Models Reference

Best overall image quality. Exceptional prompt adherence, photorealism, and detail. Supports all standard aspect ratios.

Parameters

Supported Sizes

1024x1024  (1:1)     — Default
1024x1536  (2:3)     — Portrait
1536x1024  (3:2)     — Landscape
1024x2048  (1:2)     — Tall portrait
2048x1024  (2:1)     — Wide panoramic
2048x2048  (1:1)     — Maximum resolution (+$0.02 surcharge)

Cost

View full pricing →

Example Request

from openai import OpenAI

client = OpenAI(
    base_url="https://api.fluxpool.ai/v1",
    api_key="fp_your_api_key"
)

response = client.images.generate(
    model="flux-1.1-pro",
    prompt="A bioluminescent deep-sea creature in an underwater cave, hyperrealistic, 8K detail",
    size="1024x1536",
    n=1
)

print(response.data[0].url)

Example Response

{
  "created": 1719000000,
  "data": [
    {
      "url": "https://cdn.fluxpool.ai/generations/abc123.png",
      "revised_prompt": null
    }
  ],
  "model": "flux-1.1-pro",
  "usage": {
    "credits_used": 0.03
  }
}

Notes

• •Supports multilingual prompts (English recommended for best results).
• •guidance_scale above 12 may produce oversaturated results.
• •2048×2048 generations take ~5s (vs. ~3s for 1024×1024).

Excellent Asian aesthetics and character rendering. Strong text-in-image generation. Natively supports Chinese, English, Japanese, Korean, and Malay prompts.

Parameters

Supported Sizes

1024x1024  (1:1)     — Default
768x1024   (3:4)     — Portrait
1024x768   (4:3)     — Landscape
1024x1536  (2:3)     — Tall
1536x1024  (3:2)     — Wide
2048x2048  (1:1)     — Max resolution

Cost

Example Request

response = client.images.generate(
    model="qwen-image-3-6-plus",
    prompt="夕暮れ時の日本庭園、桜と鯉、浮世絵スタイル",  # Japanese prompt
    size="1024x1536"
)

Notes

• •Best-in-class for anime, manga, and East Asian art styles.
• •Text rendering in images (e.g., signs, labels) is significantly better than most models.
• •Multilingual prompts produce style-appropriate outputs (Japanese prompt → Japanese aesthetic defaults).

Open-source workhorse. Largest ecosystem of LoRAs, ControlNets, and community fine-tunes. Fastest and cheapest image model available.

Parameters

Supported Sizes

1024x1024  (1:1)     — Default, optimal
768x1024   (3:4)     — Portrait
1024x768   (4:3)     — Landscape
512x512    (1:1)     — Fast / draft quality

SDXL is optimized for 1024×1024. Sizes outside this range may reduce quality.

Cost

Example Request

response = client.images.generate(
    model="sdxl-1.0",
    prompt="Isometric RPG village, pixel art, tavern and blacksmith, warm lighting",
    size="1024x1024",
    negative_prompt="blurry, low quality, watermark"
)

Notes

• •Cheapest model available — ideal for prototyping and batch generation.
• •negative_prompt significantly improves output quality with SDXL. Always use it.
• •scheduler parameter is unique to SDXL; other models do not support it.
• •Max resolution is 1024×1024. For higher resolution, generate at 1024×1024 then upscale with real-esrgan-4x.

Best-in-class text rendering within images. Ideal for posters, logos, signage, and any generation requiring legible text.

Parameters

Cost

Example Request

response = client.images.generate(
    model="ideogram-2.0",
    prompt='Movie poster for "NEON DYNASTY" with bold title text, cyberpunk city backdrop, rain-soaked streets',
    size="1024x1536",
    style="design"
)

Notes

• •The style parameter is unique to Ideogram. Other models do not support it.
• •For text in images, include the exact text in quotes within your prompt (e.g., "SALE 50% OFF").
• •Slower than Flux or SDXL but produces the most accurate in-image typography.

Design-oriented model. Strong at clean compositions, graphic design styles, and product mockups.

Parameters

Cost

Example Request

response = client.images.generate(
    model="playground-v3",
    prompt="Minimalist product mockup, white sneakers on marble surface, clean studio lighting, editorial style",
    size="1024x1024"
)

Notes

• •Particularly strong at graphic design, editorial, and product photography styles.
• •Limited to 1024×1024 max. Use real-esrgan-4x for upscaling.

High-quality text-to-video and image-to-video. Smooth motion, cinematic output. Strong with Asian aesthetics and character animation.

Parameters

Cost

Example Request

# Text-to-video
response = client.videos.generate(
    model="wan-video-2.1",
    prompt="A koi fish leaping out of a glowing pond, slow motion, cinematic lighting",
    duration=4.0,
    resolution="720p",
    webhook_url="https://yourapp.com/webhook/fluxpool"
)

print(response.generation_id)  # Use this to poll status or wait for webhook

Example Webhook Payload

{
  "event": "generation.completed",
  "generation_id": "gen_abc123",
  "model": "wan-video-2.1",
  "status": "completed",
  "output": {
    "url": "https://cdn.fluxpool.ai/generations/gen_abc123.mp4",
    "duration": 4.0,
    "resolution": "1280x720",
    "fps": 24
  },
  "usage": {
    "credits_used": 0.15
  }
}

Notes

• •Image-to-video: provide image_url to animate a static image. The prompt then describes the desired motion.
• •Supports CJK prompts natively.
• •Average generation time: ~45 seconds for 4s at 720p.

Realistic motion and strong character consistency. Excellent for human movement and facial expressions. Supports up to 1080p.

Cost

Notes

• •Best model for realistic human motion and facial expressions.
• •1080p output is this model's differentiator — most video models max at 720p.
• •Average generation time: ~60 seconds for 5s at 1080p.

Premium video generation. State-of-the-art motion coherence, camera movement, and scene consistency. The highest-quality video model available.

Cost

Notes

• •Most expensive video model but highest quality output.
• •The motion parameter is unique to Runway Gen-3. low produces smoother, more controlled camera work. high produces more dynamic movement.
• •Average generation time: ~90 seconds for 4s at 1080p.

Good balance of quality and cost. Strong with nature scenes, abstract visuals, and stylized content.

Cost

Notes

• •Best value video model — cheapest option for 720p video.
• •Max duration is 8s (shorter than other video models).
• •Supports CJK prompts.

Fast video generation. Best for quick iterations, prototyping, and social media content.

Cost

Notes

• •Fastest video model — ~30s for 4s at 720p. Ideal for rapid iteration.
• •Max duration is 5s (shortest of all video models).
• •Good enough quality for social media; use Runway or Kling for production-grade output.

4x image upscaling. Enhances resolution while preserving detail. Use after generating at 1024×1024 to produce 4096×4096 output.

Parameters

Cost

Notes

• •Input image must be ≤ 2048×2048.
• •4x upscale of a 1024×1024 image produces a 4096×4096 output.
• •Works with any image URL, not just Fluxpool-generated images.
• •Processing time: ~5 seconds regardless of scale.

Generate images guided by edge-detection maps from a source image. Preserves composition and structure while applying new styles.

Parameters

Cost

Notes

• •Useful for maintaining composition from a sketch, wireframe, or existing image while generating in a new style.
• •control_strength above 1.0 may produce artifacts. Start at 0.7–0.8.
• •The base_model parameter selects which underlying model does the generation — cost matches the selected base model plus a $0.01 ControlNet surcharge.