A toddler casually finishing a birthday-cake-sized slice of birthday cake. A corgi ordering an impossibly large burger. A tiny dinosaur crunching through a watermelon bigger than it is. It's one of the most-watched AI video genres on Facebook and TikTok right now — here's what's actually behind it, and how to make your own version.
This isn't a special app or model built just for "food videos" — it's a general AI video generator (see our beginner's guide to how these tools work if you're new to the category), pointed at a specific prompting style: an exaggerated mismatch in scale between the character and the food, paired with slow, satisfying, repetitive motion — biting, chewing, crunching — that reads almost like ASMR content even without added sound. The "character" is usually the same recurring figure across dozens of clips from one account, which is the giveaway that it's one consistent reference image or character description being reused, not a new setup each time.
Image-to-video starts from a single reference photo or illustration and animates it — this is how most of the accounts posting dozens of these keep the same face/character recognizable clip after clip, since re-describing a character from a text prompt every time tends to produce a slightly different-looking result each generation. Text-to-video starts from nothing but a written description, which is faster to try once but harder to keep visually consistent across multiple posts. If you want a recurring character (your own mascot, not a one-off clip), image-to-video from one fixed reference is the more reliable route.
There's no published benchmark specifically for "AI eating-motion realism," so treat any strong claim here — including ours — with some skepticism and just test your own prompt on a couple of tools before committing to one. What we can say with more confidence: Kling has the strongest general reputation for physically plausible motion, which is directly relevant to chewing/biting; Runway and Pika both have mature image-to-video support with good reference-image consistency, which matters more than raw motion quality if your priority is a recognizable recurring character; and Seedance is worth a look if you're comfortable with a slightly more technical setup (see that page for what "slightly more technical" means in practice) since it's priced per clip with no subscription required.
This is worth pausing on, because it's easy to skip past in the excitement of trying it. If you want to base a character on your own kid using a real photo, check the specific tool's policy first — most video generators put extra restrictions or verification steps around generating realistic video of real, identifiable people, and minors specifically, even when you own the photo and have every right to use it. Practically speaking, this is exactly why most of the recurring viral "AI baby" and "AI pet" accounts you've seen use one consistent, fully AI-generated or illustrated character instead of a real child's actual likeness — it's simpler under every tool's policy, and it means the "character" can keep appearing in new videos indefinitely without re-litigating consent each time. If your goal is really just "something my kid will find funny," an AI-generated character that merely resembles the vibe (same age, same general look) sidesteps the whole question and is just as fun to watch.
Same pricing as any other short clip on whichever tool you use — commonly a few cents to around $2 per 5-second generation once you're past the free tier, though budget for a few regenerations to land a clean result (see the tips above). See our full AI video generator pricing comparison for exact numbers by tool, or the cheapest way to try it first if you just want to test the genre before spending anything.
Once you've got a clip you like
The video comes out with its full generated background baked in — no built-in way to export just the character. If you want to pull your character out to reuse on a different background, layer it into a thumbnail, or turn a favorite frame into a sticker, that's a background-removal step done after generation, not something the video tool itself offers. RemoveGifBG processes AI-generated MP4s frame by frame and hands back a transparent result — see the AI video background remover guide.
Related: How AI videos are made · Tiny-world rescue trend · AI ASMR trend · Pricing compared · Cheapest way to start
No single app is behind all of them — the same handful of general-purpose AI video generators (Runway, Pika, Kling, Seedance) that make every other AI video trend are behind this one too. There's no dedicated "giant food video" app; it's a prompting style (an exaggerated food scale plus a cute or absurd character) applied to a general tool.
Check the specific tool's policy first — most restrict or require extra verification for generating realistic video of real, identifiable minors from an uploaded photo, even a photo you own. Many of the recurring viral "AI baby" accounts sidestep this entirely by using one consistent, fully AI-generated or illustrated character rather than a real child's likeness, which is also simpler from a rights standpoint if you want to reuse the same character across multiple clips.
Fine, repetitive motion around the mouth is one of the harder things for current video models to get consistently right, especially combined with a food texture (crumbs, sauce, bite marks) the model also has to render believably. Large, exaggerated bites tend to hold up better than subtle chewing — it's a real limitation of the current models, not something you're doing wrong with the prompt.
Roughly the same as any other short AI video clip on the same tool — commonly a few cents to around two dollars per 5-second generation once you're past a free tier, though you should expect to regenerate a few times to get a clean result. See our full pricing comparison for exact per-tool numbers.