Creative production

The production cost problem

A 30-second video ad produced the traditional way involves scripting, storyboarding, location scouting, talent booking, filming, editing, colour grading, sound design, and delivery. The budget: $10,000 to $50,000 for a mid-quality spot. The timeline: 4–8 weeks from brief to final cut.

For brands that need volume — social media content, product demos, localised variants, A/B test creative — traditional production doesn’t scale. A direct-to-consumer brand running ads across Instagram, TikTok, YouTube, and Meta might need 50–100 creative variants per month. At traditional production costs, that’s a multi-million dollar annual budget.

According to Wistia’s 2025 State of Video report, 91% of businesses use video as a marketing tool, but 30% cite production costs as their biggest barrier to producing more. AI creative tools are eliminating that barrier.

The AI creative toolkit

Image generation

Midjourney is the gold standard for AI image generation. It produces remarkably high-quality images from text descriptions, with fine control over style, composition, and mood. A prompt like “Product photograph of a ceramic coffee mug on a marble counter, morning light, shallow depth of field, editorial style” generates images that are nearly indistinguishable from professional product photography.

Midjourney is used by ad agencies, e-commerce brands, and publishers for:

• Product concept visualisation — Seeing what a product might look like before manufacturing
• Ad creative — Hero images for social ads, banner ads, and landing pages
• Editorial illustration — Article headers, social media graphics, presentation visuals
• Mood boards and style exploration — Rapidly iterating on creative direction

A case study from an e-commerce brand: BarkBox, the dog subscription service, used AI-generated images for social ad creative testing. They generated 200 image variants in a day, tested them against their traditionally photographed ads, and found that 40% of AI-generated concepts outperformed the professional photos in click-through rate.

DALL-E 3 (via OpenAI) is Midjourney’s closest competitor, integrated into ChatGPT and available via API. It’s particularly good at following detailed instructions and rendering text within images.

Adobe Firefly is the enterprise-safe option, trained exclusively on licensed Adobe Stock images. This is important for brands with legal teams concerned about copyright — Firefly comes with IP indemnification.

Leonardo.ai and Ideogram are strong alternatives for specific use cases: Leonardo excels at game and fantasy art styles, while Ideogram is best-in-class for generating images that include readable text.

Video generation and editing

Runway is the most versatile AI video tool. Its capabilities include:

• Text-to-video — Generate short video clips from text descriptions. “Aerial drone shot of a tropical beach at sunset, slow camera movement” produces a photorealistic 4-second clip.
• Image-to-video — Animate a still image. Take a Midjourney product photo and add subtle motion — liquid pouring, fabric flowing, camera movement.
• Video-to-video — Transform existing footage. Change the style (make live action look like animation), swap backgrounds, or extend clips.
• Inpainting and removal — Remove objects or people from video, fill in backgrounds.

A real example: a fitness brand producing social content used Runway to generate b-roll footage for their workout videos — overhead shots of gym equipment, smooth transitions between exercises, ambient gym environment clips. Previously, they sent a videographer to a gym for a full-day shoot ($2,000–$3,000). With Runway, they generated equivalent clips in an afternoon for the cost of their subscription.

Kling AI and Luma Dream Machine are strong competitors in the text-to-video space, each with different strengths. Kling produces particularly natural human movement, while Luma excels at cinematic camera movement.

Synthesia takes a different approach: AI avatar videos. You type a script, choose a digital presenter (or create a custom avatar from a short recording), and the system generates a video of the avatar delivering your script with natural lip sync and gestures. This is popular for:

• Training and onboarding videos
• Product demos and walkthroughs
• Internal communications
• Localised versions (same avatar, different languages)

Synthesia reports that over 50,000 companies use their platform, including 35% of Fortune 100 companies. Their case study with Zoom shows the company using AI avatars for help centre videos in 10 languages, reducing production time from weeks to hours per video.

HeyGen is Synthesia’s main competitor, with stronger real-time translation capabilities — it can dub an existing video into another language while matching the speaker’s lip movements.

Voice and audio

ElevenLabs has become the default tool for AI voice generation. Its capabilities include:

• Text-to-speech — Generate natural-sounding narration in 30+ languages with dozens of voice options
• Voice cloning — Record a few minutes of your own voice and create a digital version you can use to generate unlimited content (with consent protocols)
• Voice design — Create entirely new synthetic voices with specific characteristics (age, accent, tone)
• Dubbing — Translate and re-voice existing video content while preserving the speaker’s vocal characteristics

A podcast production company used ElevenLabs to expand their English-language shows into Spanish, Portuguese, and German markets. Instead of hiring voice actors in each language, they created voice clones of their hosts (with permission) and generated translated episodes. The quality was good enough that listener surveys couldn’t reliably distinguish the AI-dubbed episodes from human-narrated ones.

