media-forgehiggsfield-promptinglisted

# Higgsfield Prompting Playbook For the provider-neutral craft that applies before any of this — mode gate (T2V/I2V/V2V/R2V/FLF2V/edit/extend), the director formula, compression order, and the output contract — load `[skill:mf-video-prompt]`. This skill owns Higgsfield's own mechanisms; that one owns the shared discipline. When both are loaded and they differ on prompt shape, the provider-specific guidance here wins for Higgsfield. Verified surface facts — endpoints, auth, prompt budget, aggregator behaviour — live in `[ref:surface-prompt-profiles]` with sources and dates. **Two things to know before writing a Higgsfield prompt:** Higgsfield publishes **no prompt character limit**. The conservative profile applies: keep it compact because compactness sharpens intent, not because an API bound forces it. **Do not extend the anti-slop pass to Higgsfield image prompts.** Higgsfield's own image documentation recommends quality modifiers such as `"highly detailed"` or `"8k"`, and slop class 2 in `[skill:mf-antislop]` deletes exactly those as "borrowed image-model tokens". This is a guard, not a live fix. Today the two never meet: `mf-antislop` scopes itself to video prompts in its own description, and its only invokers are `[skill:mf-video-prompt]` and `[skill:mf-troubleshoot]` — both video paths. No image skill invokes it. So nothing is currently stripping words that Higgsfield asks for. The note exists because the obvious future "improvement" is to run anti-slop over image