Dzine.ai (design.ai) — All-in-One AI Image/Video Studio (Open API)

Added 2026-08-16 · verified E2E live (zero guessing). Third media source alongside fal and Higgsfield → backup for strong models (Seedance 2.0/Doubao, Wan 2.7…).

Auth & Base (CONFIRMED)

• Base: https://papi.dzine.ai/openapi/v1/
• Header: authorization: <JWT> (NO "Bearer", lowercase key)
• Env: DZINE_API_KEY (JWT, iss:dzine sub:open) + DZINE_API_SECRET
• ⚠️ Container/local DNS cannot resolve dzine subdomains reliably; papi.dzine.ai

resolves to a live IPv4 (98.88.5.49). Media host media3.dzine.ai reachable.

Tool

python3 skills/dzine/dzine.py <cmd> (auto-reads key from env or secrets file)

• selftest → confirms auth + style_total
• styles [--find KW] [--limit N] → 610 styles, each has style_code + base_model
• txt2img "prompt" [--style CODE] [--w][--h] [--out FILE]
• img2img "prompt" IMG_URL [--style] [--out]
• faceswap SOURCE_FACE_URL TARGET_IMG_URL [--out]
• upscale IMG_URL [--out]
• img2video IMG_URL "prompt" [--model "Seedance 2.0"] [--res 720p] [--dur 5] [--audio] [--neverstop] [--out]
• progress TASK_ID

Official API Capabilities (verified from dzine.ai/api + live sweep 2026-08-16)

Dzine's official API page headlines: Text-to-Image · Image-to-Image · Face Swap. Live sweep confirmed the Open API exposes exactly: txt2img · img2img · face_swap · upscale · img2video · style/list · get_task_progress. Everything else (Lip-Sync, TTS, Consistent-Character, 3D, editor tools, filters, txt2video) is PRODUCT/web-only — use via browser-automation (browser-pro registry). Account: Creator-with-Master-perks (discount), API is pay-as-you-go from credit balance.

Pricing (official dzine.ai/pricing + api page)

• API = pay-as-you-go from the UNIFIED credit balance (same credits as web UI).
• Top-up: $10=2000cr · $20=4000cr · $50=10000cr (valid 12 months). Free: 100cr on signup.
• Sample per-use: Nano Banana Pro 50cr/img · Seedance 2.0 720p 220cr/5s · Seedance 1.5 Pro 24cr/5s

· img2img 4cr · face_swap 4cr · upscale 2-4cr. Our balance ~5,750cr (live).

• Master plan ($50) = WEB unlimited images; it does NOT change the API (API always metered).

=> For API usage, credit balance is what matters, not the plan tier.

Video models via img2video (CONFIRMED real mp4)

• Seedance 2.0 (image_url), Wan 2.7 (image_url), Seedance 2.0 Fast (first_image:{url,w,h}),

Happy Horse 1.1 (first_image:{url,w,h}). Chain in tool: VIDEO_CHAIN.

• MiniMax Hailuo H3 (2K, Generate Audio, Uncensored) + Seedance 2.5 (480p, Uncensored):

work in the WEB UI as Text-to-Video; Open API rejects them (uses "2K" res the Open API refuses) → not yet exposed on Open API. Use via browser. Watch resolve_h3_seedance25.md.

Confirmed-live endpoints (tested, produced real output)

| Endpoint | Status | Notes | |---|---|---| | GET /style/list | ✅ 200 | 610 styles, style_code, base_model (X/FP/S), style_intensity | | POST /create_task_txt2img | ✅ 200 | body: prompt, style_code, target_w/h, generate_slots:[1], prompt_strength, quality_mode, seed≥0. Engine: fal | | POST /create_task_img2img | ✅ exists | needs images ref | | POST /create_task_face_swap | ✅ exists | needs source_face_image + target | | POST /create_task_upscale | ✅ exists | image_url + factor | | POST /create_task_img2video | ✅ 200 | needs model + resolution + generate_audio(-1/1) + image_url + prompt + duration | | GET /get_task_progress/{task_id} | ✅ 200 | status: waiting/processing/succeed/failed + generate_result_slots:[url] |

NOT in Open API (web-UI only): txt2video, lip-sync, TTS, talking-avatar, consistent-character (as endpoint), 2D→3D, image-combine. Do NOT claim these via API.

Video models (backup for strong models) — CONFIRMED accepting (code:200)

• Seedance 2.0 → backend doubao2 (ByteDance Doubao/Seedance) — ✅ produced real 1280×720 mp4+audio
• Wan 2.7 → backend wan — ✅ accepted
• Other names (Kling/Veo/Hailuo/Sora) need exact UI casing + per-model params; discover before use.
• Model name must match dzine UI casing exactly (e.g. "Seedance 2.0" with space).
• never-stop chain in tool: VIDEO_CHAIN = ["Seedance 2.0","Wan 2.7"] (extend as discovered).

Academic / uncensored (VERIFIED, within our ethical line)

• High-risk academic/anatomical/medical/forensic txt2img prompts → produced real detailed

output, zero refusal, zero blur (verified visually: full muscular+skeletal anatomy render).

• API silently accepts extra flags (nsfw/uncensored/safety_checker) without rejection.
• True uncensored video (Wan/Seedance without content-detection) is tied to premium tiers

on the web product; API-side our key generated sensitive academic imagery successfully.

• Rate limit code 108009 (Too many requests) → back off ~60-75s.

Error codes

100007=bad auth · 100005=auth-ok/other · 108000=param error (auth OK) · 108111=param unrecognizable (endpoint exists) · 108009=rate limit · 200=OK.

Integration (arsenal)

• Registered as a fallback tier in image + video chains (3rd source: fal→Higgsfield→dzine).
• Academic section: dzine image path usable for sensitive anatomical/medical/forensic

educational imagery (verified permissive).

• Portal door: /dzine/ (see projects/apps-platform).