Guides

    Awnings: first-last retracted to extended

    Two photos of the same house.

    Versely Team4 min read

    Two photos of the same house. One with the awning retracted. One with it extended. That is the whole brief. First-last-frame interpolates the motion between those two stills. It does not invent a second elevation, and it does not "open" a fabric that only exists in one picture.

    Create a first-and-last-frame transition video is the named job. Attach start and end. The agent passes first_frame_url and last_frame_url into generate_video_from_image and picks a first-last-frame-capable model. Flux 3 First Last Frame to Video is one of those rows: two stills, interpolation, native audio, 5–20 seconds, 720p or 1080p, 9 credits a second.

    Same house. Same camera height. Same time of day if you can. Retracted to extended.

    The stills are the contract

    The fabric, the arms, the fascia, the window line — those have to be in both photographs, in the two states you actually sell. Image-to-video from a single retracted shot will guess the extend. That guess is a different awning: wrong projection, wrong pattern, sometimes a different house. First-last is how you pin both ends.

    Shoot them like a pair:

    • Tripod, or the same mark on the driveway.
    • Square to the elevation.
    • Retract, shoot. Extend, shoot. Do not "fix it in the prompt."
    • Overcast beats hard noon if the fabric has a pattern the sun will blow out.

    If the two frames are compositionally unrelated — different zoom, different standing position — the model will still connect them. The connection is a warp. Keep location, lighting, and lens family aligned so the path is an extend, not a morph.

    The prompt describes the path, not the house. "Awning extends smoothly to full projection, fabric tensioning, house and sky unchanged, locked camera." Re-describing "a beige bungalow with a patio" wastes the only prose this family needs: the in-between.

    Flux.3 meters the middle

    Durations are 5s through 20s, every second. A 5-second extend and a 20-second extend are different bills at 9 credits per second. Size the interpolation to the change. A small window unit does not need 20 seconds of fabric travel. A long lateral arm might.

    Native audio is on. Plan for a soundtrack the model will generate with the transition. If you need a specific VO ("motorized, twelve-foot projection"), that is a later pass — native audio here is not lipsync against your track. Qualities go to 1080p. Do not brief 4K. Match the stills. Cost is shown before you confirm; it is for this interpolation, not a budget to invent the missing end still in the same call.

    If either end is the wrong identity — different house, different fabric — every second is a morph between two mistakes, with audio attached.

    One photo is a different job

    One hero of the extended awning plus "make it retract" is ordinary image-to-video, and it will invent the retracted state. Turn a photo into a video is that job. Useful for a breeze on a locked extend. Useless as a product demo of travel you never photographed.

    Do not text-to-video "awning opens on a suburban house." Language does not pin pixels. Two files pin pixels.

    If you only shot retracted, reshoot the extend. Generating the last still and then interpolating is a fallback, not the product method. The motion is the cheap middle. The stills are the contract.

    Label the clip if a prospect could mistake interpolation for a live install they have not bought. The pair is their house, or an elevation you own; say which. Portfolio of completed jobs and visualizer clips are different libraries.

    FAQ

    Can I run first-last with one photo and a description of the extend?

    No. This capability needs two reference photos. Describe-the-end is single-still image-to-video. It will not land on a plate you never attached.

    Do the two photos have to be the same house?

    Yes, if the point is this house's extend. Two different elevations is a morph between strangers. Same opening, two states.

    What durations does Flux 3 First Last Frame list?

    5s through 20s, inclusive, one-second steps. 9 credits meters per second. Native audio is on. 1080p is the quality ceiling on the row.

    Should I generate the extended still if I only shot retracted?

    Reshoot if you can. Generate or edit a still first, then interpolate, only if you cannot go back. A real extended photo is the better last frame.