QuiverAI's terms do permit commercial sale on a paid tier, with no attribution owed - the exposure is non-exclusivity and a one-way indemnity
The question
"Does QuiverAI's terms of service actually permit selling the AI-generated coloring-book art commercially, and what license/attribution limits apply?"
[[01-projects/printables-product/index|Scribble Works]] and the religious Christmas coloring book both route their art through QuiverAI (Arrow 2 / Arrow 2 Telos via the MCP server). Printable PDFs sell from the site and a Lulu print-on-demand paperback is in the Dec plan, so the commercial-rights answer gates everything downstream of art production. Copyrightability of the output (can RDCO own it) is a separate queue item and is deliberately out of scope here.
What we already know (from the vault)
- The Oct-07 founder decision log in [[2026-10-06-shopify-launch-plan]] already recorded a preliminary read: "paid Quiver plan confirmed; Quiver ToS 2026-03-25 allows commercial use on paid tiers, no exclusivity, we indemnify; add an 'illustrated with AI, composed and reviewed by hand' listing line; check the 'Improve QuiverAI' toggle." This brief confirms every element of that read at the source and adds three things the note did not carry.
- The same plan already flags the separate copyrightability risk in its own words: "Purely AI-generated images can't be copyrighted in the US, so others could copy them. That's a known, accepted risk for a learning project." Out of scope here, noted so the boundary is explicit.
- RDCO has shipped commercial-adjacent QuiverAI output before. The Denominator sticker set (eight Arrow 2 SVGs) and the Denominator art spec both treat Quiver output as shippable brand assets, so this is a pattern decision, not a one-off.
- RDCO has run this exact drill twice before on other vendors: [[2026-07-21-akta-monid-tos-data-output-rights]] and [[2026-09-13-anthropic-aup-agentic-use-operative-clauses]]. The house method is to quote the operative clause rather than paraphrase a vendor's marketing claim.
What the web says
All of the following was retrieved 2026-10-14 from QuiverAI's own pages. WebFetch was not needed; curl with a browser user-agent returned the pages cleanly.
- Ownership is assigned to the user, explicitly. ToS §8.2 (https://quiver.ai/legal/terms, last updated March 25, 2026): "QuiverAI does not claim ownership of your User Content and QuiverAI disclaims ownership of your Generated User Content. QuiverAI hereby assigns to you all of its right, title, and interest, if any, in your Generated User Content. This assignment does not extend to other users' Generated User Content, such as that which may be similar to your Generated User Content."
- Commercial use is tier-gated, and only the free tier is restricted. ToS §5.2 Free Tier Restriction: "If you are subscribed to our free tier, you may use the Service and any content Generated (defined below) by the Service solely for your personal, non-commercial, non-revenue generating purposes only. This restriction does not apply to your use of the Service under any of our paid tiers." Paid plans on https://quiver.ai/pricing (retrieved 2026-10-14) are Go $8/mo, Plus $20/mo, Pro $40/mo, Enterprise custom, with a "14-day Go trial (card required)". The pricing page currently advertises no free plan at all; the docs plan table at https://docs.quiver.ai/app/pricing lists only Plus and Pro and appears stale against the marketing page.
- No attribution is required anywhere. A full-text scan of the ToS for "attribut", "credit", "watermark" and "powered by" returns only two hits, neither of which is an output-attribution requirement: §5.4 says QuiverAI owes you no attribution for Feedback, and §8.4 requires that your content not "require us to obtain any further licenses from or pay any royalties, fees, compensation or other amounts or provide any attribution to any third parties." There is no credit line, no watermark obligation, and no "made with QuiverAI" requirement on the artwork, the product, or a credits page.
- No prohibition bites a coloring book, with one disclosure-shaped exception. The Acceptable Use Policy is embedded as ToS §19. Its categories are illegal activity, fraud/abuse, child safety, impersonation, and violent/hateful content. Nothing forbids resale of raw output, print-on-demand, or redistributing art as a standalone asset. The child-safety clause bars sexualizing or exploiting minors and "age-inappropriate content targeting minors", which a coloring book comfortably clears. The clause that does touch the product is under Impersonation and Deceptive Practices: users may not "mislead others about whether content was generated by AI."
- QuiverAI keeps a very broad license back over everything, including generated art. §8.1 defines "User Content" to include all content "Uploaded, Generated, or Posted", so §8.3 reaches the coloring pages: "You grant QuiverAI a worldwide, non-exclusive, irrevocable, perpetual, royalty-free, fully paid right and license (with the right to sublicense through multiple tiers) to host, store, transfer, publicly display, publicly perform ..., reproduce, modify ..., create derivative works of, distribute, and otherwise process your User Content." The Privacy Policy (https://quiver.ai/legal/privacy, last updated February 25, 2026) says the quiet part plainly: "Images generated from your Inputs may be viewable by other users on our Services or shared publicly."
