Changelog

Developer experience, model discovery, and idempotency

• RESTful task status: Poll tasks via GET /api/v1/task/{task_id} (or GET /api/v1/task/status?task_id=...) alongside POST /api/v1/task/status. Uncharged or refunded tasks report credits_deducted: 0.
• Idempotent paid submissions: Pass an optional Idempotency-Key header on POST /api/v1/task/submit/{model_key} and POST /api/v1/music/{action} to safely deduplicate network retries for 24 hours (409 idempotency_key_conflict / 409 idempotency_key_in_progress).
• Model discovery & 24h telemetry: Added GET /v1/models (OpenAI-compatible model list with Wizzx tier metadata), GET /pricing?locale=zh (English notes by default, Chinese when requested), and GET /pricing/metrics (rolling 24-hour settled reliability and average latency).
• Richer balance & error diagnostics: GET /api/v1/credits/balance now includes free_credits, paid_credits, and usd_equivalent. 402 insufficient_credits rolls back atomically before queuing and returns balance, required, shortfall, and billing_url. 429 responses include the standard Retry-After HTTP header.
• Pricing sync: Updated GPT Image 2.5 documentation to match the live GET /pricing catalog (1K = 15, 2K = 24, 4K = 35 credits per image).

Suno music actions

• New endpoint POST /api/v1/music/{action} with 33 Suno actions, each billed per call as one asynchronous suno task:
  • Creation: generate, inspo (from reference audio), sounds, lyrics, upsample-tags, upload, upload-cover, upload-extend, create-model, create-voice
  • From an existing track: extend, concat, cover, remaster, add-vocals, add-instrumental, add-stem, mashup, sample, persona
  • Editing: replace-section, remove-section, crop, fade-in, fade-out, adjust-speed, stems, stems-all
  • Analysis and export: midi, aligned-lyrics, bpm, music-video, download
• Reference earlier tracks with the Wizzx task_id of your own finished Suno task and the track’s audio_index. Invalid references are rejected with HTTP 422 before any credits are reserved.
• Each action accepts only its own fields; other fields return HTTP 422.
• POST /api/v1/task/status now returns a structured result object for finished music tasks: tracks with their index, cover image, title, lyrics, and duration, plus lyrics, enhanced style, download files, video, BPM, MIDI, timed lyrics, persona, and model IDs. result_urls is unchanged.
• generate also accepts custom_model_task_id, persona_task_id, style_weight, weirdness, audio_weight, auto_lyrics, variety, and audio_format. POST /api/v1/task/submit/suno keeps working and equals generate.
• Pricing per call: 60 credits for generation-class actions (120 with max_mode) and for remaster and midi, inspo 131 (262), stems 193, stems-all 462, create-model 1,847, and 2–47 credits for every other action. See Pricing.

• lyrics results are now returned in result.lyrics with title, text, and style for each draft.
• concat only accepts an extend task as its source; other sources return HTTP 422 before credits are reserved.
• create-voice returns a persona_id. Pass the task’s task_id as persona_task_id, the same as a persona task. Use a clean vocal recording without music.
• The error field of POST /api/v1/task/status now always returns a readable message for failed tasks, for every model.

New Model: GPT Image 2.5

• One model key, gpt-image-2.5, for text-to-image and image-to-image editing
• version: flare (default, faster) or sunburst (finer edits for product, ad, and multi-turn editing), same price
• Up to 16 reference images as HTTP(S) URLs or base64 data URIs (JPEG, PNG, WebP); reference images are not billed
• 11 aspect ratios including auto, and 1K / 2K / 4K output
• One image per request
• Pricing: 1K = 13, 2K = 21, 4K = 32 credits per image

More new models

• Kling v3 (kling-v3, Kuaishou): text-to-video and first/last-frame image-to-video, 720p/1080p, 3–15 s, optional native audio. 130/173 credits/s silent, 194/259 with audio (720p/1080p)
• Kling v3 Omni (kling-v3-omni, Kuaishou): adds reference-to-video with up to 7 images cited as <<<image_N>>>. 130/173 credits/s silent, 173/216 with audio; reference images are free
• Seedance 2.0 Mini (seedance-2.0-mini, ByteDance): text-to-video and first/last-frame image-to-video, 480p/720p, 4–15 s, audio on by default at no extra charge. 21/44 credits/s
• PixVerse V6 (pixverse-v6, PixVerse): text, image, first/last-frame transition, and 1–7 image reference modes, 360p–1080p, 1–15 s. 31/47/62/124 credits/s silent, 47/62/77/154 with audio. Image-to-video follows the input image and rejects aspect_ratio
• Grok Imagine Video 1.5 (grok-imagine-video-1.5, xAI): text-to-video or image-to-video with 1–7 references, 480p/720p, 6–15 s. 20/37 credits/s
• Gemini Omni Flash (gemini-omni-flash, Google): text, first-frame, and 1- or 3-image reference modes, 720p/1080p/4K, 4/6/8/10 s. Fixed per video: 481/577/674/770 credits at 720p and 1080p, 1,443/1,539/1,635/1,731 at 4K
• Wan 2.6 Flash (wan-2.6-flash, Alibaba): silent image-to-video from one first frame, 720p/1080p, 2–15 s. 33/54 credits/s

