From cef04fb60b28bb46b76eb91252dc0d9a0d64095b Mon Sep 17 00:00:00 2001 From: Yiling-J Date: Fri, 20 Feb 2026 10:51:04 +0800 Subject: [PATCH] Revise model recommendations in README Updated model recommendations for equipment extraction, character generation, and image generation. Added notes on budget-friendly alternatives and potential limitations of non-Pro models. --- README.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index c3b0e3f..ec0f8e8 100644 --- a/README.md +++ b/README.md @@ -82,9 +82,11 @@ Forgery requires API keys for AI generation services (Google Gemini and/or OpenA ## Model Recommendations -For optimal results in both equipment extraction and high-fidelity character generation, it is highly recommended to use the **Nana Banana Pro** model. This model provides superior precision for isolating modular assets and ensures the most consistent "looks" when remixing styles and accessories. +For text generation, Gemini 3 Flash or GPT-5 Mini are typically sufficient for most use cases. -For the Refine step, the standard Nano Banana model is typically sufficient to handle iterative adjustments and polish without the need for the Pro-level tier. +For image generation, Nanao Banana Pro delivers the highest-quality results and is generally the best choice. However, if cost is a concern, you can consider Nano Banana (non-Pro) or GPT Image 1.5 as more budget-friendly alternatives. + +Keep in mind that Nano Banana (non-Pro) may struggle during the look generation step, especially when the prompt includes multiple pieces of equipment or complex visual elements.