flux nsfw on Anyken: compare the model examples with your intended input and output. Review the capabilities documented below, then use the controls actually offered in the workspace. Available settings and credits are shown before submission.
Read source
The Flux NSFW content filter is an automated safety mechanism applied to several Flux image generation models hosted on Cloudflare Workers AI. While intended to prevent misuse, developers have reported that the filter's sensitivity leads to false positives on harmless prompts, impacting usability.
The NSFW content filter applies to multiple Flux image models, including flux-1-schnell, flux-2-klein-4b, and flux-2-klein-9b. These models are part of Black Forest Labs' Flux family, known for high-quality image generation. The filter is server-side and cannot be disabled or tuned by the user. Developers using these models for tasks like logo generation, portrait creation, or text-to-image workflows have encountered error 3030 (AiError: Input prompt contains NSFW content) even with prompts containing single words like "hamburger" or descriptions like "a man in his late 50s wearing a suit." This has led to the models being described as "unusable for production workloads" in community reports.
The false positive issue has been documented since at least October 2024, with a Cloudflare community thread remaining unresolved. Developers using img2img workflows for brand name generation have reported that completely innocuous brand names get flagged with no recourse. Adding more context to prompts sometimes helps but is not reliable. This makes Workers AI image generation impossible to depend on for any user-facing feature, as stated in the GitHub issue. The problem is compounded by the lack of any parameter to disable or tune the filter sensitivity, leaving developers without a workaround within the platform.
The Flux NSFW content filter is a mandatory safety layer on Workers AI, but its aggressive behavior has prompted calls for an opt-out mechanism. Understanding its capabilities and limitations is key for developers.
The filter operates by scanning input prompts for NSFW content before image generation. When triggered, it returns error 3030 and blocks the request entirely. There is no partial filtering or warning system; the prompt is rejected outright. Cloudflare's own SDK tests have acknowledged intermittent false positives on innocuous prompts, with test prompts being changed to avoid triggering the filter via the binding path. This suggests the issue is known internally but has not been resolved. The filter's sensitivity appears to be fixed, with no user-configurable thresholds.
The primary request from the developer community is for an optional parameter such as skip_content_filter: true or content_filter: "off" to be added to image generation models. This would allow developers to opt out of the filter when their use case doesn't require it, such as for internal tools, logo generation, or other benign applications. The GitHub issue #58 on the Cloudflare developer platform repository has received multiple +1 reactions and comments from affected developers. As of August 2026, the issue remains open with no resolution, and the feature request has not been implemented.
The Flux NSFW content filter is currently enforced on all supported Flux models within Cloudflare Workers AI. There is no way to bypass or disable it through official parameters.
The affected models—flux-1-schnell, flux-2-klein-4b, and flux-2-klein-9b—are available through Cloudflare Workers AI under the @cf/black-forest-labs namespace. They are accessible via the Workers AI API and bindings. However, the NSFW filter is applied to all prompts without exception. Developers cannot choose a different version or configuration that lacks the filter. The models themselves are functional, but the filter's false positives render them unreliable for many use cases.
Some developers have attempted to work around the filter by adding more context to prompts, but this is not consistently effective. Others have considered using alternative image generation services or self-hosting Flux models to avoid the filter entirely. Cloudflare has not provided an official workaround or timeline for adding an opt-out parameter. The community continues to advocate for a solution, with the GitHub issue serving as the primary channel for tracking progress.
There are no user-configurable settings for the model content filter. The filter is applied uniformly across all supported models and cannot be disabled or adjusted.
The NSFW content filter is enabled by default on all Flux image generation models in Workers AI. There is no parameter in the API request to disable it. The filter's sensitivity is fixed and not documented, leading to unpredictable behavior. Developers have reported that even single words like "hamburger" can trigger the filter, indicating a very low threshold. This configuration makes the models unsuitable for any application that requires consistent, unfiltered image generation.
The community has requested the addition of an optional parameter to skip the content filter. This would allow developers to choose whether to apply the filter based on their specific use case. The request includes suggestions for both a boolean flag (skip_content_filter: true) and a string parameter (content_filter: "off"). Such a change would enable production use of Flux models for benign applications while maintaining the filter for those who need it. As of the latest updates, Cloudflare has not implemented this feature.
For developers working with Flux models on Workers AI, understanding the NSFW filter's behavior is crucial for planning and troubleshooting.
When encountering error 3030, developers should first review their prompt for any words that might be misinterpreted. Adding more descriptive context can sometimes help, but this is not guaranteed. Testing with different prompt variations may identify which terms trigger the filter. For img2img workflows, the source image itself may also be subject to filtering. Developers should also check the Cloudflare community thread for updates and workarounds shared by other users.
Given the unresolved nature of this issue, developers should consider alternative platforms or self-hosting options for Flux models if they require reliable, unfiltered image generation. The GitHub issue remains the best place to track progress on the opt-out parameter request. Developers are encouraged to add their use cases to the issue to demonstrate demand. Until a solution is implemented, production use of Flux models on Workers AI for user-facing features carries significant risk of false positives.
Frequently asked questions about the model content filter and its impact on image generation.
The Flux NSFW content filter is a mandatory safety filter applied to Flux image models on Cloudflare Workers AI. It has been reported to produce aggressive false positives on benign prompts, including single words like "hamburger" and portrait descriptions. There is currently no way to disable or tune the filter. Developers have requested an opt-out parameter, but it has not been implemented. The issue is tracked on GitHub under issue #58 in the Cloudflare developer platform repository.
