Creative exchange · registry v1

Workflows you plug into your AI.

Raw assets in, finished on-brand output out. Every workflow is a versioned package with the deterministic half separated from the judgment half — so it runs the same way whether a person, an agent, or a cron job calls it.

4×5 card 432 · footage 918
2
workflows
36
ads from 4 raw files
22.6s
wall clock
7/9
stages need no model
0
brand configs written

Every model worth using, one balance

14 models · 0 verified

Sora, Veo, Runway, Kling, Luma, Hailuo, Seedance and the rest behind a single key and a single balance — no per-vendor accounts, no separate invoices, no minimums. Prices below are retail for 5 seconds of video. Rates move monthly and are re-derived on every build, so a quote is never stale.

ModelVendor5s Credits
LTX-2Lightricks$0.20020indicative
Wan 2.2Alibaba$0.26727indicative
Kling 2.1 StandardKuaishou$0.33334indicative
HunyuanVideoTencent$0.33334indicative
Vidu Q1Shengshu$0.53354indicative
Seedance 1.0ByteDance$0.60060indicative
Pika 2.2Pika$0.60060indicative
Hailuo 02MiniMax$0.66767indicative
Gen-4Runway$0.80080indicative
Ray2Luma$1.200120indicative
Veo 3 FastGoogle$1.667167indicative
Kling 2.1 MasterKuaishou$1.867187indicative
Sora 2OpenAI$3.333334indicative
Veo 3Google$3.333334indicative

14 of 14 wholesale rates are unverified. They are indicative only and must be checked against each provider's live pricing before being quoted to a customer.

The catalogue

generated from veditx.json

Each listing renders straight from its package manifest, including what does not work yet. A registry that only advertises strengths produces buyers who churn and agents that pick wrong.

Ad Multiplier

ad-multiplier · v0.1.0 · commercial · managed-run
gated

Raw footage + ICP + brand kit -> hundreds of on-brand video ads across every ratio.

videoadsbrandingcampaignmarina
tenantingestselecttagcopycomposerenderpackagedeliver
Not finished
  • tag (agent-driven, not automated)
  • deliver (gallery not templated)
  • Layout is a single template derived from State Park Marina, including its hard two-line headline constraint. Multi-brand use needs a family of layouts.
  • State Park Marina's headline font is a metric-matched substitute; source the real face and set width_scale to 1.0.
  • Derived brand kits have empty voice rules — ingest cannot parse tone or banned phrases out of prose. Supply brand.json or let an agent fill them.

19 files · 45,506B · AES-256-GCM

Reframe Batch

reframe-batch · v0.1.0 · MIT · free
open source

Drop in any footage, get every social aspect ratio back. No branding, no copy.

videoaspect-ratioutility
reframe
Download3 files · 2,670B

Built to be called by agents

live endpoints

A marketplace humans browse is a website. One agents can query, price and invoke is infrastructure. Discovery is public, static and CORS-open — no key needed to find out what exists or what a run would cost. Try them; these hit the real API.

GET/api/workflows.json

Every workflow, with inputs, stages, cost model and known gaps.

GET/api/estimate?workflow=ad-multiplier&selects=100&angles=6

Credits a run would consume, before committing to it.

GET/api/workflows/ad-multiplier.json

One workflow in full — the manifest an agent reads before deciding.

GET/.well-known/veditx.json

Service discovery. Point an agent here and it finds everything else.

Or connect over MCP and the same four capabilities arrive as tools — search_workflows, describe_workflow, estimate_cost, run_workflow.

# any MCP client
python3 api/mcp_server.py

# or straight over HTTP — no key required for discovery
curl https://veditx.netlify.app/.well-known/veditx.json
curl "https://veditx.netlify.app/api/estimate?workflow=ad-multiplier&selects=100&angles=6"
   → 1,800 outputs · 2,650 credits
curl https://veditx.netlify.app/api/openapi.json  # full OpenAPI 3.1 spec

The card is fixed. The frame is not.

measured, not eyeballed

Every ratio carries a pixel-identical 1080×432 header card — verified at a mean channel difference of 0.02. The footage area absorbs the whole difference, so a taller canvas reveals more of the shot instead of squeezing it. That is the difference between a re-composition and a crop, and it is why these read as designed rather than resized.

1×1card 432 · footage 648
4×5card 432 · footage 918
9×16card 432 · footage 1488

Deterministic

ffmpeg and Pillow. Reproducible, resumable, cheap. Never asks a model anything.

ingestselectcomposerenderpackagedeliver

Judgment

Any model, or none. Paste the brief anywhere, call any OpenAI-compatible endpoint, or write the lines yourself.

tagcopy
The quality gate lives in the code, after the model speaks. Headlines are measured against the real font and rejected if they overflow, break to the wrong number of lines, or use a banned phrase. A weak model cannot ship off-brand output — it can only fail the check and be asked again, with the failure quantified in pixels.

Install

open · gated · hosted

Open workflows install straight into Claude Code. Gated ones decrypt in your browser — enter the passcode on the card above and you get an ordinary .zip. The passcode never leaves the page.

/plugin marketplace add veditx/exchange
/plugin install reframe-batch

# any workflow, once unlocked
unzip reframe-batch-0.1.0.zip && cd reframe-batch
python3 pipeline/reframe.py ./footage --out ./out --ratios 1x1,4x5,9x16