When to Fall Back to Stock B-Roll
Purists generate everything. Pros generate the parts that matter and buy the rest.
There's a dogma among AI video practitioners that says: generate everything. Every shot, every frame, every particle of the final video should come from the model. Purity.
This is wrong, and it's slow. Professional workflows mix AI-generated content with stock footage, taking advantage of each where it's strongest. Here's the decision tree.
Generate what carries the IP
The parts of your video that must be yours — your mascot, your product, your branded scene — are generation territory. That's the stuff that can't be bought off the shelf. That's where AI video's flexibility matters.
Generate:
- Close-ups on your mascot.
- Product shots that feature your specific packaging.
- Branded scenes (the exterior of your fictional diner, the inside of your fictional laboratory).
- Interactions between multiple of your characters.
These are the parts where the creative work lives. Spend generation time here.
Stock everything else
Generic establishing shots. City skylines. Anonymous crowds. Nature footage. Generic office environments. Laboratory close-ups of unrelated hands doing unrelated things.
All of these are available from stock libraries (Getty, Shutterstock, Pond5, Artgrid) at reasonable per-clip prices. Some are available free from Pexels or Pixabay.
For a 30-second ad with five establishing shots, stock can save you 2-3 hours of generation time at the cost of maybe $50 in licenses.
The cost math
Generation time is real money:
- GPU costs for the generation.
- Engineering time to set up the generation pipeline.
- Review time to reject bad outputs and regenerate.
- Editor time to composite results.
For generic content, all of this is wasted. A stock clip of a city skyline costs $15 and is immediately usable. A generated city skyline costs computing resources, takes 2 minutes, and looks slightly worse than stock.
Stock wins for generic content. Generation wins for specific content. Mix both.
The workflow
My production workflow:
- Storyboard the video, tagging each shot as "generated" or "stock."
- Generated shots go into the AI pipeline.
- Stock shots get sourced in parallel.
- Editor composites everything together, matching color grading for coherence.
The parallelization matters. While the AI pipeline is running (which takes minutes per shot), stock sourcing happens concurrently. Total time-to-delivery is shorter than generating everything.
The quality argument
There's an implicit claim that stock footage looks lower-quality than AI generation. This used to be true. It isn't anymore.
High-end stock libraries have 4K+ footage shot by professionals. The quality floor is high. Blended correctly with AI generation (matching color grade, frame rate, and compression), stock shots are invisible in the final cut.
Low-end stock — free Pexels clips — can be obviously stock, but that's avoidable. Budget for paid stock.
The licensing
Stock licensing is usually royalty-free for commercial use, but read the terms. Some licenses restrict:
- Political advertising.
- Controversial content.
- Use as the primary element of a commercial (you can use it as B-roll, but not as the hero shot).
For DebaterX, I use stock only for incidental B-roll and only from licenses that clearly permit commercial short-form video. Legal clarity is worth the small licensing fee.
The purist objection
Some creators object to mixing stock on principle. They want their work to be "fully AI-generated." This is a creative choice for some — especially artists exploring what AI video specifically can do.
For production work — ads, brand content, commercial deliverables — the purist position is self-defeating. You're not making a statement about AI's capabilities. You're making content for a client. The client doesn't care how the individual shots were made. They care about the final product's quality and cost.
Use the tool that's best for each shot. That's often AI. Sometimes it's stock. Neither is morally superior.
The rule
Your AI generation budget — time, compute, attention — is finite. Spend it on shots that can't be bought. Save it on shots that can be.
The ratio that works for me is roughly 30% generated, 70% stock-or-reused. Your ratio may differ. The principle stays the same: optimize for what AI is uniquely good at, and buy everything else.