AI video for animal shelters and rescues
Adoption profiles, fundraising appeals and volunteer recruitment where the animal footage stays real and only the production around it is generated.
A shelter's content problem is not creativity. It is that the intake list changes every week, the people who could film are the same people cleaning kennels, and the animal who most needs a video is usually the one who photographs worst. The result at most rescues is a phone photo posted to Facebook with three lines of text, which works about as well as it sounds.
Video fixes the specific thing photos cannot do, which is show temperament. A still image of a nervous six-year-old terrier tells an adopter nothing. Fifteen seconds of that dog leaning into a volunteer's hand tells them everything. The reason shelters do not ship that video is production overhead, and production overhead is the part that can be automated.
The rule that makes all of this safe: the animal is always real footage. Everything around the animal can be generated.
Why the animal is never generated
This is not a stylistic preference. Three concrete failures follow from breaking it.
An adopter who drives ninety minutes because of a video and meets a dog that does not look like the video does not adopt, and does tell people. Generated pet imagery is recognisable to anyone who spends time in the space, and a rescue's entire fundraising position rests on being believed. And the moment one image is synthetic, every "urgent, this dog dies tomorrow" post you have ever made becomes something a donor might reasonably doubt.
There is a related trap. Versely ships one-tap templates that animate pets, and they are a genuinely fun thing to do with your own dog. Applying them to an animal currently in your care is a different act. An adoptable animal cannot consent to being puppeteered, adopters read it as a shelter treating a resident as content, and it is the single fastest way to look unserious to a foundation reviewing your grant application. Use the templates on staff pets for a light social post if you want. Keep them off the adoption feed.
So: film the animal. A phone at kennel height, thirty seconds, no music, natural light. That is your raw material.
What the generated layer is actually for
Everything that is not the animal. The list is longer than people expect, and it is the part that eats volunteer hours.
- The intro card. Same design every week, name and age swapped. This is what makes a feed of thirty different animals read as one organisation.
- Lower thirds and info panels. Weight, good with cats, house-trained, adoption fee status.
- Transition and background plates. Textures and colour fields built from your identity, used behind text so you are not cutting to black.
- The end card. Adoption link, shelter hours, the animal's ID number.
- Environment b-roll for appeals. A generic road at dawn for a transport story, a generic waiting room for a medical appeal. Never a generated animal in these frames, and never a generated version of your building.
Generated b-roll is the honest use of the technology here, because nobody is being asked to believe a specific animal exists. It is set dressing, and set dressing has always been constructed.
The three jobs, and how each one is shaped
| Job | Shape | Length | What it needs from you |
|---|---|---|---|
| Adoption profile | One animal, one temperament moment, name and ID | 20 to 30s | 30s of real footage |
| Fundraising appeal | One specific need with a number, one real animal it helps | 30 to 45s | Footage plus an actual figure |
| Volunteer recruitment | An honest shift, including the unglamorous parts | 40 to 60s | One filmed shift, once a quarter |
Adoption profiles live or die on specificity. "Sweet boy, loves everyone" describes every dog in the building. "Walks past other dogs without reacting, which is rarer than you think" describes one. Write the line before you film, then film the moment that proves it.
Fundraising appeals need a number and a use. General appeals underperform specific ones consistently, and specificity costs nothing to write. The dental extraction, the transport run, the two weeks of foster supplies. Show the animal it applies to, in real footage, and say what happens if the amount is not raised. Do not manufacture urgency you do not have, because rescues that cry wolf lose the ability to raise money when they genuinely need it.
Volunteer recruitment is the one shelters get wrong most often, by showing only puppies. People who show up expecting puppies leave in three weeks. Show the laundry, the 7am shift, the dog who does not want to come out of the kennel, and then show why people stay. Self-selection at the top of the funnel saves you the onboarding cost at the bottom.
A weekly plan a two-person shelter can keep
The trap is building something that requires a good week to execute. Build for the bad week.
- One filming block, 45 minutes, same day each week. Walk the kennels with a phone. Thirty to sixty seconds per animal, no setup, no retakes. Six to ten animals per block.
- One assembly session, 60 to 90 minutes. Drop each clip into the saved editor draft, swap the name and ID text, export. The editor is timeline-based and re-renderable, so a saved edit is a reusable structure rather than a finished file you have to rebuild.
- Iterate at preview resolution. A preview pass renders at 480p for free with a short per-user cooldown between passes, which is what lets a volunteer nudge the timing without burning through your budget. The preview and export cost guide explains where the single export charge lands.
- Caption everything. Shelter content is watched silently, on a phone, at work. Auto-transcribed burned-in captions are not optional, and the caption generator handles the transcription pass.
- Publish on a schedule, not on a feeling. Posting to connected accounts costs 2 credits per platform, which makes cross-posting cheap enough to stop being a decision.
- Post the adoption update. When an animal goes home, post it, tagged to the original video. This is your best-performing content, it closes the loop for everyone who saw the profile, and it costs you one clip from the adopter.
Two structural notes. Batch on the day, do not spread across the week, because the setup cost is what kills volunteer-run content. And keep a fixed weekly series slot for one recurring format, which is the mechanism behind most of the recurring series formats. A predictable Tuesday post trains an audience faster than better content posted at random.
The failure modes worth naming
- Generated animals anywhere in the adoption feed. Covered above, and worth repeating to every volunteer with posting access.
- Misery as the default register. It raises money once and exhausts an audience permanently. Alternate appeal content with adoption updates and volunteer content.
- Leaving adopted animals up. Nothing damages a shelter's feed like an enquiry about a dog who went home in March.
- Voice cloning a staff member without written consent. Get it in writing, keep it, and make it revocable.
- Filming another organisation's animals. Transport partners and foster networks have their own consent and confidentiality practices. Ask first.
- Publishing an animal before behavioural assessment clears them. Marketing running ahead of the intake process creates a promise the shelter cannot keep.
If you are also building a donor programme around this content, the nonprofit fundraising playbook covers the campaign side, and the animal shelters page has the sector-specific surfaces.
FAQ
Is any generated imagery acceptable in an adoption post?
Yes, as long as it is clearly not an animal and clearly not your building. Title cards, text backing plates, transitions and end cards are fine and nobody will object to them. The line is anything a viewer could mistake for documentation of a real animal or a real place they are being asked to visit.
How do we film animals who do not perform on camera?
Do not try to make them perform. A shut-down dog in a kennel is a real thing an adopter needs to see, and the video that says "she is terrified here and completely different in a home, which her foster will tell you about" places animals that a fake-happy video does not. Film the truth and write the context.
What is the smallest weekly commitment that actually works?
Three adoption profiles a week. Below that the feed goes quiet between posts and the audience stops checking. Three is achievable in a single 45-minute filming block plus an hour of assembly, and it is a volume most two-person shelters can hold through a bad month.
Should we use a cloned voice for narration?
Usually not for adoption profiles, where the volunteer's real, slightly imperfect voice outperforms a polished read. It is more defensible for logistics and recruitment content, where consistency matters and the content is not emotional. Either way, written consent from the person whose voice it is, on file, revocable.