- LTX-2.3 shifts stock footage from "close-enough" library matches to custom-generated b-roll — exact clip, exact aspect ratio, at a cost per clip that beats premium libraries once a campaign needs ~10+ clips.
- The workflow relies on prompt templates per category (lifestyle, urban, nature, corporate, product), batch generation, and disciplined tagging/version control so a library cuts together consistently.
- Commercial use runs through the LTX-2 Community License Agreement (free under $10M revenue, licensed above); traditional stock still wins for real public events, non-negotiable releases, or very small clip counts.
Agencies and producers used to license stock footage because filming the right clip cost more than buying a close-enough one. With LTX-2.3, the math has shifted: you can now generate the exact clip you need, in the exact aspect ratio you need, in roughly the time it takes to search a stock library.
The result is not "AI alternative to stock footage." The result is custom b-roll on demand, which is a different product entirely.
This guide is for the agency producer or post-production lead who needs to operationalize that shift.
It covers the LTX-2.3 stock-footage workflow, prompt templates for the b-roll categories agencies use most, the licensing terms that matter for client deliverables, and where AI generation does and does not beat the traditional library.
Prerequisites: Access to LTX-2.3 via the hosted LTX API or a self-hosted open-source deployment. Familiarity with brief intake and clip-library workflows. A working understanding of your clients' delivery specs.
Why AI stock footage now beats licensed libraries for many briefs
The shift is not about quality alone. It is about three concrete agency pain points that AI stock footage solves directly.
Specificity beyond "close enough" matches
Stock libraries are built around the most-licensed clips, not around your brief. You need a wide shot of a young woman walking through a Tokyo neighborhood at dawn carrying a paper coffee cup, and you get fifteen variants that are all almost right.
LTX-2.3 generates the exact clip you described from a text prompt or an image reference. The brief becomes the asset.
Brand-safety and exclusivity
A clip your client paid to license is also licensed by their competitors. For brand campaigns where exclusivity matters, this is a real problem. AI-generated footage produced for a specific brief is, by construction, unique to that production. That is a category change for brand work, not a feature.
Cost per clip at scale
A single licensed clip from a premium library can run $200 to $800. A campaign that needs forty b-roll inserts is a real budget line.
The LTX-2.3 API pricing is per-second, with the documented schedule available at the live pricing endpoint, so a four-second clip at 1080p has a known cost that is materially lower than premium stock per clip. The economics change again at agency scale.
The LTX-2.3 stock-footage workflow
Generating one good clip is straightforward. Generating a campaign's worth of consistent, brand-safe, deliverable clips is a workflow problem. The shape of that workflow is predictable.
Brief intake: converting a creative brief into prompts
The first job is translation. A brief like "lifestyle b-roll for a wellness brand, warm and natural, urban setting" is a creative concept, not a prompt. The translation work is to convert that into a clip list with concrete prompts.
Each clip gets a one-line description (the asset), a prompt block (what goes into LTX-2.3), and an aspect ratio (16:9, 9:16, 1:1). The translation step is where the producer's eye is most valuable.
Defining the clip library shape
Before generating, decide the structure: categories (people, places, products, transitions), lengths (typical 3 to 6 seconds for b-roll inserts), and ratios. LTX-2.3's hosted API supports resolutions up to 4K across its Fast and Pro tiers (1080p, 1440p, and 4K), plus native 1080×1920 portrait output, with maximum duration depending on the selected variant and resolution.
For b-roll, you will rarely generate clips longer than 8 seconds, because short clips are easier to nail and cheaper to iterate.
Prompt templates for repeatable b-roll
Build a template per category. A template names the camera move, the lighting, the wardrobe palette, and the energy, with a single slot for the specific subject. Once a template produces a clip you like, change only the subject slot to extend the library.
This is how you produce a coherent set of forty clips that feel like they came from the same shoot.
Batch generation patterns
Generate in small batches (5 to 10 clips per session), review each one against the brief, and iterate on the templates rather than on individual prompts. Generating in batches surfaces template-level issues (the wardrobe palette is wrong across all clips, the camera moves are too similar) faster than generating ad-hoc.
Tagging, cataloging, and version control
Every generated clip needs a filename, a prompt record, a generation timestamp, and a version. The prompt record is the most underrated artifact in this workflow. Six months later, when the client asks for "another like clip 14," the prompt is what reproduces the look. Store prompts alongside the assets in your DAM or shared drive.
