Iverywhere

Figma Weave — capability review & AI media strategy

What Figma Weave is, exactly what Claude can and can't do with it from the Iverywhere terminal, and a dated plan for turning the July AI Asset Playbook into a repeatable, one-subscription image + video engine.

Point-in-time review · Created August 24, 2026 · Facts verified against Figma's own pages that day · Companion to AI_ASSET_PLAYBOOK.html
The short version

Five things to know

Read this block and you have the decision. Everything below is the evidence.

What it is

Figma Weave, verified facts

Every row below was read from a Figma or Figma Weave page on Aug 24, 2026 — source in the last column. Rows marked ◐ are inferred, not stated.

FactDetailSource
What it isNode-based AI media platform: "Access image, video, audio, and 3D models from Google, Kling, Runway, Luma, and more without managing multiple subscriptions." Editing tools native to the canvas: layers, masking, color grading, compositing, inpainting, relighting; plus Invert, Outpaint, Crop, Mask Extractor, Upscale, Z-Depth, Image Describer, Channels, Painter.figma.com/solutions/…weave; weave.figma.com
Ownership & timelineWeavy (Tel Aviv, founded 2024) acquired by Figma and rebranded Figma Weave — Oct 30, 2025. Config 2026 (Jun 24, 2026): Weave tools inside Figma Design (open beta, free during beta), publish Weave workflows to Figma Community, "Figma node" in Weave coming; full embedding "later this year".welcome-weavy blog; connecting blog; Weave KC: Config 2026
Providers on the platformGoogle, Kling, OpenAI, ByteDance, Black Forest Labs (Flux), Runway, Luma, LTX, Wan, Grok, Recraft, Bria. Figma's own blog canvas shows a "Gemini 3 (Nano Banana Pro)" node ◐.weave.figma.com; connecting blog
Pricing (standalone Weave)Free $0 · 150 credits/mo (≈375 images) · 5 workflows · no top-ups.
Starter $24/mo ($228/yr) · 1,500 credits (≈3,750 images or ≈417 s video) · unlimited workflows · top-up $10/1,000.
Professional $45/mo ($432/yr) · 4,000 credits (≈10,000 images or ≈1,111 s video) · 3-month rollover · top-up $10/1,200.
Team $60/user/mo · 4,500 credits/user · shared pool, admin tools.
Commercial license on every plan, including Free. Topped-up credits roll over 12 months.
weave.figma.com/pricing
Weave vs. Figma billing"Figma and Figma Weave operate as separate products with independent credit systems." A paid Figma plan does not include Weave. Figma says pricing will evolve "over the next year."Weave KC: FAQ
Weave tools inside Figma DesignCurated tools (Generate mockup, Transfer style, Replace background, Change lighting, Text to vector, Apply color palette, product-shoot generators…). Requires Figma Professional/Organization/Enterprise with a Full seat and can-edit access; not on the free Figma Starter plan. Open beta, free now; "once generally available, AI credit usage will apply." Image tools only — no video inside Figma Design.Figma Learn: Use Weave tools; Config 2026
Programmatic accessNo public API on any plan. The only automation path is the Figma MCP server's six weave_* tools. "Creating and editing Weave workflows via MCP is not yet supported." Requires a Weave account on Starter/Pro/Team/Enterprise, and linking your Figma account inside Weave (Settings → Profile → Linked accounts). Only tools from the active Weave workspace are visible. Runs spend Weave credits; the agent must show the cost and get a yes before each paid run.Weave KC: MCP; Figma MCP developer docs
Rights & data"Verified by Figma" models: contract with the provider, your content is used only to deliver the service, not for training, and the contract includes indemnity. Unverified models run under the provider's own terms (still usable; badge shown on hover). SOC 2 Type II. Your content is not shared between Figma and Weave.Weave KC: Verified models

How the pieces connect

The Weave path runs beside the existing pipeline, not through Figma Design. Figma Design is optional and not part of the plan.

