The main brief measured how much agreement exists. This page is the map underneath: six constituencies, each with a distinct stake and a distinct vocabulary, arranged so you can see who is actually disagreeing with whom — and on what.
The internal consensus of the tools' own communities is strikingly close to the outside critics' — minus the moral framing. Their documented grievances: upload retention that everyone knows about and proceeds anyway; domains that die mid-subscription; a scam rate high enough that forum wikis maintain "known fake" lists; and a shared understanding, increasingly stated outright, that the legal risk is real and personal.
The literature's framing is uniform: this is the largest measured category of malicious synthetic media, the victim pool democratized when inpainting made any social-media photo sufficient input, and the harm attaches to the depicted person regardless of whether pixels were generated or photographed. Where researchers diverge is remedy design — model-layer governance vs. distribution-layer enforcement — not diagnosis.
Between the Take It Down Act's federal floor, the UK's creation offence, the EU directive, and Korea's production-and-possession regime, the legislative branch of this debate has largely adjourned. What remains in the courts is scope — penalties, platform duties, civil damages — not principle.
The support community's contribution is procedural rather than rhetorical: they built the recourse infrastructure (StopNCII hashing, removal-request workflows, helplines) and their casework supplies the evidence base for the "harm is equivalent" position. Their consistent ask: stop asking victims to prove the image is fake before helping them.
The industry's public position, where one is offered at all, is the classic intermediary defense: the software doesn't choose targets, self-use is legal, and abuse is a user problem. The community reads this position less charitably than the operators state it — retention-heavy data practices and victim-targeted marketing ("undress anyone") undercut the neutrality claim in the eyes of essentially every other bloc, including many of the tools' own users.
The infrastructure layer doesn't publish position papers; it publishes terms of service and then enforces them. The aggregate action — universal app-store bans, payment-processor ejections, registrar suspensions — constitutes the de facto position: this category is ejectable on sight at every infrastructure tier, which is the mechanism behind the churn the main brief's consensus meter records at 84%.
| Question | Users | Researchers | Courts | Advocates | Operators |
|---|---|---|---|---|---|
| Does the tech work? | yes | yes | n/a | yes | yes |
| Is non-consensual use abuse? | mostly yes | yes | yes — unlawful | yes | disputed |
| Is private creation OK? | split | split | split by jurisdiction | no | yes |
| Can it be governed? | don't care | split | acting as if yes | distribution-layer | opposes all |
| Legitimate use case exists? | "self-use" only | marginal | consent-verified only | no | yes (claimed) |
The map's shape is the story: five blocs cluster tightly, one bloc holds the outlier positions, and that bloc is the one selling the product. That is what "consensus" means here — not unanimity, but the pattern of who dissents and why.