AI copy modes: shared copy vs per-creative vision

The AI text dialog has two generation modes. The mode you pick decides whether one piece of copy is reused across all your ads or whether the AI writes tailored copy for each individual creative. This article explains both and when to use each.

For the full walkthrough of opening the dialog and filling the brief, start with How to generate ad copy with AI.

Where you choose the mode

At the top of the AI text dialog there's a Mode panel with two tiles:

• One copy for all creatives - the default. A single generation is produced and reused for every ad in the launch.
• Different copy per creative (Vision) - the AI looks at each selected creative and writes copy tailored to what's in that image.

The Vision tile unlocks when you've selected at least one video or an image with a provider-accessible preview in the creatives step. Videos do not need an external preview URL because the provider never receives their bytes. If nothing eligible is selected, the tile stays disabled with a hint to pick creatives first. A run is capped at 20 eligible creatives; the dialog blocks generation and explains the limit before any provider calls are made.

Mode 1: One copy for all creatives (shared)

This is the fast, cheap path and the right choice most of the time. uplads runs your brief once and produces a set of variants that get applied to every ad.

Use shared mode when:

• Your creatives all sell the same product or offer the same angle.
• You want the copy to vary independently of the image (Meta's optimizer mixes texts and creatives anyway).
• You want to keep cost and generation time low - it's a single call.

The whole brief still applies here: product, audience, pain points, value props, tone, variety, language, CTA hint and the per-field variant counts. After generating, you pick which variants to apply on the review screen.

Mode 2: Different copy per creative (Vision)

In Vision mode the AI actually looks at each image and writes copy that references what it sees. It makes one call per creative, so it costs roughly that many times a normal generation - the dialog shows this cost warning before you generate.

Use Vision mode when:

• Each creative shows a different product, scene or angle that the copy should speak to.
• You want headlines and primary text that match the specific image, not generic copy.
• The extra cost is worth tighter image-to-copy alignment.

A few things to know:

• Your selected model must support vision (for example GPT-4o, GPT-4.1, or Claude Sonnet/Opus). GPT-4o mini and Claude Haiku also work.
• Videos can't be read frame-by-frame yet. Video creatives fall back to brief-only copy using the filename as a hint, and they're clearly flagged as such in the review.
• Images whose authenticated or relative preview cannot be opened by the external provider are excluded up front and use shared copy instead.
• If some creatives fail to generate, the others still succeed - you get a warning rather than a failed run.

The result review uses the same master/detail pattern as the carousel builder: select a creative in the left sidebar, then see its media and exact copy in the right pane. A video can be played there when its preview is available, and its brief-plus-filename fallback is labeled explicitly.

When you apply Vision results, each creative keeps its own copy on launch. Switch between creatives in the Text step to view and edit each one; the final review and launch worker resolve the matching copy for that creative.

The brief and tuning knobs apply to both modes

Whichever mode you choose, the same controls shape the output:

• Brief fields - what you're marketing, target audience, pain points, and value propositions / USPs / proof. You can also auto-fill these from a landing page URL.
• Campaign objective - the app automatically chooses an internal sales, lead, app or awareness strategy. There is no prompt-template choice to configure in the launch dialog.
• Tone - Professional, Casual, Urgent, Witty, Empathetic, Bold, or Story-driven.
• Variety - Tight, Balanced, Diverse, or Wild, controlling how different the variants are from each other.
• Language - English, Deutsch, Español, Français, Italiano, Português, Nederlands or Polski.
• CTA hint - biases copy toward a Meta call-to-action verb, or leave it at No preference.
• Variant counts - set 0-5 for primary texts, headlines and descriptions independently.

Your saved Brand context and voice are applied in both modes. The product and audience selected in the dialog stay identical across the run; Vision then adapts each bundle to its creative.

Which should I pick?

Start with shared mode - it's cheaper, faster, and good enough for most launches where the creatives share an offer. Reach for Vision when your images genuinely differ and you want the copy to call out what's in each one, and you're comfortable paying per creative.

To watch spend across generations, see How to track AI usage and estimated cost and How to set a monthly AI spend cap. If a generation is blocked or failing, see Why is AI copy generation blocked or failing?.