Turn a product photo into a clean reference shot
Reference-to-video is only as good as the picture you feed it, and the picture most people have is a phone snap with a price sticker and a countertop in frame. Upload it, pick a setting, and get back a shot worth feeding into the video step -- without a reshoot.
How it works
- 1. Upload the product photo you already have. Any phone photo where the whole product is in focus and in frame. It does not need to be a studio shot -- that is the point of this page.
- 2. Pick a setting. Three presets, each written as a full shot direction rather than left to guesswork: On a real desk (Looks like someone photographed it at home. Best for UGC ads.); Clean studio (Neutral backdrop, no distractions. Best when the product is the ad.); Held in a hand (Gives scale and a person. Best as the first frame of a video.).
- 3. Get one, two or four shots back. The product itself is kept exactly as supplied -- only the scene around it changes -- and each generation is priced before you press anything, at $0.11 (3 credits) per shot.
- 4. Send it to the video step. The tool below hands the result straight to the composer, because a reference image is not the finished ad -- it is the input to the generation that is.
What photos work, and what does not
The bar is lower than a studio shot: in focus, whole product in frame, is genuinely enough. What hurts a result is clutter the model has no instruction to remove -- a shopfront sign or someone else's branding in the background gets carried into the new scene rather than dropped, the same way it would in the reference-to-video step after this one. A tighter crop on just the product beats a wide shot of the whole counter it was sitting on.
Where the shot ends up
Two of the three presets render at 9:16, the vertical frame every short-form ad slot uses -- TikTok, Instagram Reels, YouTube Shorts and RedNote all crop to it. The clean-studio preset renders square (1:1), closer to a listing photo than a video frame. Pick the aspect ratio for where the finished clip is actually going, not for how the reference looks on its own.
Questions before you upload
What kind of photo works best as the input?
A phone photo of the actual product, in focus, with the whole thing in frame. That is genuinely the bar -- this exists because a raw phone snap is usually the weakest link in an AI ad, not because it needs a studio shot to start. What hurts more than blur is clutter the model has no instruction to remove: a price sticker, a shopfront sign, someone else’s logo in the background. Those get carried into the new scene rather than dropped, so a tighter crop on the product itself beats a wider shot of the whole counter.
Does it redraw my product, or just the scene around it?
Just the scene. Every preset direction ends with the same instruction: keep the product exactly as supplied. Reference-to-video, the step after this one, has no seed to fall back on if the product drifts between shots, so redrawing it would be the failure that looks like success -- a good-looking clip of a slightly different product. If a generation changes the product itself rather than the background, that is a bug, not a style choice.
Will it add a logo or text I did not give it?
It is instructed not to, and the honest caveat is about the SOURCE, not the setting: a model reproduces a wordmark it can actually read in your photo and tends to invent one when the source mark is too small or blurred to read. A tighter, sharper photo of the product’s own printing is the fix, the same way it is for a shopfront sign in the background.
What do I do with the image afterwards?
Feed it to the video step. This page produces a reference image, not a finished ad -- reference-to-video is the only generation path that holds a product’s identity across shots, and it is only as good as the reference it is given. The button at the end of a generation hands the image straight to the composer rather than only offering a download, because that is the actual next step for almost everyone who lands here.
What does one generation cost?
Shown on this page before you press anything, in credits and in dollars, computed by the same pricing function the generation itself bills against -- never a number typed into a marketing page. One press makes one, two or four images at that same per-image rate; there is no bulk discount, because the underlying image cost is linear too.