TrueFusion Models

Understand the TrueFusion family — Skytells' native image models — and choose the right variant for every use case. speed, quality, or cost.

What you'll be able to do after this module

Select the exact right TrueFusion model for your product — whether you need a real-time preview in under 2 seconds, a high-quality production render, or a photorealistic final output.

The TrueFusion family

All TrueFusion variants share the same API. Switching models is a single string change:

Model	Quality	Speed	Cost/pred	Best for
`truefusion`	Good	Fast	$0.02	Drafts, prototyping
`truefusion-pro`	High	Medium	$0.04	Default — production apps
`truefusion-max`	Very high	Slow	$0.06	Portfolio work, print
`truefusion-ultra`	Ultra	Slow	$0.08	Maximum photorealism
`truefusion-2.0`	Next-gen	Medium	$0.06	Best coherence, complex scenes
`truefusion-edge`	Good	Very fast	$0.01	Real-time previews (Edge only)
`truefusion-pano`	High	Medium	$0.05	Panoramic / wide-format output

truefusion-pro is the right default for most products. Only move up to truefusion-ultra or truefusion-2.0 when quality genuinely matters for the end user experience.

TrueFusion Pro — the production default

truefusion-pro hits the best balance of quality, speed, and cost for most products:

curl -X POST https://api.skytells.ai/v1/predictions \
  -H "x-api-key: $SKYTELLS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "truefusion-pro",
    "input": {
      "prompt": "Professional headshot, studio lighting, neutral background, sharp focus",
      "negative_prompt": "blurry, watermark, text, distorted, low quality",
      "width": 1024,
      "height": 1024,
      "num_inference_steps": 30,
      "guidance_scale": 7.5
    }
  }'

TrueFusion Edge — for real-time apps

When your app needs near-instant responses (live previews as the user types, real-time generation), use truefusion-edge. Average generation time: under 2 seconds.

truefusion-edge requires the Edge API and is only available on Business and Enterprise plans.

// app/api/preview/route.ts
export const runtime = 'edge'; // Next.js Edge Runtime

import Skytells from '@skytells/sdk';

export async function POST(req: Request) {
  const { prompt } = await req.json();

  const client = Skytells(process.env.SKYTELLS_API_KEY, {
    baseUrl: 'https://edge.skytells.ai/v1',
  });

  const prediction = await client.predictions.create({
    model: 'truefusion-edge',
    input: {
      prompt,
      width: 512,
      height: 512,
      num_inference_steps: 4, // 4 steps = ultra-fast, good preview quality
    },
  });

  return Response.json({ url: prediction.output![0] });
}

TrueFusion 2.0 — complex scenes and coherence

truefusion-2.0 is the recommended upgrade when you need:

Complex multi-subject compositions
Accurate text rendering within images
Better spatial coherence (objects in the right place)
Following detailed or unusual instructions precisely

{
  "model": "truefusion-2.0",
  "input": {
    "prompt": "A coffee shop interior with 3 people working on laptops, afternoon light, plants in the background, warm atmosphere",
    "width": 1024,
    "height": 768,
    "num_inference_steps": 35,
    "guidance_scale": 8.0
  }
}

Third-party models via the same API

Beyond TrueFusion, Skytells gives you access to top models from other providers — same x-api-key, same endpoint:

Model	Provider	Strength
`google-imagen-4`	Google	Photorealism, following instructions
`flux-2-pro`	Black Forest Labs	Artistic, creative styles, text in images
`gpt-image-1`	OpenAI	Instruction-following, editing
`nvidia-sana`	NVIDIA	Very fast, good quality

# Switch to any model — just change the model string
for model in ["truefusion-pro", "flux-2-pro", "google-imagen-4"]:
    prediction = client.predictions.create(
        model=model,
        input={"prompt": "A majestic eagle in flight over mountains"},
    )
    print(f"{model}: {prediction.output[0]}")

Input parameters reference

Parameter	Type	Default	Description
`prompt`	string	required	What to generate
`negative_prompt`	string	`""`	What to avoid
`width`	int	1024	Output width (multiples of 8)
`height`	int	1024	Output height (multiples of 8)
`num_inference_steps`	int	20	Quality: 4 = fast preview, 30 = production
`guidance_scale`	float	7.0	Prompt adherence: 1 = loose, 20 = strict
`seed`	int	random	Set for reproducible outputs
`num_outputs`	int	1	Generate multiple variations (max: 4)

Reproducibility with seeds

Use seed to get the same image every time with the same prompt:

# Same prompt + same seed = same image
prediction = client.predictions.create(
    model="truefusion-pro",
    input={
        "prompt": "A red fox sitting in autumn leaves",
        "seed": 42,  # Fix this to reproduce exactly
        "width": 1024,
        "height": 1024,
    },
)

Tip: Generate variations by keeping the same prompt and changing only the seed. This lets you show multiple options to users while maintaining a consistent subject.

Choosing the right model — quick guide

Your situation	Recommended model
Building a prototype	`truefusion`
Production app, default	`truefusion-pro`
Live preview (user is typing)	`truefusion-edge`
Complex scene, text in image	`truefusion-2.0`
Maximum photorealism	`truefusion-ultra`
Artistic / creative output	`flux-2-pro`
Panoramic / wide-format	`truefusion-pano`

Summary

You know the full TrueFusion lineup and when to use each variant.

Start with truefusion-pro — it's right for 80% of use cases
truefusion-edge for real-time (sub-2s) — requires Business/Enterprise + Edge API
truefusion-2.0 when scene complexity or text accuracy matters
Third-party models (FLUX, Imagen, GPT-Image-1) are one model-string change away

Next: prompt engineering — how to write prompts that reliably produce the outputs you want.

NextPrompt Engineering for Images

On this page