AI for Out-of-Home Advertising: Print, DOOH, and 3D
How AI produces OOH creative at scale: print-resolution stills via upscaling, DOOH motion loops, the anamorphic 3D billboard trend, and localized creative. Specs and cost math.
Out-of-home advertising has two hard requirements that trip up most AI image workflows: resolution and scale. A billboard printed at 14 by 48 feet needs source art that holds up at sizes where a 1024-pixel generation would look like mud, and a national DOOH buy needs the same creative localized across dozens of markets without a photoshoot per city. Both problems are now tractable. This guide covers the four OOH workflows worth building on AI, the model routing behind each, the resolution specs that actually matter, what it costs, and the mistakes that put a distorted hand on a 48-foot billboard.
TL;DR
- Print resolution is a solved problem. Generate the composition at native model resolution, then upscale with Topaz or Clarity to the print size. A 4K generation upscaled 4x lands well above the DPI a large-format print needs at viewing distance.
- DOOH is where motion pays off. Digital screens run short looping video. A 6-to-10-second loop generated in Kling 3.0 or Veo 3.1 costs a few dollars and refreshes as often as you want, versus a fixed printed panel you commit to for a full flight.
- Localization at scale is the biggest unlock. One master concept, regenerated per market with local language, landmarks, and casting, turns a national campaign into a fifty-market campaign without fifty shoots.
The 4 OOH workflows
1. Static print at large-format resolution
The core static-OOH move is to generate the composition at the highest native resolution your image model supports, then upscale to the physical print size. Large-format print is not as DPI-hungry as people assume, because viewing distance does the work: a highway billboard is designed to read from hundreds of feet, so it prints at 10 to 15 DPI at final size, while a bus shelter or transit poster seen from a few feet wants closer to 100 to 150 DPI. Either way, you are upscaling a generation, not shooting a gigapixel plate.
The workflow on 8frame: compose the still in Nano Banana Pro or Flux 1.1 Ultra, the two highest-fidelity image models on the canvas. Get the composition, product accuracy, and text-safe zones right at native resolution. Then run the approved frame through the Clarity upscaler (or Topaz for the heaviest enlargements) to reach print dimensions. For a bus shelter poster at 4 by 6 feet at 120 DPI, you need roughly a 6,000-pixel long edge, comfortably reached by a 4x upscale of a high-res generation.
Keep live text out of the generation. Models still mangle small type, and OOH type has to be pixel-clean. Generate the image, then set the headline and legal in a real layout tool over the top.
2. DOOH motion loops
Digital out-of-home screens play short video loops, which is exactly what the video models produce. A 6-to-10-second loop of product motion, an environmental scene, or a simple animated logo reveal gives a digital panel the movement that makes it out-perform a static creative in the same slot.
Model pick: Kling 3.0 for fast, clean environmental and product loops, Veo 3.1 when you want cinematic weight and the higher resolution holds up on a large LED wall. Generate at 16:9 for landscape screens or a vertical ratio for portrait street furniture, and design the clip to loop, hold the first and last frame close so the repeat is seamless. A tested approach: generate a 5-second clip, mirror it, and cross-fade for a 10-second breathing loop with no visible seam.
Loop generation cost is a few dollars per screen format, which means you can run a different loop per daypart or per screen network without a production line item per version.
3. Anamorphic and 3D billboard concepts
The forced-perspective 3D billboard, the illusion that something is breaking out of the screen, has been the most-shared OOH format of the last few years. Verified examples give you the reference vocabulary: Maybelline's London Underground campaign, where a tube train appeared to sweep mascara across giant lashes, became one of the most talked-about OOH stunts of 2023; Spirit Halloween's Times Square execution had a character smashing through the digital frame and reaching toward the street; the Charles Schwab campaign had a robot appearing to climb out of the board toward the intersection below.
AI's role here is pre-visualization and content generation, not the final anamorphic warp. The anamorphic effect is a mathematical distortion mapped to a specific screen and a specific viewing angle, and that mapping is a specialist DOOH production step. What AI accelerates is everything before it: generating the hero object and its "breaking out" motion, iterating the concept with the client, and producing the environmental plates. Generate the 3D hero action as a clean clip on the canvas, get client sign-off on the idea cheaply, then hand the approved asset to the anamorphic mapping vendor. You compress the expensive part, concepting and revisions, from weeks to an afternoon.
4. Localized OOH at scale
A national OOH buy usually means one master creative stretched thin across every market. AI makes true localization affordable: regenerate the master concept per market with the local language headline, a recognizable local landmark, or locally representative casting, from one composition.
The workflow: build the master still on 8frame and save it as a reusable workflow template. For each market, swap the prompt variables, the background landmark, the on-model casting, the seasonal cue, and regenerate. Because the composition and lighting are locked in the template, the outputs stay on-brand while the local specifics change. Run the print-upscale pass on each finished market version. Fifty market variants is a batch job, not fifty projects.