Your phones + LightroomLane 1 capture, house grade applied. 18 graded stills + 5 atmosphere frames + 7 hero clips already in assets/art-direction/
→
Claude (this terminal)Picks the input, uploads it, runs the tool, polls, downloads, runs garment QC, files + logs the asset
⇄
Figma MCP servermcp.figma.com · one OAuth login · weave_list_tools · weave_get_tool_inputs · weave_upload_asset · weave_run_tool · weave_get_tool_run_output · weave_cancel_tool_run
⇄
Figma Weave (paid)Your four published tools → Nano Banana / Flux 2 / Veo 3.1 / Kling nodes → QC nodes (Compare, Image Describer)
→
broll/ai bank + capture logGraded, logged, ready for the content calendar

What this means for you: the human work is building four graphs once (a few hours total, starting from Community templates). After that, "make me the Como plate at 4:5 and 9:16" or "animate the packing-table still" is a sentence to me, a cost-confirmation from you, and a logged asset in the bank.

Capabilities

What Claude can and can't do — honest inventory

Checked against the installed Figma plugin (v2.2.96), Figma's developer docs, the Adobe connector installed here, and this session's actual connection state.

CapabilityStatusDetail
Run your published Weave tools (upload → run → fetch result)Can, after loginSix weave_* tools on the Figma MCP server. Needs: paid Weave plan, Figma↔Weave accounts linked, and you completing one OAuth login when I ask. Each credit-spending run pauses for your confirmation — Figma's design, not mine — so we batch runs into sessions where you're present.
Build or edit a Weave workflowCannot"Not yet supported" via MCP. You build graphs in the browser; I can write the node-by-node recipe and the prompt blocks for you to paste.
Generate images or video nativelyCannotI have no image/video model of my own. I orchestrate, prompt, QC, file, and log.
Visual QC of outputsCanI can view any image and judge palette/light/texture against the art-direction doc. For the garment QC I can go further than eyeballing: a short Python script overlays the source garment on the composite, measures colour drift per channel and flags edge mismatch — a numeric version of the playbook's 200% check.
Figma Design files (read/write)Can, after loginSame login unlocks 24 Figma tools — use_figma (scripted edits), get_screenshot, upload_assets, download_assets, create_new_file… Useful later for laying out email/journal templates; not needed for the media pipeline.
Weave tools inside Figma Design (Transfer style, Replace background…)Not for us nowRequires Figma Professional with a Full seat (a second subscription) and is image-only, UI-driven — no MCP path for it. Revisit only if you start designing in Figma anyway.
Adobe: batch house grade, generative expand, resize to ratios, social variations, Enhance Speech, quick-cut reelsCan, after loginThe Adobe for creativity connector is installed (skills: batch-edit, resize, social-variations, quick-cut, retouch). Also not signed in this session. This remains the Lane 2 engine — Adobe-stock-trained expand is still the commercially safest fill.
Files, naming, capture log, calendarCanRename to iv-ai-<workflow>-<scene>-<nn>, drop into broll/ai/, write the lane/tool/used-in-post row, and update the .ics.
Verify your Figma plan / Weave accountCouldn'tweave.figma.com showed a signed-out page in Chrome; figma.com is blocked for my browser extension. See "Unverified" at the end.
Fit

Where Weave slots into the four lanes

The playbook's lanes decide tool, rules and disclosure. Weave changes the tool column only. Credit prices are per run, from Weave's model-comparison pages (updated Mar–Apr 2026; Weave says in-app prices are authoritative).

