Skip to content
Updated for 2026

AI video adds marks because fine detail drifts toward pattern as a clip runs, and naming what you don't want — even in a negation like "no tattoos" — can put that exact thing into the generation. Fix it by describing the surface you want positively, asking for less motion, and regenerating; a rerun often comes back clean.

Why Does AI Video Add Marks That Weren't in the Photo?

By MakeThisVid Editorial Updated Methodology

A generated clip comes back with a dark patch on skin that reads like a bruise or a tattoo, texture that's gone patchy and blotchy, a film-strip bar rendered into the frame, or a pattern spreading across what was a plain, smooth surface — and none of it is in the source photo. This isn't random. It has two distinct causes, and telling them apart matters, because the fix for one does nothing for the other and the wrong fix can make things worse.

Make a video from one prompt. $2.99 a video · Ready in about a minute · No subscription.

Key facts

Cause 1 Drift into pattern Fine detail — skin texture, fabric weave, hair — gets progressively reinterpreted as the clip advances. Subtle becomes coarse; coarse becomes a mark.
Cause 2 The prompt summoned it Mentioning a thing at all can put it in the frame, including as a negation. Writing "no tattoos" puts tattoos in the description.
Fix Describe the surface you want Positive description — "clear, even skin," "plain sleeve" — is the only reliable form of control here.
Honest limit Wording reduces odds, doesn't eliminate them A mark that recurs across several runs is usually being fed by the source photo, not the prompt.

How to Use MakeThisVid

From prompt to downloadable MP4, ready to deploy.

  1. Never name what you don't want

    Writing "no tattoos" or "no blemishes" puts the concept of tattoos and blemishes into the description. Models are unreliable at handling negatives — the safest move is to leave the unwanted thing out of the prompt entirely, not to mention and negate it.

  2. Describe the surface you do want, concretely

    Say "clear, even skin" instead of "no blemishes." Say "plain sleeve" instead of "no pattern." Positive, specific description of the end state is the only reliable form of control here — it gives the model something to render toward instead of something to avoid.

  3. Ask for less motion and a shorter apparent action

    Every frame the model redraws is another chance for fine detail to drift toward pattern. A subject that holds still, a camera that barely moves, and a simple, short action all mean less redrawing and less opportunity for texture to compound into a mark.

  4. Regenerate before you rewrite

    This varies between runs on the same prompt. A second attempt with no changes at all often comes back clean. Try that before you start editing the wording.

  5. If it keeps happening, look at the source photo

    You cannot eliminate this with wording — only reduce the odds. If a specific mark keeps recurring across several runs, the prompt usually isn't the problem: a shadow, a texture, or a crease in the source photo is being read as something else. Changing the source photo beats fighting the prompt.

Make a video — $2.99

Who Uses MakeThisVid for This

Skin close-ups

Portraits, hands, and any tight shot of skin are where drift shows up first — smooth tone breaking into blotchy, patchy texture as the clip runs.

Plain fabric and flat surfaces

Solid-color sleeves, walls, tablecloths, and other smooth, unpatterned surfaces are exactly where a repeating pattern tends to grow — there's nothing else for the model to hold onto.

Prompts that list what to avoid

Any prompt built as a list of things not to include — no scars, no wrinkles, no stains — is reliably the prompt most likely to produce them.

Frequently Asked Questions

Two mechanisms, usually. Fine detail drifts toward coarser pattern as the clip advances, and separately, naming an unwanted thing in the prompt — even as a negation — can introduce it. Check your prompt for negatives first; that's the more fixable cause.
Mentioning a concept, even to rule it out, puts that concept in the description the model works from. It is unreliable at handling negatives, so "no tattoos" can function as a request for tattoos. Describe the clean result you want instead of naming what to exclude.
Drift compounds. Each frame reinterprets detail from the frame before it, so small texture changes early in the clip accumulate into visible marks or patterns by the end.
Often, yes. Results vary between runs even with identical wording, so a second attempt with no changes at all can come back clean. Try that before rewriting the prompt.
Yes. More motion and more action means more of the frame gets redrawn, and every redraw is another chance for fine detail to drift toward a mark. A subject that holds still and a short, simple action reduce that opportunity.
That pattern points at the source photo, not the prompt. A shadow, a texture, or a crease in the original image is probably being read as the mark and fed into every render. Changing the source photo is more effective than continuing to adjust wording.
You can reduce the odds, not eliminate them. Positive description, no negatives, and less motion all help, but this is a property of how the model resolves ambiguity — some variation between runs is unavoidable.

Rewrite the prompt, then try again

$2.99 makes one clean video, ready in about a minute — or add 3 more takes for $6.99, so you can regenerate until a take comes back clean. Pay once. No subscription required.

Make a video — $2.99

Broken renders are remade or refunded.

or 3 more takes — $6.99 Any photo, any look · About $2.33 a take · No subscription required.