Model routing for OOH
| Workflow | Primary model | Secondary | Why |
|---|---|---|---|
| Static print composition | Nano Banana Pro | Flux 1.1 Ultra | Highest still fidelity; Flux for print-res detail |
| Print upscaling | Clarity upscaler | Topaz Video AI | Clean 4x enlargement to billboard/transit size |
| DOOH landscape loops | Kling 3.0 | Veo 3.1 | Fast clean loops; Veo for large-LED cinematic |
| 3D/anamorphic hero asset | Veo 3.1 | Kling 3.0 | Cinematic motion for the breakout action |
| Localized market variants | Nano Banana Pro | Flux 1.1 Ultra | Template-driven batch regeneration |
Resolution spec table
| OOH format | Typical physical size | Working DPI at distance | Pixel long-edge target |
|---|---|---|---|
| Highway bulletin | 14 x 48 ft | 10-15 DPI | ~7,000-8,600 px |
| Transit/bus shelter | 4 x 6 ft | 100-150 DPI | ~6,000-9,000 px |
| Street furniture (digital) | Portrait HD/4K screen | Native screen res | 1080 x 1920 or 2160 x 3840 |
| Large LED spectacular | Varies (pixel pitch driven) | Match screen native | Per vendor spec sheet |
Always get the exact spec sheet from the OOH vendor before you finalize. Pixel pitch on digital screens and printer requirements on static vary enough that a generic target will occasionally under- or over-shoot. The rule: compose at native, upscale to their number, never generate directly at billboard pixel dimensions.
Unit economics
A traditional large-format OOH creative, photoshoot, retouching, and layout, runs $5,000 to $30,000 for a national-quality static campaign, more if it involves talent and locations. A DOOH motion spot produced traditionally adds a shoot and post budget on top.
AI generation on 8frame:
- Static print composition (Nano Banana Pro + Clarity upscale): $0.10 to $0.40 per finished panel in compute
- DOOH loop (Kling 3.0, two format ratios): $0.60 to $1.60 per loop
- 3D/anamorphic hero asset (Veo 3.1, iterated): $3 to $10 in concepting compute, before the specialist mapping step
- 50 localized market variants (template batch): roughly $5 to $20 total in compute
The specialist steps that remain, the anamorphic warp, the final large-format print production, the vendor spec compliance, are where your budget goes. The creative development that used to dominate the timeline and cost is now the cheap part.
3 mistakes that put a bad panel on a billboard
1. Generating live text into the image. Models still corrupt small type. Every OOH headline, price, and legal line must be set as real text in a layout tool over the generated image, never generated inside it. This is non-negotiable for print, where a broken letterform is visible at 48 feet.
2. Upscaling a weak composition. Upscaling adds resolution, not information. A generation with a distorted hand, a warped product, or a flawed perspective gets bigger, not better. Fix every artifact at native resolution and approve the composition before the upscale pass. The billboard magnifies your mistakes.
3. Ignoring the vendor spec sheet. OOH vendors have exact requirements for pixel dimensions, color profile, bleed, and file format that vary by site. Producing at a generic 4K and hoping it fits is how a campaign gets bounced the week of the flight. Pull the spec sheet first, then work backward to your generation and upscale targets.
FAQ
Can AI-generated images really print at billboard size?
Yes. Large-format OOH prints at low DPI because it is viewed from a distance, so a high-resolution generation upscaled 4x with Clarity or Topaz comfortably exceeds what a highway bulletin needs at final size. The constraint is not raw resolution, it is composition quality: upscaling magnifies both detail and flaws, so the discipline is to perfect the image at native resolution before enlarging. Set live text separately in a layout tool, since models still mangle small type.
Can AI make the 3D "breaking out" anamorphic billboards?
Partly. AI handles the concepting and content, generating the hero object and its breakout motion, iterating with the client, and producing environmental plates, which is the expensive, slow part of those campaigns. The anamorphic effect itself is a mathematical distortion mapped to a specific screen and viewing angle, and that mapping is a specialist DOOH production step done by the screen vendor or a 3D studio. So AI compresses the creative development, and you hand off the approved asset for the final warp.
How does AI help localize OOH across many markets?
Build the master creative once as a reusable template with the composition and lighting locked, then regenerate per market by swapping only the local variables: language headline, a recognizable landmark, seasonal cue, or locally representative casting. Because the template holds the brand-defining elements constant, every market output stays on-brand while carrying local relevance. Fifty variants becomes a batch job costing a few dollars in compute, versus fifty separate photoshoots.
OOH is a resolution-and-scale problem, and both are now cheap to solve. Compose on the canvas, upscale to spec, and localize by template. Start an OOH project on 8frame with every leading image and video model in one place, or browse reusable workflow templates to batch a national buy. For the digital-screen cousin of this format, see AI for CTV ads.