• Qwen Image 2.0 (qwen-image-2.0) and Qwen Image 2.0 Pro (qwen-image-2.0-pro, Alibaba): text-to-image and editing with up to 3 references, 7 aspect ratios, 1K/2K at the same price. 31 and 74 credits per image
• Seedream 5.0 Pro (seedream-5-pro, ByteDance): text-to-image and single-reference editing, 10 aspect ratios plus auto, 1K/1.5K/2K, JPEG or PNG, one image per request. 44 credits per image at 1K/1.5K, 89 at 2K

• Suno (suno): Suno V6 songs and instrumentals in inspiration or custom (lyrics) mode, versions v6, v6-wild, v6-mini. Each request usually returns two tracks in result_urls. 60 credits per request, 120 with max_mode

Veo 3.1 Quality

• veo3 now accepts model: "veo3_quality" at 720p, 1080p, or 4K for 1,924 / 1,924 / 2,885 credits per video.
• Quality supports 6 or 8 seconds and does not support reference-to-video. Fast and Lite are unchanged.

Runway Gen-3 is offline

• runway-gen3 no longer accepts new tasks. Submissions return HTTP 503 with error code model_unavailable, and the model is no longer listed in GET /pricing.
• Tasks accepted before the change still complete and settle normally, and you can keep polling them with POST /api/v1/task/status.
• For new video tasks, choose a model listed in GET /pricing, such as Veo 3.1, Kling AI, or Seedance.

Veo 3.1 Lite and fixed per-video pricing

• Added veo3_lite alongside veo3_fast on the existing veo3 endpoint.
• Fast and Lite support 720p, 1080p, and 4K with 4, 6, or 8-second output.
• Pricing is fixed per video by variant and resolution; duration does not multiply the charge.
• Lite costs 180/210/900 credits and Fast costs 360/390/1080 credits at 720p/1080p/4K.

New Models: Nano Banana 2 Series

• Up to 14 reference images for image-to-image generation
• Resolution options: 1K, 2K, 4K (default 1K)
• Output format: png or jpg (default jpg)
• Prompt length up to 20000 characters
• Pricing: 1K = 48, 2K = 72, 4K = 108 credits per image

• Budget option at a fixed 24 credits per image
• Up to 10 reference images
• Same 15 aspect ratio options as Nano Banana 2
• No resolution or output format controls

• Use Nano Banana 2 for 2K/4K output, PNG format, or more than 10 references
• Use Lite for high-volume drafts and cost-sensitive workloads

New Model: WAN 2.2 Animate Move

• Image + Video animation: Combine an image with a video to create animated content
• No prompt required: Unlike WAN 2.5/2.6, this model is purely image+video driven
• Resolution options: 480p, 580p, 720p
• Input requirements:
  • Video: mp4, mov, mkv (max 10MB)
  • Image: jpeg, png, webp (max 10MB)
• Fixed 5-second billing

New Model: Seedance 1.5 Pro

• Audio generation: Generate synchronized audio (voice, sound effects, background music) with generate_audio parameter
• First + Last frame support: Control both start and end frames with up to 2 images in image_urls
• Flexible duration: 4-12 seconds or auto (-1)
• Text-to-video and image-to-video modes
• Resolution options: 480p, 720p (no 1080p)
• Aspect ratios: 16:9, 4:3, 1:1, 3:4, 9:16, 21:9, adaptive
• return_last_frame option for consecutive video generation

• Use 1.5 Pro for audio generation and first+last frame control
• Use Pro Fast for 1080p resolution and simpler workflows

New Model: ElevenLabs TTS Turbo 2.5

• 50% cheaper than TTS V2
• Faster generation, optimized for low-latency scenarios
• Same 21 voice options and parameters as V2
• Pricing: 40 credits per 1000 characters

New Model: ElevenLabs TTS V2

• First audio model on the platform
• High-quality multi-language text-to-speech
• 21 voice options (Rachel, Brian, Sarah, etc.)
• Text length: 1-5000 characters
• Voice controls: stability, similarity_boost, style, speed
• Multi-language support with automatic detection
• Pricing: 80 credits per 1000 characters

• Professional/formal: Brian, Daniel, George
• Warm/friendly: Rachel, Sarah, Alice
• Energetic/youthful: Aria, Charlotte, Lily