LanePlaybook tool todayIn WeaveVerdict
1 · RealCamera + Lightroom house preset—No change. Weave never touches Lane 1.
2 · Assisted
extend to ratio, clean-up
Photoshop generative expand (Firefly)Flux Pro Outpaint (6 cr), Bria Content-Aware Fill (5), Bria Remove Background (0.6), Relight 2.0 (10)Keep Photoshop as default (safest training data, layer control). Use Weave for volume batches of ratio variants.
3 · Composite
real garment, generated world
Nano Banana Pro in GeminiGemini 3 Pro edit node (15–30 cr) ◐ = Nano Banana Pro; Gemini 3.1 Flash "Nano Banana 2" (4–18); Seedream V5 Edit (4); Flux Kontext Multi-Image (10)Move here. Same model, but now with the reference frame, the locked-garment prompt and a Compare node baked into one tool.
4 · Atmosphere
no people, no product
Midjourney v7 (optional) / Nano Banana ProMidjourney is not in Weave (it has no API). Replacements: Flux 2 Pro (5), Imagen 4 (6), Recraft V3 (5), Luma Photon (2), Flux 2 Flex (14)Move here and drop the Midjourney line item. Flux 2 Pro with a style-reference image gets the filmic plate.
Motion
still → 6–8 s quiet clip
Veo 3.1 via Gemini/Flow; Kling 3.0 backup; Runway Gen-4.5 for motion-brushVeo 3.1 Image-to-Video, 8 s: Fast 120 / Standard 320. Kling 3: Std 3 s 60 → 15 s 300. Kling 2.5 Turbo Pro 5 s 35 / 10 s 70. Runway Gen-4.5 5 s 70 / 8 s 112; Gen-4 Turbo 5 s 30. Sora 2 present (96/288) — the playbook says don't build on it.Move here. Draft on cheap Kling/Runway, render the keeper on Veo 3.1 Fast. All image-to-video, never text-to-video.
AudioIn-app IG audio; ElevenLabs for owned surfacesWeave lists audio models; not assessed hereKeep ElevenLabs — the playbook chose it for licence clarity, and Weave's audio nodes weren't in the pages I could verify.
NeverVirtual try-on / AI models — declined"Kolors Virtual Try On" exists (8 cr)Still declined. Absolute #2 (no synthetic people as real) is the compliance moat.

Why this is worth doing

One bill replaces Google AI Pro ($20) + Midjourney ($10–30): Starter is $24.

The playbook's "prompt library system" becomes literal — fixed blocks live in nodes, one variable changes per run.

QC moves inside the graph (Compare node, Image Describer as a checker) and I run the rest.

Commercial licence on every plan; verified models carry indemnity.

What it doesn't fix

Weave credits are pricier per Veo clip than a direct Google sub if video volume gets high — hence draft-cheap/render-once.

Platform AI labels (SynthID from Nano Banana, C2PA) still apply. The capture log stays non-negotiable.

Pricing and packaging are explicitly in flux "over the next year". Don't prepay annually.

Every paid run asks you for a yes. Fine in a batch session, annoying as a background job.

The build

Four Iverywhere tools — one per playbook workflow

A "Weave tool" is a workflow published with a simplified input form. Build order follows risk: cheapest and safest first. Each has a stated builder (you or Haley, in the browser) and a stated operator (me, over MCP).

1 · iv-atmosphere — Lane 4 plate, 4:5 + 9:16 pair

Inputs
City recipe (dropdown: Copenhagen / Amsterdam / Watson's Bay / Lisbon / rainy morning / still-life / the six b-roll skeletons) · one canonical graded reference frame · ratio pair.
Graph
Text node (recipe) + Text node (the grade sentence, fixed) + Text node (negative block, fixed) → LLM node assembles the five-block prompt → Flux 2 Pro ×2 (4:5, 9:16, same seed) with style reference → Image Describer node checks "no people, no product, no text" → outputs. ≈10–12 credits per pair.
Operator
I run a batch of 4 pairs per scene, download, apply the Sézane filter, hand you the shortlist; you grade the keeper in Lightroom; I file + log Lane 4.

