Recycling MRFs: sort-line still, then Kling motion after the line is locked

Lock the line photo. Kling I2V for belt motion. Do not let the stream morph into a different plant.

A materials recovery facility listing is this sort line: the belt, the optical, the eddy, the bunker a buyer already walked. Motion is allowed after that plate is true. A paragraph about “a modern recycling plant” will invent a cousin hall, a cousin stream, and pickers you did not hire. Photograph the line. Animate the belt.

The sort-line photo is the plant

Procurement pauses on hardware they can match: belt width, chute colour, the optical over PET, the bunker stencil. Text-to-video has no plate. It samples an average MRF — a different optical, mixed paper where your still showed bottles. That is a different plant.

Photograph square to the belt. Stream readable. Crop people. A picker in the still becomes a picker in the motion, and a generated cousin of that picker is a private person you do not have. Empty geometry is the honest MRF plate.

Logistics companies is nearby freight: dock motion, a trailer square to the bay. That page is not a sort line. Manufacturers is the plant tour. An MRF can film a real hall when operations has cleared the shot list. The listing a hauler pauses on is still the belt photo, not a capability reel that invents a seventh sorter.

Do not prompt “busy recycling line, workers sorting.” The model will pick a stream. Average is not your PET.

Kling V3 Standard I2V for belt motion

Image-to-video starts from the frame you upload. Frame one is the file. Kling Video V3 Standard Image to Video is the still-to-clip row for that plate: camera controls and element composition, image required, 3s–15s, 63 credits listed for the first five seconds, native audio on. Aspects: 16:9, 1:1, 9:16. No max-resolution token on this slug. Do not invent 4K.

Prompt camera and physics only. Belt travels. Camera locked. No new optical. No new chute. Re-describing “PET bottles on a high-speed line” in the paragraph is how the model redraws the stream. Element composition will not grow a sorter that is not in the still. Swap the still.

Iterate at 3s, 4s, or 5s — inside the 63-credit first bill. Extra seconds add on the matrix. Fifteen seconds of the wrong bunker is a long miss. Native audio is on: write silence, belt rumble, no voice, or a silent prompt still mints a soundtrack you did not pick. Do not prompt a plant-manager VO. The belt does not talk.

If the still would fail a site-walk audit, stop. Video is a terrible identity debugger for a sort line.

Seedance 2.5 when the hold has to last

A three-to-five-second belt tick is the usual listing. When the same locked photo has to hold longer in one pass, that is the Seedance family, still from the image-to-video tool.

Seedance 2.5 is the still-to-motion family: 4–30s, 720p, 47 credits, native audio. I2V is that job on this family. The published URL is the family page. Do not invent a second Seedance I2V slug. Thirty seconds is for a hold that actually lasts thirty seconds. A belt tick is usually four to eight. Do not pad because the slider goes there.

Native audio is on here too. Write “no speech, belt tone only,” or you will get a mouth you did not cast. 720p is the ceiling. Identity of this optical matters more than a 4K invention of a different hall. Chase resolution on the same locked still, not by switching to text-to-video.

Do not morph the stream into a different plant

The failure mode is a beautiful belt of the wrong material. PET becoming mixed paper mid-clip is a claim. An eddy that grows an optical the still never had is a claim. A buyer who walked your floor will notice before they notice your logo.

Do not:

• Text-to-video a “state-of-the-art MRF” with no upload.
• Generate pickers, a plant manager, or a talking host walking recovery rates.
• First-last two different bunkers and call it one line.
• Prompt the model to “improve” the hall, add a second optical, or restage the baler.

Hours, a material list, a “tours by appointment” line — type after picture lock. Add a text overlay burns one authored line, whole clip, top / center / bottom. It does not transcribe. Kling will not letter the bunker.

Pause on the optical. If it is the box on the floor, ship. If not, you used text-to-video or you first-last’d two plants. Go back to the still.

FAQ

Can I prompt text-to-video to “match our MRF” instead of uploading the line photo?

No. A sentence is not a sort line. The model will invent the hall and the stream. Upload the line photo to Kling V3 Standard image-to-video and describe only the belt.

Kling V3 Standard or Seedance 2.5?

Kling V3 Standard when the brief is a short belt move with camera control: 3s–15s, 63 credits for the first five seconds, native audio, image required. Seedance 2.5 when the same locked still needs a longer still-to-motion hold: 4–30s, 720p, 47 credits, native audio, opened from the image-to-video tool. Neither row is a talking plant manager.

Should we generate pickers so the line looks staffed?

No. Occupancy of a person is a likeness you do not need for a hardware plate. Photograph the empty belt. Animate that. If people appear, they should be real, released, and already in the photograph.

Do I need a last frame of a different stream?

Only when you photographed two approved states of the same line and the shot has to land. PET as first and mixed paper as last is a morph. A single hero still with a slow belt tick is the usual listing.