Higgsfield releases Virality Predictor with 15-second clip scoring

• Higgsfield opened an experimental preview called Virality Predictor that scores uploaded clips for viral potential, hook strength, and related short-form metrics, according to Higgsfield's launch post.
• The preview works on clips up to 15 seconds, and Higgsfield's MCP demo says the current test version does not consume credits.
• Higgsfield's Ad Reference post positions the tool inside a larger MCP workflow: feed in top-performing ads, let an agent recreate the format, then run the outputs back through scoring.
• Higgsfield's try-it link points users to Higgsfield's MCP page, while the Marketing Studio intro is the product surface attached to the Ad Reference demo.

You can open the MCP page, skim Higgsfield's Marketing Studio intro, and watch the demo thread chain Claude, MCP, Ad Reference, and Virality Predictor into a loop that keeps regenerating and rescoring short ads. The odd bit is that the same thread says the scoring tool is free in preview, which turns it into a lightweight measurement layer rather than another paid generation step.

Higgsfield framed the release as a clip scorer, not a generator. The launch post says you upload a video up to 15 seconds long and get back a viral-potential readout plus a hook score.

That makes the product surface unusually narrow in a good way. The company is selling feedback on creative, not just more creative.

The workflow Higgsfield showed has four steps:

1. Drop top-performing videos into Ad Reference through MCP, per Higgsfield's demo.
2. Let the agent recreate the format without manual prompting, again per the same demo.
3. Score each generated output with Virality Predictor, according to Higgsfield's thread.
4. Run the process as a scheduled loop, which Higgsfield described as a compounding pipeline.

The earlier Ad Reference post adds one more detail: Higgsfield built that feature for agent pipelines on OpenClaw and other agents, so the company is clearly aiming at automated ad iteration rather than one-off prompt work.

The availability detail is simple but useful. Higgsfield's demo thread and the follow-up link post both say Virality Predictor is in experimental preview, and the thread note says it currently does not consume credits.

That gives the release a different shape from a normal model feature drop. The creative loop in Higgsfield's demo is already automated, but the scoring layer is temporarily free to test.