AI Agents for Image and Video Generation
by DeepLearning.AI
Build media agents that generate images and video, score their own output, and retry when it misses.
Overview
AI Agents for Image and Video Generation is a DeepLearning.AI short course built in partnership with Google and taught by Katie Nguyen, a Developer Relations Engineer at Google Cloud AI, and Wafae Bakkali, a Staff Generative AI Specialist at Google. It runs about 1 hour 34 minutes across ten lessons with six code examples and one graded quiz. The organising idea is that visual media has no single correct output, so evaluation gets as much attention as generation. Early lessons cover generative media architectures, then prompt engineering for image generation and for video generation as two separate lessons, because video adds temporal consistency and motion that a good still-image prompt will not produce. The evaluation lesson introduces three complementary techniques used together: SigLIP image-text similarity scoring, LLM-as-judge, and structured rubrics. The final third assembles those pieces into working agents — an image agent that turns brand guidelines into mockups, and a video agent that plans a multi-scene explainer, generates reference frames, animates them with audio and evaluates temporal consistency across scenes. Generation runs on Google's Nano Banana for images and Veo for video, with Gemini CLI used to package the agent's skills. The transferable part is the loop itself: generate, score automatically, and iterate when the score misses, which applies regardless of which image or video model you swap in.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Course
- Cost
- Freemium
- Duration
- ~1 hour 34 minutes, self-paced (10 lessons, 6 code examples, 1 graded quiz)
- Provider
- DeepLearning.AI
- Hands-on
- Yes — code/exercises
- Certificate
- Available
What You’ll Learn
- ✓Write prompts for image generation and, separately, for video generation
- ✓Score image-text alignment automatically using SigLIP similarity as a machine evaluation signal
- ✓Design LLM-as-judge and structured rubric evaluators for subjective visual output
- ✓Build an image agent that converts brand guidelines into usable product mockups
- ✓Build a video agent that plans multi-scene explainers and animates reference frames with audio
- ✓Evaluate temporal consistency across scenes rather than judging each frame in isolation
- ✓Close the generate-evaluate-retry loop so the agent iterates without human review
Highlights
- •Evaluation gets equal billing with generation, which is rare in generative-media tutorials
- •Three stacked evaluation techniques (SigLIP, LLM judge, rubrics) instead of one subjective eyeball check
- •Taught by two Google engineers working directly on the models the course uses
- •Ends with a working multi-scene video agent rather than a single-image demo
- •Extends the agentic patterns most engineers only know from text into visual media
Who It’s For
Best For
- ✓AI engineers extending existing agent workflows beyond text into images and video
- ✓Developers automating marketing, product-demo or explainer asset production
- ✓Teams that need automated quality gates on machine-generated visual output
Prerequisites
- •Familiarity with Python
- •Basic experience calling LLM APIs from code
- •No computer vision, video editing or design background required
FAQ
What is AI Agents for Image and Video Generation?
A DeepLearning.AI short course built with Google on extending agentic workflows past text into visual media. You learn prompt engineering for images and video separately, three stacked automatic evaluation techniques, and then combine them into agents that iterate on their own output. Afterwards you can build an agent that plans a multi-scene explainer video, generates it, and checks consistency without a human in the loop.
Is AI Agents for Image and Video Generation free?
AI Agents for Image and Video Generation offers free content, with paid options for certificates or premium features.
What level is AI Agents for Image and Video Generation for?
AI Agents for Image and Video Generation is aimed at a intermediate audience. Recommended background: Familiarity with Python, Basic experience calling LLM APIs from code, No computer vision, video editing or design background required.
How long does AI Agents for Image and Video Generation take?
Expect roughly ~1 hour 34 minutes, self-paced (10 lessons, 6 code examples, 1 graded quiz). Most learners work through it at their own pace.
What will I learn from AI Agents for Image and Video Generation?
You'll learn: Write prompts for image generation and, separately, for video generation; Score image-text alignment automatically using SigLIP similarity as a machine evaluation signal; Design LLM-as-judge and structured rubric evaluators for subjective visual output; Build an image agent that converts brand guidelines into usable product mockups; Build a video agent that plans multi-scene explainers and animates reference frames with audio; Evaluate temporal consistency across scenes rather than judging each frame in isolation; Close the generate-evaluate-retry loop so the agent iterates without human review.
Topics
Sources
This page was written from 2 sources, 1 on domains other than deeplearning.ai.