Runway Agent Brand Kits from a PDF vs Sume Format package files
Runway Enterprise Brand Kits can be built from PDF text. Sume keeps brand rules as files in a Format package, edited by the Contents API.

Runway's Brand Kits with Agent is an Enterprise feature: Agent applies a kit to its work and can build one from PDF text, and the page says there is no website URL import. Sume has no Brand Kit object; the equivalent is a Format whose package files hold the rules, which you edit by the Contents API. That is a different shape, and it is the only brand-persistence route in Sume's docs.
Where the brand lives
| Question | Runway Brand Kits with Agent | Sume Format package |
|---|---|---|
| Plan | Enterprise | Any account that can create a Format |
| Source | Applied kits; can build from PDF text | Files you write: instruction and reference files |
| From a website URL | Not supported | Pass the URL as run input if the Format reads it |
| Edit by API | Not described on the page I read | Contents API with an If-Match guard |
| Applied to | Agent work | Every run of that Format |
Editing the package safely
Read the package first, then write with If-Match set to the revision you read. If someone else changed it, the write is refused instead of overwriting, so you never clobber a teammate's edit. A multi-file change is one commit.
Keep the run body small
Hold the stable rules in the package and send only the per-run variables as input: product, offer, audience. Anything in input is data rather than instructions, so do not put brand rules there.
{
"instruction": "Make a 20-second ad for the product below.",
"input": {"product": "Trail mug", "offer": "Free shipping"},
"generation_spend_cap_usd": 15
}Checklist before you rely on it
- Write colors, tone and banned phrases as plain files in the package.
- Run once with a low cap and read the receipt before a batch.
- Change a rule in one place and every later run follows it.
Sources
Related posts
More in Comparisons
- Runway Agent custom preferences vs a Sume run instruction
Runway's custom generation preferences are guidance Agent weighs, not rules. On a Sume run the instruction is composed after the Format and wins on conflicts.
- Runway Agent long sessions vs Sume previous_run_id continuation
Runway says very long Agent sessions may degrade; start fresh per project. Sume continues a Format run with previous_run_id; Agent Completions cannot.
- Runway Agent long videos in Final Cut vs a Sume run deadline
Runway Agent has no hard length limit and joins clips in Final Cut. A Sume Format run has a 90-minute deadline. How to plan a long video on each.
- Runway Agent message credits: Opus 5.5, GPT-6 Astra, and Sume
Runway charges 24 credits per message on Opus 5.5 and 40 on GPT-6 Astra. Sume lets a run pick its orchestrator with one model field. The math and the gap.
Written by Sume