- Training on your prompts and output is on by default and is a self-serve toggle. https://quiver.ai/security and https://docs.quiver.ai/developers/platform/data-controls (both retrieved 2026-10-14): "Model training: Whether QuiverAI may train its models on inference inputs and outputs. On by default for every organization. An Owner or Admin turns it off with Improve QuiverAI in Data Controls." Turning it off stops training but not storage; no-storage is contract-only, not self-serve.
- Identical output for someone else is contemplated, not disclaimed. ToS §15: "THE SERVICE MAY GENERATE CONTENT FOR OTHER USERS THAT IS SIMILAR OR IDENTICAL TO THE GENERATED USER CONTENT, AND YOU AGREE THAT SUCH CONTENT GENERATED FOR OTHER USERS IS NOT GENERATED USER CONTENT."
- The indemnity runs one way only. ToS §14 obliges the user to "indemnify, defend ..., and hold harmless QuiverAI" against third-party claims arising from "your User Content, including Uploading, Posting, and Generating any User Content through the Service" and from "your violation of any third-party right, including any intellectual property right". There is no reciprocal IP indemnity to the customer anywhere in the terms. QuiverAI, Inc. is a Montréal-domiciled company (7186 Blvd. Saint Laurent, Montréal, QC) and §18.9 states "The Service is intended for visitors located within the United States."
Convergences and contradictions
- Convergence, and it is clean. The founder's Oct-07 read matches the source on all four points: paid-tier commercial permission, the 2026-03-25 ToS date, non-exclusivity, and user-side indemnity. Nothing in the vault needs correcting. The brief adds three things that note did not carry: that no attribution is required at all, that §8.3's license-back explicitly reaches generated art (not just uploads), and that the Privacy Policy independently reserves public display of your generated images.
- Contradiction inside QuiverAI's own surfaces. ToS §5.2 is drafted around "our free tier", but the live pricing page advertises no free tier, only a card-required 14-day Go trial. Whether a trial counts as "subscribed to a paid tier" is genuinely ambiguous and the terms do not resolve it. Practically this is moot if the account is on a billed plan, but it is the reason the tier should be verified rather than assumed.
- The upstream-model layer cannot be verified, and that is itself the finding. QuiverAI names no foundation-model provider and publishes no subprocessor list anywhere in its terms, privacy policy, security page, or developer docs. The models are presented as first-party ("Our most capable model", API ids
arrow-2andarrow-2-telos, token-priced at $4/$20 and $6/$30 per 1M), andmcp__quiverai__list_modelsreturns only those two. The single public hint of an upstream component is marketing copy in the Sept 7 2026 launch post: Arrow 2 Telos "combines Arrow's speed with the refinement capabilities of frontier models." Telos's 1,050,000-token context window is circumstantial evidence of a frontier base model, but that is inference, not a source. Two searches found only third-party aggregators repeating QuiverAI's own copy. No upstream provider's terms can therefore be checked for conflict, and none can be shown to impose one.
Synthesis for RDCO
The permission question is answered and the answer is yes. On any paid tier, QuiverAI's terms do more than tolerate commercial sale: §8.2 affirmatively assigns to the user whatever right, title and interest QuiverAI holds in the generated art, and §5.2 confines the non-commercial restriction to the free tier in one sentence with an explicit carve-out. No attribution, credit line, or watermark is owed in any form or any place. Nothing in the Acceptable Use Policy prohibits resale, print-on-demand, or shipping the art as a standalone asset. For the narrow question "may we sell printable PDFs and a Lulu paperback illustrated with Arrow 2 output", the terms say yes with unusual clarity for a tool at this stage.
What the terms do not give is exclusivity or cover. Three clauses stack into one operational posture. §8.2's assignment stops at the user's own output and explicitly does not reach "other users' Generated User Content, such as that which may be similar to your Generated User Content." §15 then says in capitals that the Service may generate identical content for other users. And §8.3 hands QuiverAI a perpetual, irrevocable, sublicensable right to publicly display and distribute the coloring pages, with the Privacy Policy confirming generated images "may be viewable by other users on our Services or shared publicly." The practical read: RDCO owns a non-exclusive asset that the vendor may also show the world and, with "Improve QuiverAI" on by default, may train its next model on. None of that blocks the sale. It does mean the moat is the book, not the line art - the composition work, the page sequencing, the physical product, the audience. That is already the plan's premise, and this is independent confirmation of it rather than a change of direction.
The exposure that deserves a real decision is §14. The indemnity runs entirely one way: RDCO defends QuiverAI against third-party IP claims arising from art RDCO generated, and QuiverAI owes RDCO nothing in return. Adobe Firefly, OpenAI and Google all offer some form of customer-facing IP indemnity for enterprise generative output; QuiverAI offers none, holds no SOC 2 or ISO 27001 ("We do not hold third-party certifications today"), and will not name the models behind Arrow. For a $7.99 PDF that risk is academic. For a physical print run with a vendor relationship, an ISBN, and a brand name on the spine, it is the clause an attorney would want to see, and the one that argues for keeping the first run small.