2 · iv-motion — still → quiet motion, 8 s, 9:16

Inputs
An approved graded still (real or AI) · motion block (dropdown of the playbook's four starter prompts, or the "camera-only" block when product is in frame) · quality (draft / final).
Graph
Image node (first frame) + Text node (motion prompt with the locked "nothing deforms" clause) → draft: Kling 2.5 Turbo Pro 5 s (35 cr) · final: Veo 3.1 Image-to-Video 8 s Fast (120 cr) → mute audio → output. Product-in-frame rule enforced by the prompt block, not the model.
Operator
I run 3 drafts, you pick, I run one final. I watch every frame at full size for warps before it's banked.

3 · iv-composite — real garment, generated world (Lane 3)

Inputs
Clean graded product photo (A6/P2/P5 frames) · scene pick (trattoria chair / teak bench by canal / villa window) · one reference frame for the look.
Graph
Image (garment) + Image (reference) + Text (the playbook's "keep the garment EXACTLY as photographed" template, fixed; scene is the only variable) → Gemini 3 Pro edit node (Nano Banana Pro) ×4 → Compare node (source vs output, side-by-side) → outputs. ≈60–120 credits per batch of 4.
Operator
I run the batch, then the numeric garment QC (overlay, colour drift, edge registration) and kill any candidate that touched the garment. You re-grade the keeper lightly. Logged Lane 3; captioned as still-life, never trip.

4 · iv-extend — one hero → every ratio (Lane 2, volume only)

Inputs
Graded hero · target ratios (1:1, 4:5, 9:16, 16:9, 2:3) · which side may grow (sky / water / wall).
Graph
Image → Outpaint node (Flux Pro Outpaint, 6 cr) per ratio, empty prompt first → Compare → outputs. The 40%-new-canvas rule stays: if it would need to invent architecture, people or product, it's Lane 3 or a re-crop.
Operator
Default remains Photoshop via the Adobe connector; this tool is for weeks with many heroes. I zoom the seams at 200% before banking.

Rules that ride along in every graph, verbatim: (1) never generate the product; (2) never generate people presented as real; (3) AI plays atmosphere, not autobiography; (4) never animate the garment. Prefer "Verified by Figma" models for anything published — hover the node to see the badge; Google/Flux/Kling nodes are the ones to check first. Grade after generation, grain last, log the lane.

Money

Credit budget at the planned cadence

Content plan cadence: 3–4 Instagram posts/week, 2–3 Story touches on posting days, product ≤1 in 4. Assumes AI supplies roughly two assets a week. Per-run prices from Weave's comparison pages; treat as ±30% until the first real month.

Monthly workRunsModel · credits/runCredits
Atmosphere plates4 scenes × 8 candidatesFlux 2 Pro · 5160
Composites4 scenes × 4 candidatesNano Banana 2 · ~10 (4–18)160
Ratio extends4 heroes × 3 ratiosFlux Pro Outpaint · 672
Motion drafts4 clips × 3 draftsKling 2.5 Turbo Pro 5 s · 35420
Motion finals4 clips × 1Veo 3.1 I2V 8 s Fast · 120480
Total≈1,290

Reading the number

  • Free (150): one Veo clip or ~30 plates. Evaluation only — and no MCP.
  • Starter (1,500): covers the plan at ~86% with a $10 top-up as the safety valve. The right plan for Sept–Oct.
  • Pro (4,000, 3-month rollover): for a holiday month where motion doubles, or if Veo Standard (320) becomes the norm. Decide Nov 1 from October's real burn.

Versus the July stack

  • Playbook: Google AI Pro ~$20 + optional Midjourney $10–30 = $20–50/mo.
  • Weave Starter $24 replaces both and adds orchestration + MCP. Roughly cost-neutral; strictly better on repeatability.
  • Adobe stays (already owned). ElevenLabs stays if/when the house music bed is needed.
The plan

Dated phases — nothing before launch week ends

Launch is Sept 1; the marketing plan's "quiet last word" is Tue Sept 8. Every phase below is on the Content sub-calendar.