Model Update: Kling AI V2.6

• NEW: Text-to-video generation support (V2.6)
• NEW: Audio generation with sound parameter (V2.6 only)
• NEW: Aspect ratio control with aspect_ratio parameter (V2.6 only: 16:9, 9:16, 1:1)
• NEW: kling-v2-6 model version - latest and recommended
• Supports both text-to-video and image-to-video modes
• Pricing: 65 credits/s (no sound), 130 credits/s (with sound)

New Model: WAN 2.6

• Text-to-video and image-to-video generation
• Extended video duration: up to 15 seconds (vs 10s for WAN 2.5)
• Extended prompt length: up to 5000 characters (vs 800 for WAN 2.5)
• Resolution options: 720p, 1080p
• Pricing: 720p = 80 credits/s, 1080p = 110 credits/s

• Use WAN 2.6 for longer videos (15s), detailed prompts, and improved quality
• Use WAN 2.5 for aspect ratio control, negative prompts, seed control, and prompt expansion

Documentation: Error Codes Reference

• Complete API error codes reference for all HTTP status codes (400-503)
• Detailed error types: authentication, invalid request, quota, permission, rate limit, server errors
• Example responses and solutions for each error code
• Error handling best practices with retry logic examples
• Available in both English and Chinese

New Model: WAN 2.5

• High-quality video generation powered by Alibaba
• Text-to-video and image-to-video modes
• Resolution options: 720p, 1080p
• Video duration: 5s or 10s
• Aspect ratios: 16:9, 9:16, 1:1 (text-to-video only)
• Advanced features: negative prompts, LLM prompt expansion, seed control
• Pricing: 720p = 60 credits/s, 1080p = 100 credits/s

New Models: Flux 2 Series & Nano Banana

• Megapixel (MP) based pricing for flexible cost optimization
• Flux 2 Pro: 20 credits for first MP + 10 per additional MP
• Flux 2 Flex: 40 credits per MP with advanced controls (guidance & steps)
• Support up to 8 reference images for image-to-image generation
• Total MP = Input MP + Output MP

• Fixed 24 credits per generation - simple and predictable
• Support up to 10 reference images (more than Pro version)
• Smart mode selection automatically chooses text-to-image or image-to-image
• Wide range of aspect ratios: 1:1, 16:9, 9:16, 21:9, and more

New Models: Seedream 4.0 & 4.5

• High-quality image generation powered by ByteDance
• Multi-image fusion: combine up to 14 reference images
• Batch generation: create up to 15 images in one request
• Resolution options: 1K, 2K, 4K
• Seedream 4.0: 30 credits base price
• Seedream 4.5: 40 credits for enhanced quality

New Models: Video Generation Expansion

• Professional image-to-video generation
• Multiple model versions: v1, v1-5, v1-6, v2-master, v2-1, v2-1-master, v2-5-turbo
• Quality modes: Standard and Pro
• Advanced features: camera controls, dynamic brushes, trajectory animation
• Video duration: 5s or 10s
• Pricing: 10-40 credits (varies by quality mode and duration)

• Fast video generation with ByteDance Seedance
• Text-to-video and image-to-video modes
• Resolution options: 480p, 720p, 1080p
• Video duration: 5s or 10s
• Pricing: 10-50 credits/second based on resolution

New Models: Flux Kontext Series

• Multi-reference image support (up to 4 images)
• Flexible aspect ratios from 21:9 (cinematic) to 9:21 (portrait)
• Prompt upsampling for enhanced creativity
• Safety tolerance controls (0-6 levels)
• Kontext Pro: 30 credits - fast and cost-effective
• Kontext Max: 60 credits - premium quality output

New Models: Professional Video Generation

• Industry-leading video generation technology
• Text-to-video and image-to-video modes
• High-quality output: 720p and 1080p
• Video duration: 5s or 10s
• Aspect ratios: 16:9, 9:16, 1:1
• Pricing: 5-10 credits based on duration and resolution

• Google’s state-of-the-art video generation model
• Three generation modes:
  • Text-to-video: pure text prompts
  • Image-to-video: first/last frame control
  • Reference-to-video: style reference generation
• Multiple aspect ratios supported
• Fast and standard model options
• Pricing: 120 credits per second

Platform Launch

• High-quality text-to-image generation
• Fast generation speed
• 40 credits per image
• Aspect ratios: 1:1, 16:9, 9:16

• Advanced image generation with resolution control
• Resolution options: 1K, 2K, 4K
• Multi-reference support (up to 8 images)
• Pricing: 10-20 credits based on resolution

• AI-powered virtual garment try-on
• Single garment and combo outfit support
• Model versions: v1 (single item) and v1-5 (top+bottom combo)
• 65 credits per generation