AI Agent Ad Creative Workflow: Generate Product Images, UGC Assets, and Video Variants with Restory

Direct answer: a creative agent can use Restory to generate ad assets by discovering available models, checking credits, producing product images, editing reference shots, creating UGC-style visuals, animating winning frames into videos, and returning completed CDN URLs for review or publishing.

The ad creative bottleneck is no longer one hero image. It is volume with control. Growth teams need hooks, angles, thumbnails, before/after frames, UGC variants, product demo stills, and short videos. Restory gives agents a programmable media layer so they can create those assets without waiting for a designer to manually export every variation.

Last verified: June 11, 2026. Check capabilities JSON before selecting models, settings, reference limits, variants, and credit costs.

The Creative Agent Stack

A useful ad agent has four layers:

Strategy layer: offer, audience, pain point, hook, visual angle.
Production layer: Restory image generation, restoration, edits, and video generation.
Verification layer: credit check, job polling, failed-job handling, asset manifest.
Distribution layer: handoff to Postiz, ad platforms, CMS tools, or internal review.

Restory sits in the production and verification layers. The agent can still use its own reasoning for hooks and campaign logic, but the media operations become API calls with structured outputs.

The Workflow, Step by Step

1. Turn the Campaign Brief into Asset Types

Start by forcing the agent to produce an asset plan. For one product launch, a practical batch might be:

Asset	Purpose	Suggested Restory workflow
Product hero	Landing page and ad thumbnail	Image generation
UGC still	TikTok and Reels hook	Image generation or reference edit
Problem/solution visual	Carousel and static ad test	Image generation
Reference edit	Keep product identity, change scene	Image generation with reference
Motion ad	Short video variant	Video generation from image

2. Check Capabilities and Credits

Before batch work, agents should confirm what they can run and how much budget remains.

npm install -g @restory/cli
export RESTORY_API_KEY="rst_..."
restory capabilities
restory credits

Direct API clients can use:

GET /api/agent/capabilities
GET /api/agent/credits
Authorization: Bearer rst_...

This makes the batch predictable. If the agent sees too few credits for twenty video variants, it should propose a smaller test batch instead of failing halfway through.

3. Generate Product Hero Assets

For static ad tests, generate clean product visuals with explicit composition and text-safe areas.

restory image:generate \
  --prompt "premium product photo of a matte black travel mug on a stone counter, morning light, clean ecommerce composition, text-safe right side" \
  --model nano-banana-2 \
  --resolution 2K \
  --aspect-ratio 4:5

The agent should vary one element at a time: background, lighting, framing, CTA area, or audience context. Random prompt churn makes performance attribution impossible.

4. Generate UGC-Style Ad Frames

UGC-style visuals work better when they feel like a real scene rather than a perfect product render. Ask the agent for smartphone framing, natural light, simple gestures, and a clear first-second hook.

restory image:generate \
  --prompt "vertical UGC ad still, creator holding a skincare bottle near bathroom mirror, natural morning light, authentic smartphone framing, text-safe top third" \
  --model nano-banana-2 \
  --resolution 2K \
  --aspect-ratio 9:16

For a larger test, produce three hooks:

Problem hook: "I almost returned this."
Transformation hook: "Seven days later..."
Proof hook: "The texture is what surprised me."

5. Use Reference-Based Edits for Product Consistency

When the product identity matters, upload or pass a reference image and instruct the agent to preserve the product while changing the setting.

restory image:generate \
  --prompt "keep the product identical, place it in a bright kitchen counter scene, realistic shadows, clean ad composition" \
  --model gpt-image-2 \
  --quality high \
  --image-size 1024x1024 \
  --reference ./product.png

Local references are uploaded first and passed to Restory as media URLs. The agent does not need to manually host files elsewhere.

6. Animate the Winning Static into Short Video

Once a static frame has strong composition, create one or two video variants. For an image-to-video ad:

restory video:generate \
  --prompt "slow push in, subtle hand movement, premium UGC product ad, natural bathroom light, keep product label readable" \
  --model seedance-2-0 \
  --variant fast-image \
  --image ./winning-static.png \
  --duration 8 \
  --resolution 720p

Poll the video record until it completes:

restory jobs:get <videoRecordId> --type video

7. Return an Asset Manifest

The most useful output is not a paragraph saying "done." The agent should return a structured manifest with every completed URL and campaign note.

{
  "campaign": "skincare launch",
  "assets": [
    {
      "type": "image",
      "angle": "product hero",
      "recordId": "imageRecordId",
      "resultUrl": "https://restory.b-cdn.net/images/...",
      "recommendedUse": "Meta 4:5 static ad"
    },
    {
      "type": "video",
      "angle": "UGC mirror hook",
      "recordId": "videoRecordId",
      "resultUrl": "https://restory.b-cdn.net/videos/...",
      "recommendedUse": "TikTok/Reels 8s test"
    }
  ]
}

Copy-Paste Prompt for an Ad Creative Agent

Act as a performance creative agent using Restory for media generation. Read capabilities and credits first. Build a small asset plan for the campaign brief. Generate product hero images, UGC-style vertical frames, and one short video from the strongest frame. Poll all jobs until completed. Return a JSON-style asset manifest with result URLs, prompts, model settings, and recommended ad use. Do not expose API keys.

Creative Batch Template

Batch	Count	Goal
Static hooks	3	Find one visual angle worth animating
Product scenes	3	Test background and context
UGC frames	3	Find platform-native first frames
Video variants	1-2	Animate the strongest still

Common Failure Modes

No discovery: the agent uses stale settings and fails validation.
No credit check: the batch starts but cannot finish.
No manifest: completed assets are hard to review or schedule.
Too many variables: every asset changes style, hook, background, and format at once.
No polling: the agent reports before final URLs exist.

FAQ: AI Agent Ad Creative on Restory

Can Restory generate assets for ads?

Yes. Agents can generate product images, UGC-style frames, reference edits, restoration assets, and short video variants through Restory workflows.

Can the agent use local product images?

Yes. The CLI uploads local reference files when needed and passes Restory CDN URLs into generation endpoints.

Should an agent generate images or videos first?

Usually images first. Static frames are cheaper and faster to evaluate. Animate the strongest image into video after the hook and composition are clear.

How should teams keep creative tests organized?

Use an asset manifest with prompt, model, settings, record ID, result URL, campaign angle, and recommended channel.

Can completed URLs go into a scheduler?

Yes. Completed Restory jobs return stable CDN URLs that a publishing agent can pass to Postiz, CMS workflows, ad review systems, or download steps.

Conclusion

The best ad creative agents are structured operators. They do not ask for random art. They create a batch plan, discover model capabilities, check credits, generate controlled variants, poll every job, and return usable asset URLs.

Start here: Restory agent recipes
Agent stack: Inside the Restory AI Agent Stack
TikTok slides: Automate TikTok Slides with AI Agents
Image generation: /app/image-generation
Video generation: /app/video-generation