The Rig

Week 1 · SLOP8760

Phase 1 of 4 · The Rig

This week adds

The rig itself: one budget you can name, one licence you have read, and one record that survives the frame.

The ramp

Weeks 1–2: The Rig — the budget, the graph, the log
Weeks 3–6: Four Generators — anime-first: image, then video
Weeks 7–9: Holding It Together — your own workflow, continuity, the board
Weeks 10–12: The Episode — one episode, produced and delivered

The budget has two shapes

A local rig is bounded by VRAM; an API rig by a credit balance. Wan2.2 publishes the floor for both of its sizes, and the floor decides which models exist for you at all.

Wan2.2 model	Single-GPU VRAM
TI2V-5B	at least 24GB
T2V-A14B, I2V-A14B	at least 80GB

Source: Wan2.2 README — https://github.com/Wan-Video/Wan2.2

A local rig is bounded by VRAM, not by price or processor speed — VRAM sets the largest model, the highest resolution, and the longest clip a single pass can produce. Running out of it doesn’t mean slower; past a point, it means the model does not run at all. An API rig is bounded by a credit balance instead: no ceiling on model size, but every call has a metered cost, and a workflow that loops without a limit burns through a semester’s budget in an afternoon.

The README’s own words: TI2V-5B “can run on a GPU with at least 24GB VRAM (e.g, RTX 4090 GPU)”, while the A14B text-to-video and image-to-video models want “at least 80GB VRAM”. The A14B pair is a mixture-of-experts design — “27B parameters but only 14B active parameters per step” — which is why the name and the memory floor disagree.

Render time is not shooting time

9minutes, at most

Wan2.2’s own figure for one five-second 720P clip out of TI2V-5B, on a single consumer-grade GPU. That is one take, before the takes you throw away.

Source: Wan2.2 README — https://github.com/Wan-Video/Wan2.2

A licence is three separate questions

Weights arrive with terms attached, and the terms are not interchangeable. One model’s licence answers all three at once.

Question	MiniMax H3’s answer
Revenue	separate authorisation above US$20 million yearly revenue
Territory	the EU, the UK, the Republic of Korea and the USA are excluded
Output	MiniMax claims no rights over the outputs you generate

Source: MiniMax H3 Community License Agreement — https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/LICENSE

Local model weights come from public repositories — one base release, then a long tail of community finetunes built on top of it — and every file carries a licence that is not interchangeable with any other: some permit commercial use outright, some are research-only or non-commercial, and a few carry a responsible-use clause restricting specific output regardless of what was paid for access. An API model publishes the same information differently, as terms of service rather than a licence file, but the obligation to read it before building a shot around it is identical.

The licence’s own definitions section names the Excluded Territories as the European Union, the United Kingdom, the Republic of Korea and the United States, and requires separate authorisation from any entity with “more than 20 million US dollars … in yearly revenue”. The model card adds that the local weights are “released under the MiniMax H3 Community License Agreement” rather than an open-source licence.

Licensor claims no rights in the Output You generate using the Model.
CreativeML Open RAIL-M, §6

Source: CreativeML Open RAIL-M licence text — https://huggingface.co/spaces/CompVis/stable-diffusion-license/raw/main/license.txt

The artefact is the record, not the frame

Keep whatever lets someone rerun the decision. A node graph exports to API format: UI metadata dropped, every node carrying a class_type, its inputs, and a link written as ["4", 0]. An API rig has no graph, but its request log does the same job.

Source: ComfyUI docs, Workflow API Format — https://docs.comfy.org/development/api-development/workflow-api-format

On a local rig, the thing worth keeping is the workflow graph: a node-based description of the operations that produced a frame, saved so the same recipe runs again with a new seed for a comparable result. On an API rig there’s no graph to save, but there is a request log — the exact parameters sent and the response received — and it does the same job. Either way, today’s frame is not the point. What’s worth keeping is whatever lets someone else rerun the decision that produced it.

The export is File → Export Workflow (API), and the docs are explicit that the API format “omits UI metadata”: no node positions, no colours, nothing about how the graph looked. The frame on the right is one still from this course’s reference episode; the record behind it is a file of exactly this shape.

Before class

Pick one rig and commit to it: install a local workflow engine and download one base model plus one small community finetune, or open an account with a generative video API and note its per-call cost. Either way, arrive with two comparable checkpoints or settings to test against each other.

This week’s exercise

Generate one five-second clip on default settings, then generate it again after changing exactly one parameter. Keep both graph files or both request logs, and log both render times and, if relevant, both costs. Bring both clips, their graphs or logs, and your timing notes to Wednesday’s Dailies.

Reading

Wan2.2
Wan-Video2025
can run on a GPU with at least 24GB VRAM (e.g, RTX 4090 GPU)
Hugging Face Hub: Licenses
Hugging Face
Remember to seek out and respect a project's license if you're considering using their code or data.
CreativeML Open RAIL-M licence text
CompVis2022
Licensor claims no rights in the Output You generate using the Model
Responsible AI Licenses
RAIL Initiative
RAIL empower developers to restrict the use of their AI technology in order to prevent irresponsible and harmful applications.
ComfyUI: Workflow API Format
Comfy Org
API format omits UI metadata (positions, colors, groups, node sizes) that is only needed for visual editing in the frontend.

The Rig#

Week 1 · SLOP8760#

This week adds#

The ramp#

The budget has two shapes#

Render time is not shooting time#

A licence is three separate questions#

The artefact is the record, not the frame#

Before class#

This week’s exercise#

Reading#