Two mechanical items close the gap between "the terms allow it" and "we are clear to ship." First, confirm at the source which plan the account is actually on, because §5.2 is a binary switch between permitted and prohibited and account state is not a terms question. The founder stated on Oct 07 that the paid plan is confirmed; that statement has not been checked against the Billing page, and a screenshot dated to the week art ships is a thirty-second paper trail. Second, turn off "Improve QuiverAI" in Data Controls before the production run, since training is on by default for every new organization and the setting is self-serve for an Owner or Admin. Separately, the AUP's "mislead others about whether content was generated by AI" and KDP's own AI-content disclosure both point the same way as the listing line already drafted in the launch plan, so that line should be treated as a compliance item rather than a marketing nicety.
BOTTOM LINE: AMBER. Ship, with three named mitigations: (1) verify and screenshot the paid plan in QuiverAI Billing on the week art ships, (2) switch "Improve QuiverAI" off in Data Controls before the production run, (3) carry the "illustrated with AI, composed and reviewed by hand" disclosure on every listing. The single most load-bearing clause is ToS §5.2: "If you are subscribed to our free tier, you may use the Service and any content Generated ... solely for your personal, non-commercial, non-revenue generating purposes only. This restriction does not apply to your use of the Service under any of our paid tiers." Everything turns on an account-state fact, not a legal one, and it is cheap to verify. The reason this is AMBER and not GREEN is §14: with no reciprocal IP indemnity and no disclosed model provenance, RDCO carries 100% of any third-party infringement claim, which is worth a real attorney's eyes before a large print run even though it does not block a first one.
This brief is a reading of published terms, not legal advice. A licensed attorney should review §8.3, §14 and §15 before any print run large enough that a claim would matter.
Why this is in the vault
It closes decision 4 ("Rights check: confirm QuiverAI's terms allow selling the art") in [[2026-10-06-shopify-launch-plan]] with sourced clause text rather than a recollection, and it sets the two pre-production gates - paid-tier screenshot and the Improve QuiverAI toggle - that the Dec coloring-book launch and any Scribble Works paid printable must clear before art ships.
Open follow-ups
- Does QuiverAI's Enterprise contract offer a customer-facing IP indemnity, a no-training commitment, or a waiver of the §8.3 public-display right? The security page says contracts can prohibit training and storage, which implies the standard terms are negotiable at that tier; what an Enterprise agreement actually covers is not published.
- Does Lulu's or Shopify's own content policy impose an AI-generated-art disclosure or restriction beyond KDP's, and does either require a rights representation that §14's one-way indemnity makes uncomfortable to sign?
Related
- [[2026-10-06-shopify-launch-plan]]
- [[2026-07-21-akta-monid-tos-data-output-rights]]
- [[2026-09-13-anthropic-aup-agentic-use-operative-clauses]]
- [[2026-07-04-anthropic-max-plan-tos-productized-agentic-use]]
- [[2026-09-11-scribble-works-proxy-demand-signals]]
Sources
Primary (all retrieved 2026-10-14):
- QuiverAI Terms of Service, last updated March 25, 2026 - https://quiver.ai/legal/terms (operative: §5.2, §5.3, §6, §8.1-8.4, §14, §15, §18.6, §18.9, §19 Acceptable Use Policy)
- QuiverAI Privacy Policy, last updated February 25, 2026 - https://quiver.ai/legal/privacy
- QuiverAI Security and data handling - https://quiver.ai/security
- QuiverAI Pricing - https://quiver.ai/pricing
- QuiverAI Docs, Data controls - https://docs.quiver.ai/developers/platform/data-controls
- QuiverAI Docs, App pricing - https://docs.quiver.ai/app/pricing
- QuiverAI Docs, Models - https://docs.quiver.ai/developers/models
- QuiverAI Docs, MCP server - https://docs.quiver.ai/app/mcp
- QuiverAI blog, "Introducing Arrow 2 and Arrow 2 Telos", Sept 7 2026 - https://quiver.ai/blog/introducing-arrow-2-0
mcp__quiverai__list_models(read-only), returnedarrow-2andarrow-2-telosonly
Vault:
- /Users/ray/rdco-vault/01-projects/coloring-books/2026-10-06-shopify-launch-plan.md
- /Users/ray/rdco-vault/01-projects/personal-brand/assets/stickers/notes.md
- /Users/ray/rdco-vault/02-sops/packs/design/denominator/art.md
Consulted and rejected as a basis for the verdict (third-party aggregators repeating QuiverAI's own copy, no independent model-provenance disclosure): llmreference.com QuiverAI pages, Vercel AI Gateway arrow-2-telos page.