Murf AI is a competitor focused on professional voiceover use cases — e-learning, corporate presentations, and advertisements.

For music and sound design, Udio and Suno generate original music from text descriptions. “Upbeat corporate background music, 120 BPM, acoustic guitar and light percussion, 60 seconds” produces a usable track in seconds. This replaces the process of searching stock music libraries or commissioning custom compositions.

Descript bridges video and audio editing with AI — it lets you edit video by editing the transcript text, remove filler words automatically, and generate studio-quality audio from rough recordings.

Post-production and enhancement

Topaz Labs uses AI to upscale video resolution, remove noise, and stabilise shaky footage. A 720p clip can be upscaled to 4K with remarkable quality. This is useful for repurposing older content or improving footage shot on phones.

How the production workflow comes together

A typical AI-assisted creative workflow for a product video ad:

1. Script and concept — The creative director writes the concept and script. Claude or ChatGPT can help brainstorm angles, draft scripts, and write variant copy for A/B testing.

2. Visual asset creation — Midjourney generates product imagery and lifestyle shots. Multiple concepts are generated in parallel — 20–30 variants in an hour.

3. Motion and video — Selected images are animated in Runway to create video clips with camera movement and subtle motion. Additional b-roll clips are generated from text prompts.

4. Voiceover — The script is narrated using ElevenLabs. Multiple voice options are generated and reviewed. If the ad runs in multiple markets, localised voiceovers are produced simultaneously.

5. Music and sound — Background music is generated via Udio to match the ad’s mood and pacing. Sound effects are added from AI-generated or stock libraries.

6. Editing and assembly — A human editor assembles the final cut in Premiere Pro or DaVinci Resolve, adding transitions, timing adjustments, colour grading, and text overlays. This step is still fundamentally human — AI generates the raw materials, but the creative assembly requires editorial judgment.

7. Format adaptation — The final ad is exported in all required aspect ratios and durations: 9:16 for TikTok/Reels, 1:1 for feed posts, 16:9 for YouTube, and 15/30/60 second cuts.

Total timeline: 24–48 hours from script to final delivery. Total cost (excluding subscriptions and editor time): under $500 for tool usage, compared to $15,000–$30,000 for equivalent traditional production.

When AI creative works — and when it doesn’t

Where AI excels

• Performance marketing creative — Ads optimised for clicks, conversions, and engagement. Volume and iteration speed matter more than individual artistic perfection.
• Social media content — Short-form, high-volume content that has a short shelf life. The cost of traditional production per piece is hard to justify.
• Product visualisation — Showing products in contexts, lifestyles, and environments without physical photoshoots.
• Internal content — Training videos, presentations, documentation. The audience doesn’t expect blockbuster production quality.
• Prototyping and concepting — Testing creative directions rapidly before committing to full production.

Where AI still falls short

• Nuanced human emotion — AI-generated faces still occasionally fall into uncanny valley territory, especially in close-ups.
• Complex physical interactions — Characters manipulating objects, sports footage, dance choreography. Physics remains inconsistent.
• Brand campaigns — Hero-level brand content where artistic vision, cultural sensitivity, and emotional resonance are paramount. A Super Bowl spot still needs human creative direction end to end.
• Talent-driven content — Influencer content, testimonials, and user-generated content require real humans by definition.

The pragmatic approach: use AI for the 80% of creative that needs to be good and produced quickly, and invest traditional production budgets in the 20% that needs to be exceptional.

Tools referenced in this guide

• Midjourney — AI image generation for marketing and product visuals
• Runway — AI video generation, editing, and effects
• ElevenLabs — Voice generation, cloning, and multilingual dubbing
• Udio — AI music generation
• Suno — AI music and song generation
• Synthesia — AI avatar video creation
• HeyGen — AI video translation and avatar generation
• DALL-E 3 — OpenAI’s image generation model
• Adobe Firefly — Enterprise-safe AI image generation
• Kling AI — Video generation with natural motion
• Luma Dream Machine — Cinematic AI video generation
• Descript — AI-powered video and audio editing
• Topaz Labs — AI video upscaling and enhancement