--- title: "Feed AI - Choosing the Right Model" url: "https://smartfeeds.productfruits.help/en/article/feed-ai-copy" language: "en" last_updated: "2026-11-06T14:41:54.510Z" --- # Feed AI - Choosing the Right Model Learn about the text and image AI models available in **Feed AI**, their technical capabilities, credit consumption, and how to pick the right model for your catalog needs. * * * ## Table of Contents 1. [Overview](https://smartfeeds.productfruits.help/en/article/feed-ai-copy#overview) 2. [Text Models: Which One to Choose](https://smartfeeds.productfruits.help/en/article/feed-ai-copy#text-models:-which-one-to-choose) 3. [Image Models: Which One to Choose](https://smartfeeds.productfruits.help/en/article/feed-ai-copy#image-models:-which-one-to-choose) 4. [Key Features & Capabilities](https://smartfeeds.productfruits.help/en/article/feed-ai-copy#key-features-&-capabilities) 5. [Quick Decision Guide](https://https//smartfeeds.productfruits.help/en/article/feed-ai-copy#key-features-&-capabilities) 6. [Related Articles](https://smartfeeds.productfruits.help/en/article/feed-ai-copy#key-features-&-capabilities) * * * ### Overview When creating a task in **Feed AI**, you choose between **Text** and **Image** processing. This selection is fixed at creation time and determines which models are available: - **Text tasks**: Rewrite titles, optimize descriptions, extract attributes, or categorize products. - **Image tasks**: Generate lifestyle staging, create product variations, or produce clean packshots. All models are available by default across all contracts. > **Note on credit estimation:** Before launching any task, SmartFeeds calculates an upfront credit estimate based on your catalog size, selected model, and input data. You are only charged for products successfully processed. If an item fails, no credits are deducted. ### Text Models: Which One to Choose | ***Model*** | ***Provider / Hosting*** | ***Estimated Credits (per 1,000 products)*** | ***Quotas & Speed*** | ***Best For*** | | --- | --- | --- | --- | --- | | **GPT-6 Luna** *(Default)* | OpenAI (Outside EU) | 🟢 **~2.0 credits** *(2.8 with image)* | ⚡ Up to 30,000 req/min | **Default choice for all standard catalogs and bulk rewrites.** | | **Mistral Small 3.1** *(New)* | Mistral / Vertex (EU) | 🟢 **~1.5 credits** *(2.4 with image)* | 🐢 300 req/min (~18k products/h) | **European data sovereignty at the lowest credit cost for small catalogs.** | | **Gemini 3.5 Flash Lite** | Google Vertex (EU) | 🟡 **~9.5 credits** *(11.8 with image)* | ⚡ Up to 10,000 req/min | **High-volume catalogs requiring strict EU data residency.** | | **Gemini 3.1 Flash Lite** | Google Vertex (EU) | 🟡 **~6.2 credits** *(8.1 with image)* | ⚡ Up to 10,000 req/min | Legacy tasks (prefer 3.5 for new setups). | *\* Estimates assume an average of 1,000 input tokens and 400 output tokens per product. Figures in parentheses include reading a 1024×1024 product image.* #### Recommendations: - **Default choice**: Use **GPT-6 Luna**. It offers the best performance-to-credit ratio and the highest throughput for large product feeds. - **European Data Sovereignty**: - - For sovereign EU hosting at the lowest credit consumption, pick **Mistral Small 3.1** (hosted on Vertex AI in Europe, data never leaves the EU). *Note: Due to lower rate limits (300 req/min), reserve it for small catalogs and straightforward text generation.* - For large catalogs requiring EU hosting with high throughput, choose **Gemini 3.5 Flash Lite**. ### Image Models: Which One to Choose | ***Model*** | ***Provider*** | ***Available Qualities*** | ***Estimated Credits (per 1,000 images)*** | ***Avg. Speed*** | ***Transparent BG*** | ***Best For*** | | --- | --- | --- | --- | --- | --- | --- | | **GPT Image 2.5 – Flare** *(Default)* | OpenAI | `low`, `medium`, `high` | 🟢 **~88 credits** *(medium)* | 13s – 22s | Yes | **Everyday catalog image generation & packshots.** | | **GPT Image 2.5 – Sunburst** | OpenAI | `low` to `max` | 🟡 **~351 credits** *(high)* | 18s – 120s | Yes | **Hero campaign visuals and high-visibility assets.** | | **Nano Banana 2** | Google Vertex | 1K, 2K, 4K | 🔴 **~672 credits** *(2K)* | Fast | No | **Native high-resolution outputs (2K/4K).** | #### Recommendations: - **Default choice**: Use **Flare** set to **`medium`** **quality**. It delivers fast rendering, optimized credit consumption (~88 credits for 1,000 images), and native background cutout support. - **Premium visuals**: Use **Sunburst** when maximum image fidelity is required (e.g., hero banners or small premium collections). - **High Resolution**: Use **Nano Banana 2** when native 2K/4K resolution is mandatory. ### Key Features & Capabilities `Feed AI` incorporates built-in technical safeguards and optimizations: #### 1\. Multimodal Vision & Input Fidelity All text models (including **GPT-6 Luna**, **Gemini**, and **Mistral Small 3.1**) and GPT-based image models can directly inspect source product images: - **Accurate details:** The model looks at your product photos (colors, materials, shapes) to write accurate descriptions without guessing. - **Credit cost:** Scanning a product photo adds a small cost (~0.8 extra credit per 1,000 products on GPT-6 Luna) #### 2\. Guaranteed Structured Output - **Text Tasks**: Formatting strictly matches your configured schema. Every requested attribute returns in its dedicated column with 100% reliability. - **Dual Output on Image Tasks (Flare & Sunburst)**: A single task execution can generate the new product image **and** extract accompanying text fields (such as image alt-text, color tags, or SEO labels) simultaneously. - *Note: Nano Banana 2 creates images only, without accompanying text fields.* #### 3\. Native Transparent Backgrounds Flare and Sunburst can automatically remove the background (transparent PNG). You get clean product cutouts without needing Photoshop or extra tools. #### 4\. Aspect Ratios & Variations - Supported formats: **Square (1:1 / 1024×1024)**, **Landscape (1536×1024)**, and **Portrait (1024×1536)**. - Generate between 1 and 3 visual variations per product in a single run. ### Quick Decision Guide - **Standard catalog rewriting at scale?** → Choose **GPT-6 Luna**. - **Strict EU data sovereignty on a budget?** → Choose **Mistral Small 3.1**. - **High-volume EU data residency?** → Choose **Gemini 3.5 Flash Lite**. - **E-commerce lifestyle staging or cutouts?** → Choose **GPT Image 2.5 – Flare** (`medium`). - **Top-tier hero campaign visuals?** → Choose **GPT Image 2.5 – Sunburst** (`high` or `max`). * * * ## Related Articles For further assistance with SmartFeeds, consider reviewing these articles: - [Join](https://smartfeeds.productfruits.help/en/article/join) - [Mapping](https://smartfeeds.productfruits.help/en/article/mapping) - [Feed AI : Getting Startted](https://smartfeeds.productfruits.help/en/article/feed-ai) For additional help, feel free to reach out via our [Contact Us](https://smartfeeds.productfruits.help/en/article/contact-us) page.