PhotoGPT Text2Img Inference Advance Options Comprehensive Guide

PhotoGPT's Text2Img studio hides a powerful set of inference controls. Mastering these options gives you granular control over style, fidelity, and iteration speed.

Understanding the Key Controls

1. Sampler

Choose how the model walks through the latent space. Start with Euler a for quick previews, then switch to DPM++ 2M Karras when you need crisp final renders. Experimenting across samplers can unlock new textures without rewriting prompts.

2. CFG Scale

The "Classifier Free Guidance" slider balances prompt adherence and creativity.

• 5–7 keeps the output faithful while leaving room for organic details.
• 8–11 locks tightly to your prompt—ideal for brand guidelines.
• Higher values can introduce artifacts, so increase gradually.

3. Steps

More steps equal more refinement. Use 20–24 steps for speed runs and 35+ for production-quality imagery. Pair longer runs with the High-Resolution Fix option if you're generating large portraits.

4. Seed

Lock a seed to reproduce a composition or vary it to explore alternatives. PhotoGPT exposes the seed value for every render so you can revisit previous looks later.

5. Aspect Ratio & Resolution

Set aspect ratios that match your destination—1:1 for social avatars, 3:4 for print. The resolution field automatically scales the diffusion pipeline to avoid stretching.

Workflow Tips

1. Draft Fast – Begin with low steps and a conservative CFG to explore. Save promising outputs to a "Shortlist" collection.
2. Refine the Winners – Reuse the seed, bump up steps, and tighten CFG to lock down details.
3. Stack Negative Prompts – Remove artifacts (blurry, extra limbs) to keep portraits clean.
4. Leverage Variations – PhotoGPT can generate multiple seeds per run, giving you options without rerunning the prompt manually.

When to Use Each Setting

Armed with these controls, you can guide PhotoGPT to produce exactly the imagery your campaigns demand.