How do other AI image services (DALL·E, Midjourney) handle similar flagged content and appeals?
Executive summary
DALL·E and Midjourney both enforce strict content policies but take different technical and community approaches: Midjourney historically relied on banned-word filters and later shifted to AI-powered moderation, while OpenAI’s DALL·E is described in reporting as more policy-detailed and comparatively stricter in enforcement and indemnification [1] [2] [3]. Both platforms have drawn user complaints about opaque or ineffective appeal experiences, and academic reporting finds failures in presenting clear moderation criteria and in supporting appeals across multiple generative-AI tools [4] [5] [6].
1. Midjourney’s evolution: banned words to AI moderation, and the consequences
Midjourney’s earliest public moderation mechanism used a banned-word list that blocked prompts containing specific terms — including sexual, violent, and even some political or religious names — and that approach drew criticism for bluntness and overreach [1] [7]. Starting in May 2023 Midjourney reported a shift toward AI-powered moderation intended to interpret prompts holistically and allow context-dependent uses of words previously banned, but users continued to report unexpected false positives and deleted outputs as the system evolved [1] [7] [4]. The company’s community rules allow moderators to warn, time-out, or block users and provide users the ability to flag content through Discord or the website, but moderation actions can still remove images from public galleries and limit creators’ visibility [8] [7].
2. DALL·E’s posture: policy depth, stricter gating, and design tradeoffs
Reporting and product comparisons present DALL·E as having more “in-depth policies” and stronger ownership and indemnification language, framing its moderation as comparatively restrictive and deliberate — a design that can slow generation but aims for legal and safety guardrails [2] [3] [9]. Analysis-oriented commentary on DALL·E 3 suggests its censorship outcomes follow thematic boundaries intended to uphold ethical and safety guidelines, indicating a context-aware filter that blocks prompts deemed to fall outside safe creative parameters [10]. That posture trades speed and creative latitude for policy clarity and risk management, according to multiple comparisons [2] [9].
3. Where appeals and transparency break down — common user complaints
Cross-platform research and subreddit sentiment analyses document a recurrent complaint: users say appeals are slow, opaque, or ineffective, and that moderation reasons are often unexplained — a pattern visible in both Midjourney anecdotes and broader generative-AI forums [4] [6] [5]. Midjourney users report receiving generic “our AI moderator thinks this prompt is probably against our community standards” messages and describe a broken appeal pipeline that deters people from contesting blocks, while broader studies flag failures by several tools to give clear criteria or practical remediation paths [4] [6] [5]. Midjourney’s own support channels are limited — for example, its email support focuses largely on billing — which complicates timely human review for contested moderation actions [2].
4. Practical differences that affect flagged-content outcomes
Operational contrasts matter: Midjourney’s default community visibility of generated images means flagged content can be immediately seen and reported by other subscribers, increasing moderation pressure, whereas DALL·E’s platform design and stated policies prioritize tighter gating and indemnification that can prevent some outputs from being attempted or surfaced in the first place [2] [8]. Midjourney maintains an actively curated banned-word list and a public-facing moderation stance that evolved in response to user feedback and legal pressure, whereas DALL·E’s publicly-cited strength is more formalized policy infrastructure that underpins its moderation choices [7] [3].
5. Competing narratives, business incentives, and unanswered questions
Industry coverage presents two competing narratives: one that platforms like Midjourney overreach with blunt filters and community moderation, and another that platforms like DALL·E are more conservative but clearer in rules — each narrative masks incentives: Midjourney’s open-community model benefits from rapid iteration and low-friction sharing, creating pressure to limit harms via blunt filters, while OpenAI’s corporate posture and indemnification incentives favor formal policy and cautious blocking [7] [2] [3]. Major gaps remain in public reporting: academic and media sources document user frustration and systemic opacity but do not provide a comprehensive, platform-by-platform accounting of appeal success rates or the exact mechanics of DALL·E’s human-review processes, so definitive comparative claims about appeal outcomes are limited by available evidence [4] [5] [6].