v4.1.0 — June 2026 model additions

Adds 10 new registry entries for the kie.ai models that landed in June 2026. All slugs verified live against POST /api/v1/jobs/createTask on 2026-07-02; input schemas scraped from docs.kie.ai; every entry carries a full Averiguare-style research field.

Added — Video (9)

Registry key	Pricing	Highlights
happyhorse-1-1/text-to-video	22.5–29 cr/s	Native sync audio + 7-language lip-sync (#2 with-audio on Arena)
happyhorse-1-1/image-to-video	same	Single first-frame image
happyhorse-1-1/reference-to-video	same	Up to 9 refs, [Image N] prompt addressing
kling/v3-turbo-text-to-video	18–22.5 cr/s	3.0's vCoT + 6-shot at speed tier; caps 1080p
kling/v3-turbo-image-to-video	same	Same feature bundle, I2V
bytedance/seedance-2-mini	9.5 cr/s @480p (probed), 20.5 @720p	Full multimodal refs (9img+3vid+3aud) at 2.0 budget
grok-imagine-video-1-5-preview	1.6 @480p, 3.0 @720p (probed)	Cheapest with-audio video — but I2V-only (docs mis-say nullable)
omnihuman-1-5	27 cr/s	Audio-driven full-body avatar; per-character routing via masks
volcengine/video-to-video-lip-sync	8 cr/s	Re-dub existing footage; no prompt, takes video + audio URLs

Added — Image (2)

• nano-banana-2-lite — Gemini 3.1 Flash-Lite (June 30), ~4s generation. Empirical 4 cr/image (kie's site says 3 but the balance delta says 4 — same price as full NB2; the win is speed).
• omnihuman-1-5/subject-detection — Returns mask images for up to 5 subjects. Empirically FREE (verified end-to-end). Feed masks back to omnihuman-1-5 via mask_url for multi-person scenes.

Fixed

• extractResultUrls now understands resultObject.mask_urls (subject-detection result shape).
• getCostEstimate reports "free (0 credits)" for zero-priced models instead of "unknown".

Notes

• New per-second video entries flagged in PRICING_ESTIMATED because rates vary by resolution (and Seedance Mini by video-input presence).
• Sora entries untouched (paused upstream since v4.0.6 — see #18).
• generate_video tool description keeps the Sora-paused warning while promoting the new models.
• kie's new Suno Voice API was deliberately scoped out — tracked separately.

Full diff: v4.0.6...v4.1.0

🤖 Generated with Claude Code