Smart Banner Resizer

Smart Banner Resizer — Composition-Aware Image Resize

What is Smart Banner Resizer?

Smart Banner Resizer is a composition-aware image-to-image model on Segmind. Provide one source image and a list of target dimensions, and the model returns a recomposed output for every size — not a naive crop or pad. It is powered by Nano Banana Pro with vision-LLM-guided prompting, so it understands subjects, backgrounds, text zones, and visual hierarchy before laying out each variant. The result is one source asset turned into many platform-ready sizes without losing brand or layout integrity.

Key Features

• •Single source, many sizes: pass 1–10 target dimensions in one call as <width>x<height> strings.
• •Mixed aspect ratios in a single request: 1:1, 4:5, 9:16, 16:9, and arbitrary custom dimensions all work together.
• •Composition-aware recomposition: keeps the subject centered, preserves decorative and brand elements, avoids awkward cropping.
• •1K, 2K, and 4K resolution tiers for drafts versus final hero assets.
• •Multiple variants per size (1–4) for A/B testing.
• •Output as png, jpeg, or webp.
• •Optional layout prompt and reproducible seed.

Best Use Cases

• •Performance marketing teams turning one product hero into Meta, Google, TikTok, and YouTube ad variants.
• •Agencies generating banner sets across IAB sizes from a single design.
• •E-commerce teams reformatting product photography for PDP, listings, and social.
• •Growth teams running A/B tests across creative variations and platforms.
• •Founders and content teams that need fast, on-brand resize without a designer in the loop.

Prompt Tips and Output Quality

Defaults (1K, png, no prompt, safety_tolerance=4) produce strong results for most ad imagery — confirmed on a 16:9 product hero with splash and confetti elements recomposed cleanly into both 1:1 and 16:9 outputs. The optional prompt is a layout hint, not a generation instruction; use it for things like "leave headline space on the left" or "keep logo top-right". For final delivery, set resolution to 2K; reserve 4K for hero assets. Generation is recomposition-heavy (~60–120s for a multi-size batch) — use a fire-and-poll pattern.

How to Use Smart Banner Resizer

Smart Banner Resizer turns a single source image into multiple banner-ready sizes through composition-aware recomposition rather than cropping. Use it when you need one product or campaign visual to ship across feed, story, banner, and display formats without manual layout work.

Quick Start

Pass image_url (a publicly reachable jpg / png / webp URL) and target_sizes (an array of <width>x<height> strings, 1–10 entries) to https://api.segmind.com/v1/smart-banner-resizer. Defaults handle everything else. The response is a JSON object listing every generated image with its dimensions, aspect ratio, and resolution tier.

Recommended Settings

Confirmed in pre-release testing on a 16:9 product hero (orange sneakers with water splash and confetti on a peach gradient) recomposed cleanly into both 1024×1024 and 1920×1080:

• •target_sizes: ["1024x1024", "1920x1080"] for square + widescreen; add "1080x1920" for stories and "1200x628" for LinkedIn / Facebook link cards.
• •resolution: 1K for drafts, 2K for production ads. 4K is charged double — reserve it for hero assets.
• •num_images_per_size: 1 for speed, 2–4 if you want A/B options.
• •output_format: png for transparency or quality, jpeg for ad-server uploads, webp for the web.
• •prompt: leave empty for clean recomposition. Use it as a layout hint only — e.g. "reserve a headline strip on the left", "keep logo top-right".
• •seed: set while iterating on prompt; unlock for final batches.

Best Practices

1. •Use a clean hero shot — uncluttered backgrounds recompose most reliably.
2. •Group similar aspect ratios in one call for consistent treatment; run outliers (e.g. very tall vs. very wide) separately.
3. •For text-heavy creatives, anchor brand text with an explicit prompt rather than relying on default detection.
4. •Run many target sizes per call to keep cost predictable and reduce orchestration overhead.

Performance

Multi-size generation takes ~60–120 seconds depending on resolution and variant count. The Segmind sandbox times out synchronous calls at 45s, so use the Request History fire-and-poll pattern for any production integration. Do not expect sub-15s synchronous responses for batches.

Does it preserve text in my source image?

Text is treated as a visual element. For text-heavy creatives, anchor it with an explicit prompt.

Can I get multiple options per size?

Yes — set num_images_per_size to 2–4.

Which aspect ratios are supported?

Any <width>x<height> pair you pass, including mixed ratios in a single call.

Is the output deterministic?

Set seed to lock recomposition while iterating on prompt.

How many sizes can I request at once?

Up to 10 per call.

What input formats work best?

Clean hero shots in jpg, png, or webp recompose most reliably.