# Image Generation APIs -- TESTED RESULTS (2026-02-27)

## CONFIRMED WORKING (with your actual API keys)

### 1. NVIDIA NIM -- FLUX.1-dev (BEST QUALITY)
- **Status**: WORKING
- **Speed**: 4.8 seconds per image
- **Resolution**: 1024x768
- **Anime Quality**: EXCELLENT (professional anime key visual level)
- **Free Credits**: 1,000 on signup (up to 5,000 with business email + 90-day enterprise license)
- **API**:
```python
import requests, base64

r = requests.post(
    "https://ai.api.nvidia.com/v1/genai/black-forest-labs/flux.1-dev",
    headers={"Authorization": "Bearer YOUR_NVIDIA_KEY", "Accept": "application/json"},
    json={
        "prompt": "anime style, young warrior girl, silver hair, blue eyes, glowing sword, fantasy forest, masterpiece",
        "mode": "base",
        "cfg_scale": 3.5,
        "width": 1024,
        "height": 768,
        "seed": -1,
        "steps": 30
    }
)
img_bytes = base64.b64decode(r.json()["artifacts"][0]["base64"])
with open("output.png", "wb") as f:
    f.write(img_bytes)
```

### 2. NVIDIA NIM -- FLUX.1-schnell (FASTEST)
- **Status**: WORKING
- **Speed**: 3.4 seconds per image
- **Resolution**: 1024x1024
- **Anime Quality**: EXCELLENT
- **API**: Same as above, endpoint: `black-forest-labs/flux.1-schnell`, steps=4

### 3. HuggingFace -- FLUX.1-schnell (MOST GENEROUS FREE)
- **Status**: WORKING
- **Speed**: 1.4-3.8 seconds per image
- **Resolution**: up to 1024x768
- **Anime Quality**: EXCELLENT
- **Free Tier**: Rate-limited but generous (3 rapid requests all succeeded)
- **API**:
```python
import requests

r = requests.post(
    "https://router.huggingface.co/hf-inference/models/black-forest-labs/FLUX.1-schnell",
    headers={"Authorization": "Bearer YOUR_HF_TOKEN", "Content-Type": "application/json"},
    json={
        "inputs": "1girl, silver hair, blue eyes, holding sword, anime style, masterpiece",
        "parameters": {"negative_prompt": "low quality, blurry", "width": 1024, "height": 768}
    }
)
with open("output.png", "wb") as f:
    f.write(r.content)  # Returns raw image bytes
```

### 4. HuggingFace -- SDXL Base (GOOD PAINTERLY STYLE)
- **Status**: WORKING
- **Speed**: 9.8 seconds per image
- **Resolution**: up to 1024x768
- **Anime Quality**: EXCELLENT (more painterly/artistic style)
- **API**: Same as above, model: `stabilityai/stable-diffusion-xl-base-1.0`

---

## NOT WORKING / NOT FREE

| Service | Status | Reason |
|---------|--------|--------|
| OpenRouter | NO free image models | All 29 free models are text-only |
| Prodia | Needs paid upgrade | Free accounts can't create API tokens |
| Google Gemini | Quota = 0 on free tier | Image gen removed from free tier entirely |
| Google Imagen | Paid only | "Imagen 3 is only available on paid plans" |
| Pollinations | HTTP 530 | Cloudflare blocking (may work from other networks) |
| Together AI | No free tier (2026) | User confirmed free tier discontinued |
| DeepInfra | Needs auth/credits | 403 without authentication |
| fal.ai | Needs auth/credits | 401 without authentication |
| HF FLUX.1-dev | DEPRECATED | "No longer supported by provider hf-inference" |
| HF Animagine XL | 404 | Not available on inference API |
| Stability AI | Needs API key | 401 without key |

---

## RECOMMENDED STRATEGY

### Primary: HuggingFace FLUX.1-schnell
- Fastest (1.4-3.8s)
- No visible hard limit yet
- Excellent anime quality
- Use for bulk image generation

### Secondary: NVIDIA NIM FLUX.1-dev
- Best quality (30-step generation)
- 1,000-5,000 free credits
- Use for hero/key moment images
- Save credits for important scenes

### Fallback: HuggingFace SDXL
- Different artistic style
- Good for variety
- 9.8s per image (slower)

### Combined Monthly Estimate
- HuggingFace: Unknown hard limit, but generous for free tier
- NVIDIA: 1,000+ credits (unknown images-per-credit ratio)
- Pollinations: ~1,000 images (if accessible from your network)
- **Total: Likely 1,000-5,000+ images/month for $0**

### Images Per Chapter
- A 3-5 minute episode needs ~20-30 keyframe images
- At 2,000 images/month = ~66-100 chapters/month
- More than enough for production

---

## QUALITY RANKING (for anime)

1. NVIDIA FLUX.1-dev (30 steps) -- Most detailed, best composition
2. NVIDIA FLUX.1-schnell -- Great quality, fastest NVIDIA
3. HuggingFace FLUX.1-schnell -- Slightly different style, very fast
4. HuggingFace SDXL -- More painterly, artistic feel

All four produce professional-grade anime art suitable for our project.