Stability AI
Listed at https://platform.stability.ai
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | SD3.5: $0.0065/image, SDXL: $0.01/image, Stable Code: $0.50/M tokens | per million tokens |
| Output | 按图像/视频/音频输出计费，非 token 计费 | per million tokens |
🤖 Supported Models (8)
✨ Pros
- ✓Stable Diffusion 3.5 open-source image generation benchmark
- ✓Stable Video 4D video generation innovation
- ✓Stable Audio high-quality AI music generation
- ✓Stable Code coding assistant model
- ✓UK company, outside US export controls
- ✓Simple REST API interface
⚠️ Cons
- ×Image quality below Midjourney
- ×API pricing higher than local ComfyUI deployment
- ×Stable Assistant (LLM) capabilities limited
- ×China requires proxy (non-US but still restricted)
- ×Company financial stability questioned
🎯 Best For
Image generation (Stable Diffusion); video/audio generation experiments; open-source model customization
💰 Pricing & Plans
| Service | Credits | USD (1 cr = $0.01) | Best For |
|---|---|---|---|
| Free signup credit | 25 credits | $0.25 value | Trying the API |
| Stable Diffusion 3.5 Flash | 2.5 | $0.025 / image | Fast, low-cost generation |
| Stable Diffusion 3.5 Medium | 3.5 | $0.035 / image | Balanced quality/speed |
| SDXL 1.0 (legacy) | from 0.9 | $0.009 / image | Straightforward generation |
| Stable Diffusion 3.5 Large | 6.5 | $0.065 / image | High quality + prompt adherence |
| Stable Image Ultra | 8 | $0.08 / image | Flagship detail (SD3.5 Large) |
| Stable Audio 3.0 | 26 | $0.26 / clip | Up to 6 min text-to-audio |
| Creative Upscaler | 60 | $0.60 / upscale | 4K upscaling with prompt guidance |
| Stable Fast 3D | 10 | $0.10 / asset | 3D assets from a single image |
🔧 API & Developer Experience
- •REST API: gRPC + REST endpoints under api.stability.ai (v2beta), bearer-token auth
- •SDKs: Official Python (stability-sdk) + JS/TS client; community Go/Ruby wrappers
- •Response format: Base64 image bytes or PNG/JPEG/WebP artifacts; JSON error envelope
- •Billing: Prepaid credits (1 cr = $0.01), metered per operation, no subscription required
- •Free tier: 25 free credits on signup; no credit card needed to start
- •Model families: Image (SD3.5 / SDXL), Audio, Upscale, Edit, Control, 3D — one account, one key
- •Open weights: SD3.5 / SDXL weights downloadable (Community License) for local/ComfyUI use
🎨 Open-Weights Image, Audio & 3D Generation
Stability AI's differentiation is open weights: Stable Diffusion 3.5 and SDXL ship as downloadable checkpoints under the Community License, so teams can prototype on the hosted API and then self-host on their own GPUs with ComfyUI when volume grows. The platform now spans more than image generation — Audio 3.0 covers text-to-audio up to six minutes, Creative Upscaler delivers prompt-guided 4K upscaling, and Stable Fast 3D turns a single 2D photo into a usable 3D asset. All of it sits behind one API key and a single prepaid credit pool, which keeps cost attribution simple and avoids juggling separate vendors for image, audio, and 3D. For teams that need predictable per-asset costs rather than per-token billing, the credit meter (1 credit = $0.01) makes budgeting unusually transparent.
🌐 China Access & Latency
Stability AI is a UK company (London-based) and therefore sits outside US export controls, which is a meaningful advantage for teams wary of US-routed inference. In practice, though, api.stability.ai is served from Cloudflare and origin infrastructure in Europe and North America, so direct connections from mainland China are unreliable — expect packet loss and timeouts without a proxy. Through a clean overseas proxy, round-trip latency from Shanghai or Beijing typically lands in the 250-450 ms range for image and audio calls, and generation itself dominates wall-clock time (SD3.5 Large is several seconds per image regardless of region). Payment is another friction point: the platform requires an international credit card for credit top-ups.