Prompt patterns for common b-roll categories
Lifestyle and people
Lifestyle b-roll lives or dies on naturalism. Specify activity, environment, lighting, and avoid over-direction.
Template: "A medium handheld shot of {subject} {activity} in {environment}, {lighting condition}, natural body language, slight handheld movement, ambient light only."
Example fill: "A medium handheld shot of a young woman reading on a window seat in a sunlit apartment, late-morning natural light through sheer curtains, natural body language, slight handheld movement, ambient light only."
Urban and architecture
Urban b-roll is mostly about light and time of day. Specify both precisely.
Template: "A slow {camera move} of {urban subject} in {location}, {time of day} light, {weather}, {camera height}."
Example fill: "A slow dolly forward of an empty cobblestone alley in Lisbon, late golden-hour light, clear sky, eye level."
Nature and landscapes
Landscape b-roll benefits from scale cues. Name the foreground, middle ground, and background.
Template: "A {camera move} across {landscape}, foreground {detail}, middle ground {detail}, background {scale element}, {light source}."
Corporate and office
Corporate b-roll is the hardest stock category to make feel non-generic. Specify activity over scenery; show people doing something specific.
Template: "A medium shot of {role} {specific activity} in {workspace type}, soft natural light from {direction}, contemporary office, focused expression, subtle camera movement."
Product close-ups and macro
Product b-roll needs control over texture, light, and rotation. Reference materials by name.
Template: "A macro shot of {product} on a {surface}, {lighting setup}, shallow depth of field, slow {rotation or push-in}, the {material} catching the light."
How do you keep quality consistent across a clip library?
Locking style across a clip library
Cross-clip consistency comes from prompt-template discipline, not from cherry-picking individual generations. If every prompt in the library shares the same lighting language ("late golden-hour"), the same camera language ("handheld at eye level"), and the same energy descriptors, the clips will cut together.
If each prompt freelances, the library will look stitched.
Color and grading consistency
AI-generated clips do not come out of the model perfectly graded. Treat the LTX-2.3 output as raw footage and run it through your standard grading pipeline (DaVinci Resolve, Premiere) before delivery. This is where you lock palette and contrast across the library.
Aspect ratios for delivery
Plan deliverables by ratio: 1920×1080 for traditional broadcast and YouTube and 1080×1920 for vertical social — both natively supported resolutions — and 1:1 for square placements via the separate Reframe (video-outpainting) endpoint, which expands a generated clip to a new aspect ratio (16:9, 9:16, 1:1, 4:5, or 5:4) without altering the existing pixels.
Some clients will need all three from the same library. Generate the master ratio first, then reframe into the alternate ratios rather than cropping.
Cost: traditional stock vs. AI generation at agency scale
The math depends on volume and clip mix. For a campaign with 5 hero clips and 30 b-roll inserts, premium stock will typically run $5,000 to $15,000 in licensing alone.
The same library generated through the LTX-2.3 API costs a fraction of that in per-second inference, even accounting for the iteration cycles every AI workflow needs. The producer time is roughly comparable: searching the right stock clips and writing the right prompts both take real attention.
The math flips for very small projects. A single 8-second clip from a stock library may still be cheaper than the producer time to brief it through an AI workflow. Use AI when the clip count justifies the producer overhead, which in practice means roughly ten clips or more per campaign.
When should you still license stock?
Three cases where licensed stock still wins. First, real footage of real public events (a real concert, a real city skyline at a specific date) that needs to be the actual place at the actual time.
Second, footage where a release is non-negotiable for legal reasons (recognizable individuals, branded properties).
Third, projects where the clip count is small enough that the producer overhead of an AI workflow does not pay back. For everything else (wellness lifestyle, urban abstract, corporate atmosphere, product macros), AI-generated b-roll wins on specificity, exclusivity, and cost.
Summary
LTX-2.3 produces agency-grade stock footage when the workflow is built around prompt templates rather than one-off generations, when brief intake explicitly translates creative concepts into prompts and ratios.
The cost math favors AI generation past roughly ten clips per campaign, the brand-safety math favors AI for any work where exclusivity matters, and the specificity math favors AI for any clip a stock library cannot match exactly.
Start by templating five clips in a single category, run them through your standard grade, and put them next to your most-recent licensed stock buy. The comparison usually answers the rest of the questions.
