Eleven of the looks in this library are AI effect guides, and each one separates what we ran from what a vendor states. We ran the effect ourselves for ten guides: polaroid, action figure and anime film style on 2026-10-21; dance on 2026-10-26, with one Kling 3.0 Motion Control clip on Higgsfield; caricature on 2026-10-26 and yearbook on 2026-10-27, each with one prompt on three image models; old photo restoration on 2026-10-27, with one restore prompt on three image models and one Kling 2.6 clip; couple photo on 2026-10-27, with two fictional adults on three image models; and pet dance and hug on 2026-10-27, each with one start photo on two video models. The kiss guide reports vendor terms and is tested in 0 tools. The five effects on our radar stayed at or above 25% of their own peak for 14 to 46 of the last 52 weeks, the five trends we track for 1 to 3 (Google Trends US, weeks to 2026-10-18).
Effect, trend or template: what’s the difference?
An effect is a repeatable result that people search for all year, such as a kissing video or an anime film look. One dated wave around a specific prompt or meme counts as a trend. Templates are presets inside a tool that produce an effect in a few clicks. Our Higgsfield templates hub covers the presets.
The difference shows in the data. “ai kissing video” stayed at or above 25% of its peak in 46 of 52 weeks. “ghibli ai”, the search behind our anime film style guide, did so in 43. “ai polaroid” follows with 40, “ai dance video” with 36 and “ai action figure” with 14 (Google Trends US via DataForSEO, weeks of 2025-10-26 to 2026-10-18; the last week was still running at fetch on 2026-10-21). Single waves sit on the trend radar, each with its own status.
Lifespan is not popularity. Each curve is scaled to its own peak, so a 46-week effect may still draw fewer searches than a 14-week one. The week count only tells you how long interest lasted.
What “tested” and “vendor states” mean in our tables
As of 2026-10-21, every table cell on an effect guide carries a date and says whose claim it is. “Vendor states” means the vendor wrote it on its own site or help page, and we read that page on the date shown. It is not our measurement.
“Tested” means we ran it ourselves, with our fictional, AI-generated test character as the only subject. All our runs so far come from one tool, Higgsfield. Each run lists model, date, credits charged and tries.
“Tested in N tools” on a guide counts the tools we ran for that effect. One run is a single sample, not a ranking. The full method is on our test method page.
“Not tested” means we did not run it, and we never fill that cell with a guess. Where we could not confirm a vendor claim, the cell reads “not verified as of” plus the date. Same-input tests across several vendors are planned for a later update. Until then the tables hold vendor-stated terms and our Higgsfield runs.
Effect guides
- AI kissing video: what four vendors state about a free first try, checked 2026-10-20. Tested in 0 tools: a kiss needs two consenting adults, and we have not run one with our two fictional test characters.
- AI couple photo: one prompt with two reference photos, our test character and a second fictional adult, on Nano Banana Pro, Seedream 5.0 Lite and GPT Image 2, one try each on 2026-10-27, plus a Valentine’s variant on Nano Banana Pro. By eye Seedream 5.0 Lite drifted her face; the other two kept both faces.
- AI hug video: one AI-generated start frame of two fictional adults through Kling 3.0 and Hailuo 02 image-to-video, one try each on 2026-10-27. In both clips a cheek press can read as a peck; photos of people who have died are not tested.
- AI dance video: how a photo plus a dance clip becomes a video, and what four vendors state about a free first try (checked 2026-10-20). We made a full-body start frame on 2026-10-21 and one Kling 3.0 Motion Control clip from it on 2026-10-26, so it is tested in 1 tool, Higgsfield; none of the dance apps is tested.
- AI pet dance video: one AI-generated fictional dog photo through Kling 3.0 and Hailuo 02 image-to-video, one try each on 2026-10-27. Pet motion transfer and cats are not tested.
- AI polaroid: the flash-and-curtain prompt on Nano Banana Pro and Seedream 5.0 Lite, one charged image each on 2026-10-21. Our log marks a Christmas variant on Nano Banana Pro as identity drift.
- AI action figure: the blister-pack prompt on the same two models, one try each on 2026-10-21.
- AI caricature from a photo: one caricature prompt on Nano Banana Pro, Seedream 5.0 Lite and GPT Image 2, one try each on 2026-10-26, plus one follow-up edit that fixed the hands. Our log marks only the Nano Banana Pro result as a true caricature.
- AI yearbook: one 1990s yearbook prompt on Nano Banana Pro, Seedream 5.0 Lite and GPT Image 2, one try each on 2026-10-27. Only the GPT Image 2 result has the pink-violet laser backdrop.
- AI old photo restoration: one keep-the-face restore prompt on Nano Banana Pro, Seedream 5.0 Lite and GPT Image 2, one try each on 2026-10-27, on a synthetic damaged photo of our test character, plus one Kling 2.6 clip of the restored portrait. By eye only the Seedream 5.0 Lite result kept the face.
- Anime film style: a hand-painted look from a prompt that names no studio. Both models returned the look on the first try on 2026-10-21.
- AI Action Figure: Prompt, Two-Model Test and Free Routes
- AI Caricature From a Photo: One Prompt, Three Model Results
- AI Couple Photo From Two Photos: One Prompt, Three Models Tested
- AI Dance Video: How to Make One and What Each Vendor Says Is Free
- AI Dog Dance Video: Kling 3.0 vs Hailuo 02 on One Dog Photo
- AI Hug Video: Kling 3.0 vs Hailuo 02 on One Start Frame
- AI Kissing Video: What Each Vendor States About Free Limits
- AI Polaroid Trend: Prompt and Tested Result
- AI Restore Old Photo: One Prompt, Three Models, One Face Test
- AI Yearbook Photo: One Prompt, Three Model Results
- Studio Ghibli AI: a hand-painted anime look without naming the studio
The consent rule for kiss, couple and hug effects
Use your own photo, or a photo of an adult who has agreed to appear. Everyone shown must be an adult. In a two-person effect, both people must consent before you upload. Leave out minors, celebrities, ex-partners and anyone who has not said yes.
We apply the same rule to what we publish. Our pages show no real person’s face, and we do not teach likeness-swap prompts with real people. The full rules sit in our content policy.
Where each effect can be made without paying
Each of the eleven effects has at least one vendor-stated free route, and every guide lists its free routes first, with dates. The amounts stay on the guides: vendor terms change, and a second copy here would go stale. Gemini and ChatGPT both accept an uploaded photo with a prompt, which covers the six one-photo effects; for the couple photo, Gemini’s help page says it takes several uploaded images (checked 2026-10-27). Their free limits are not verified as of 2026-10-21, and our prompts there are not tested. What a free Higgsfield account includes is on the Higgsfield free plan page.