Welding: motion + PPE stills you photographed
Do not generate workers' faces.
Do not generate workers' faces. Photograph the PPE. Generate the motion of the work — the bead, the spark, the coupon on the table — without a person attached. A shop video that needs a crew still needs a crew you filmed. A generated welder is a private person in a helmet, and "it's just an extra" is not a grant.
A deepfake of a private person is not a style. TikTok's private-figure line is permission, not labelling. Your fitter, your apprentice, a stock-prompt face that happens to look like last year's hire: if someone who knows them would say "that is them," you have a replica. If they do not exist, you still do not get to put them in the shop as if they clocked in.
Photograph the PPE. Leave the faces off the prompt.
The stills that prove you run a professional cell are objects you already own: hood on the bench, gloves, jacket, curtains, the coupon with a bead you actually ran. Shoot them. Overhead light, phone, the same bay the buyer will visit. Those frames are evidence. A generated hood with a generated jaw inside it is a costume.
PPE in the feed is how a trades buyer reads safety without a speech. Construction and trades video is the wider context. This page is narrower: faces in the feed are people you photographed, or no people at all.
Do not upload a crew headshot as a "style reference" for a helmeted extra. That is how you get their head. A toolbox-talk release does not authorise generating new performances. Most shops should just shoot the still. Minors are a hard stop. Do not generate an "apprentice."
Motion of the work, not of a worker
The motion job is the process: puddle, spark, a torch travel, a grind pass, smoke under a hood that has nobody in it. That is text-to-video of physics, not a portrait.
Kling 3 Turbo text-to-video is the silent plate for that: audio is false, 3s through 15s, 720p or 1080p, aspects 16:9, 9:16, and 1:1. You get motion. You do not get a voice, and you should not get a face you did not ask for — so do not ask for one. Turbo is a silent plate you score later. Start at 3s, 4s, or 5s. A 15-second miss of a hallucinated welder is a long miss.
Prompt the work. "Close-up of a TIG bead traveling on stainless, no people, no faces, helmet and gloves on the bench at the edge of frame." If a face appears anyway, throw the take. Do not "fix 15% of the jaw." Do not ship a shop extra who does not exist.
The AI video generator is the door. A real coupon or booth still belongs in a locked still first, then an image-to-video row, not in a Turbo paragraph. The PPE stills you photographed can sit on the grid as stills. They do not need to become a generated crew.
Type the cert. Do not let the model invent it.
ASME stamps, procedure numbers, thickness — those are captions you write. Kling will not typeset a WPS you can defend. After picture lock, burn the line with a text overlay or time several with timed text overlays. Auto-captions transcribe speech. A silent spark plate has no speech. If a cleared welder talks on camera, transcribe them.
The shop that already has a phone in the bay does not need a synthetic crew. It needs the bead, the PPE stills, and a sentence that is true.
FAQ
Can I generate a welder if I never show the face — just the helmet?
A helmeted body that is a real private person you referenced is still a replica. A fully invented extra is safer on likeness and still a fake employee if you imply they work for you. Prefer process close-ups and PPE stills you shot. If you need a person, film one.
Why not image-to-video from a photo of my welder?
Because that is how you animate a private person. I2V holds identity. That is the feature, and it is the problem. Photograph PPE and coupons. Animate process without the crew, or film the crew.
Is Kling 3 Turbo going to add dialogue if I write a toolbox-talk script into the prompt?
No. audio is false. Quoted speech will not become a stem. You may get a silent mouth. That is worse. Keep people out of the prompt.
Can we show a generated injury as a safety lesson?
No. That is a generated victim in a shop. Film a demonstration with real PPE, or use a diagram. Do not invent a burned face.