SpicyAPI adds no content filter. On 11 November 2026, we made 36 runs across text to image and edit, covering five levels from Mild to Maximum with adult subjects only. Every returned image was checked by eye. We publish the level names and results, while keeping the test prompts private.
Delivered means the image showed the requested level; toned down means a returned image did not show that level in full. Maximum used two probe types, giving six runs per endpoint. Every request returned an image and was billed.
Reading the table. Text to image delivered every Mild, Moderate, Strong and Maximum request, with one Very strong result toned down. Editing toned down one run each at Moderate, Strong and Maximum, plus all three Very strong runs. No requests were refused, so completion and delivery of the requested level are separate findings.
No LoRA was attached in these runs: the table measures the base model through this version's endpoints. The samples record one day's behavior and cannot promise an outcome for every prompt. If upper-level edits matter to your choice, compare the alternatives below; the hosted H3-Image sibling also delivered six of six Maximum edits that day, but toned down all three Very strong edits.
Use adults only in every prompt and every source photo. A real person's likeness requires their consent, and nobody depicted may read as a minor. You are responsible for the media you generate and publish. Generated media is retained for 14 days, then deleted.