Embroidery: the stitch still, then maybe I2V

Macro still first. The product is the stitch — density, thread colour, the mark as sewn — not a camera orbit you have not earned. If the still is wrong, every second of video is a regenerate of the logo. Lock the still before you spend on motion is the routing rule. This post is that rule on a hoop.

Embroidery fails in video the way chest prints fail: the graphic is small in the latent, high-frequency, and the one surface a buyer inspects. A model that re-draws letterforms every frame is not "adding life." It is unpicking the job.

The stitch is the identity

Shoot a macro of the actual work. Fill the frame with the mark. Even light. No busy background competing for the consistency budget. If a six-word slogan is tiny on a sleeve, it will be soup in motion — the same scale rule as on-screen text and as a back-print that only occupies a few pixels.

A still is not locked because you generated one pretty hoop. It is locked when someone can point at the frame and say the thread colour, the stitch direction, and the spelling are the ones you would ship. If the logo is already wrong on frame one, video cannot lock a graphic the first frame never had.

Iterate at image prices. Do not debug identity at video prices.

Motion is a push, not a restage

AI image-to-video treats the still as frame one. Composition, lighting, and the mark are already decided. The prompt describes only what changes: a slow push, a slight hoop rotation, thread sheen. Leave the logo out of the paragraph. Repeating the brand in text is how you invite the model to renegotiate the letterforms.

Keep the clip short. A three-to-five-second locked mid-shot of a large, sharp mark will often hold. A ten-second orbit of dense fill almost never will. That is shot design. Drift is the model running out of information about stitches it has never seen. Shorten, ask for less motion, keep the camera on one side of the work.

First-and-last-frame runs exist when both ends are pictures you own: hoop at rest, hoop with the needle up. Do not feed a finished hero to a restage endpoint and ask it to "just add motion." It will recast the room.

Text-to-video is how you get someone else's monogram

A paragraph that says "elegant custom embroidery on a black hoodie" will invent a mark. It will not match the PO. Text-to-video is allowed when there is no subject to lock — abstract thread, a texture plate nobody has to match to a SKU. It is banned as an identity tool.

Reusable products belong in a saved still, not in a weekly rewrite. If the same crest has to appear on a cap, a towel, and a jacket, lock three macros. Animate each once. Do not hope one prompt remembers the digitizing file.

The hold test

Print the still. Would you send that photo to the client as proof?

If no, stop. If yes, buy a short I2V push on that file and inspect the mark at full size on every frame you would actually post. A morphing letter is a failed take, not a grade of footage. Reroll motion on the same still. Do not "fix" it by describing the crest harder.

FAQ

Can I skip the macro if the model is good at logos now?

No. Better models reduce drift. They do not create a contract. Tomorrow's job still needs a still it can see. Prompt memory is not the digitizing file.

Why not generate the stitch from text and then caption the brand name?

Because the stitch is the brand. Captions will not rescue a monogram the picture invented. Lock the sew-out, then maybe move it.

Should every listing be I2V?

When the composition is the ad, yes — image-to-video from the approved macro. When the crest must appear in a room you do not have, that is a different contract (reference, not a restage of the hero). Text-to-video is how you get a crest that is almost yours.

The still looks right and the clip still crawls. Now what?

Then you have a motion problem: too long, too much orbit, mark too small. Shorten. Fill the frame. Do not rewrite the logo into the prompt.