Aug 24 → Sep 8Phase 0 · lockout

Change nothing. Launch week runs on the banked library and the Adobe lane. The only optional 10-minute task: create the free Weave account with contact@iverywhere.com so the trial starts cold on the 9th.

Sep 9 → Sep 22Phase 1 · free trial · $0

You drive, I QC. Run three Community Weave workflows (a product-shoot template, a style-transfer template, an image-to-video template) against five graded frames from assets/art-direction/candidates/graded/. Send me the outputs; I score them on palette · light · texture · motion · the Sézane test and the garment QC. Also: link your Figma account in Weave (Settings → Profile → Linked accounts) — needs a Figma account; a free one is fine.

Go/no-go Sep 22: do ≥2 of 3 templates produce a frame that could sit in the feed after the house grade? If not, the playbook's current stack stands and this doc closes.

Sep 23 → Oct 13Phase 2 · Starter · build

Upgrade to Starter ($24, monthly — not annual). Build the four tools in the order above, roughly one per five days, each starting from the closest Community template. As each is published, in a Claude Code session say "connect Figma" — I complete the login, run whoami and weave_list_tools, then prove the tool end-to-end on the playbook's first-batch assets (#1 extend, #2 vest-on-chair composite, #3 Como plate, #4 that plate animated, #5 packing-table living photo).

Oct 14 → Nov 1Phase 3 · production month 1

Bank the holiday season. The b-roll doc's rule — "October banks holiday" — becomes the first real workload: candlelight, kraft + Brunswick ribbon, garlands as atmosphere plates and quiet-motion loops; composites of the vest as the family gift. I track credits per asset in the capture log.

Nov 1 decision: stay on Starter, or move to Pro for the Q4 volume the marketing plan expects (Q4 ≈ a third of annual revenue for gift categories).

From Nov 2Phase 4 · the weekly loop

Sat capture session (b-roll lists) → Sun grade in Lightroom → Mon I run the tools in one batch session with you present for cost confirmations → Tue QC, bank, log → the week's calendar posts assemble from the bank. Add the winning prompt blocks back into the playbook's library with the asset they produced.

Before we start

What I need from you

Watch-list

Risks and what would change the plan

RiskWhy it mattersWhat we do
Pricing/packaging changesFigma says Weave's fit within Figma tiers is still being decided; in-Figma tools will start consuming credits at GA.Monthly billing only. Re-check pricing at the Nov 1 gate.
Model rights driftUnverified models run under the provider's terms; Sora is present but the playbook flagged its discontinuation.Publish only from Verified-badge nodes; never build on Sora.
Auto-labellingMeta auto-detects SynthID/C2PA regardless of tool. A surprise "AI" label with no context reads worse than a quiet one.Capture-log row on every asset; Lane 3–4 captions never claim the trip.
MCP linkage doesn't expose toolsTools come only from the active Weave workspace; wrong workspace = empty list.First session runs whoami + weave_list_tools before anything else.
Founder hoursBuilding four graphs is ~4–8 hours of browser time in the first post-launch month.Start from Community templates; I write the recipes; one tool per five days, not all at once.
Honesty

Unverified / assumptions

• Your Figma account/plan and Weave account state — weave.figma.com was signed-out in Chrome and figma.com is blocked for my browser tool.

• Per-model credit prices come from Weave help pages last updated Mar 29–Apr 26, 2026, which say the in-app prices are authoritative. Budget math is ±30% until month one.

• "Gemini 3 Pro edit node = Nano Banana Pro" is inferred from Figma's Jun 24 blog screenshot label "Gemini 3 (Nano Banana Pro)" and the comparison table; confirm on the node's tooltip.

• Which models carry the "Verified by Figma" badge — the list page needs a login; check in-app before first publish.

• The Sora discontinuation claim is carried over from the July playbook, not re-verified today.

• Whether the login here actually surfaces the weave_* tools depends on the account linkage — proven only by running weave_list_tools.

Appendix

Sources (all read Aug 24, 2026)