Do not use --highvram. Do not set ZE_AFFINITY_MASK in dual mode.
Campaign cap: 180 W. Idle cap: 150 W via h3 stop. Discover hwmon, never hard-code.
H3_SKIP_UNLOAD=1 between acts in multi-clip queues.
Speed numbers from single observations must be labeled as such.

5. Content rules

Main example story: Paper Star. Cast: Hana (23), Ren (20), two young adults. Season 1 is ten episodes plus a Season 2 teaser.
Additional cartoon examples (fox, cookie heist, moon lantern) live under examples/ with their own jobs and a gallery in docs/GALLERY.md.
Cartoon / anime is the default example style. People LoRA at 0 for that look.
Lock wardrobe in every act prompt.
MiniMax prompt IR: integrated_multimodal_description, timed Shot 2, overall_soundscape, non_diegetic_music.
I2V / Ref2VA FL2VA do not work on this XPU. T2V only.

Install intel-arc-pro-b70-videogen-cookbook and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "sergiiob-intel-arc-pro-b70-videogen-cookbook-intel-arc-p-6e4191", "kind": "skill", "name": "intel-arc-pro-b70-videogen-cookbook", "description": "For coding agents working on this repo. Last updated: 2026-08-26.", "publisher": "SergiioB", "version": "1.0.0", "capabilities": { "domains": [ "coding", "creative" ], "tags": [ "skill-md", "github-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "For coding agents working on this repo. Last updated: 2026-08-26." ], "primary": false, "metadata": { "source": { "provider": "github-agents", "repository": "https://github.com/SergiioB/intel-arc-pro-b70-videogen-cookbook", "path": "AGENTS.md", "ref": "15089da5b5245f6317c5d6fc694048fb1d164d91", "url": "https://github.com/SergiioB/intel-arc-pro-b70-videogen-cookbook/blob/15089da5b5245f6317c5d6fc694048fb1d164d91/AGENTS.md", "key": "SergiioB/intel-arc-pro-b70-videogen-cookbook/AGENTS.md" } }, "instructions": "# AGENTS.md -- Intel Arc Pro B70 Video Cookbook\n\nFor coding agents working on this repo. Last updated: 2026-08-26.\n\n## 1. Purpose\n\nPortable recipes to run MiniMax H3 T2V on one or two Intel Arc Pro B70 32 GB cards via ComfyUI + PyTorch XPU.\n\n## 2. Stack\n\n- ComfyUI 0.33.0 + PyTorch 2.13.0+xpu. T2V only.\n- Dual split: CLIP `xpu:1`, UNet+VAE `xpu:0`. Not tensor parallel.\n- OmniXPU ESIMD SDP + oneDNN FP8 GEMM (optional, SDPA is the default).\n- Metered queues via `scripts/queue_metered.py`. Power via `xe` hwmon `energy1_input`.\n- Concat + lanczos upscale via ffmpeg.\n\n## 3. Hardware assumptions\n\n- T", "cost": { "context_tokens": 513 } }

intel-arc-pro-b70-videogen-cookbook

About

AGENTS.md -- Intel Arc Pro B70 Video Cookbook

1. Purpose

2. Stack

3. Hardware assumptions

4. Rules for scripts and docs

5. Content rules

Install intel-arc-pro-b70-videogen-cookbook and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews