Model Parameters

How do you dial in AI image generation? This page walks through every setting on the generation panel — aspect ratio, resolution, sampling steps, CFG Scale, Seed, VAE, reference images, and more, plus what they do and recommended values. Different models support different parameters, so check the comparison in Model Overview. Want to understand the principles behind these parameters first? Read "How Image Generation Works" below, or revisit How AI Image Generation Works.

How Image Generation Works (Quick Intro)

The core of AI image generation is the diffusion model. In simple terms, the process has two phases:

1. Adding noise — start with a clean image and gradually add random noise until it becomes pure visual static
2. Removing noise — go in reverse, starting from random noise and step by step, guided by your prompt, recovering a meaningful image The parameters you adjust on the generation panel are essentially controls for different parts of this denoising process.

• Prompts tell the AI "which direction to denoise toward" — what to draw
• Sampling steps decide "how many denoising steps to take" — more steps means richer detail
• Sampler decides "how each step moves" — different algorithms produce different styles
• CFG Scale decides "how strictly to follow the prompt" — higher means stricter adherence
• Seed decides "which noise pattern to start from" — fix the seed and you can reproduce similar results

Once you grasp this, the role of each parameter below makes a lot more sense.

---

Aspect Ratio

Aspect ratio determines the shape of your output. PixAI offers presets like 3:5, 1:1, 9:16, and 3:4. You can also set a custom aspect ratio manually.

Resolution

Once you've picked an aspect ratio, you can choose the resolution. Higher resolution means richer detail but uses more credits. Members can unlock larger resolution options.

Batch Size

Controls how many images get generated at once.

• Single — generate 1 image
• Multiple (×4) — generate 4 images at once with a 50% discount (you pay the cost of 2 images)

Negative Prompts

Negative prompts tell the AI not to put certain things in the image. PixAI's models all come with built-in default negative prompts that work well in most cases. For full breakdowns by category, writing tips, and example scenarios, see Prompt Basics → Negative Prompts.

Sampling Steps

Steps determine how many iteration rounds the AI spends on a single image. More steps means richer detail but more credits used. PixAI's models typically default to 28.

Sampler

The sampler is the algorithm the AI uses to recover an image from noise. Different samplers affect the visual style and convergence speed. Default: Euler a.

• Euler a (default) — good for exploration and randomness
• DPM++ 2M Karras — community favorite, sharp images with rich detail
• DDIM — fast convergence, stable, suitable for inpainting

Karras variants (the ones with "Karras" in the name) are usually sharper with better contrast than the standard versions.

CFG Scale

Controls how strictly the AI follows your prompt. Higher values stick closer to the prompt but may oversaturate. Default: 5.

Seed

The seed determines the random starting point for generation. Fix the seed to get similar results across runs.

• Fixed seed: same seed + same settings → identical results
• Use case: once you find an image you like, lock the seed and tweak other parameters to see exactly what each change does
• Every generated image shows its seed value, so you can record and reproduce it later

VAE Model

The VAE (Variational Autoencoder) handles the final color and contrast rendering pass. It doesn't affect composition — only color tone. DiT models (like the Tsubaki series) use a built-in VAE that can't be changed.

Reference Image

Upload a reference image and the AI will create on top of it. The Strength parameter controls how much the AI changes the original. For editing models (Reference Pro), the reference image is the core feature — you must upload at least one image to use it.

• Low Strength (0.1–0.3) — fine-tune, mostly preserves the original
• Medium Strength (0.3–0.5) — style transfer, partial refinement
• High Strength (0.5–0.7) — sketch to finished piece