How to Use AI Video Remake: From Reference Clip to New Video

Turning an already-edited reference clip into a new video of your own usually takes the most time not in generating the footage, but in understanding its shot structure first and then rebuilding it shot by shot. PixMind's Video Remake chains those two tasks together: it first breaks the reference clip into a shot plan with reference frames, then generates a brand-new video from a single editable merged script. To start, just open Video Remake and upload a clip.

Here is how it works: upload an MP4, MOV, or WEBM up to 100MB and 60 seconds, and the workspace lists each shot's number, reference frame, duration, and camera movement, then assembles everything into an editable remake script; swap the subjects, scenes, and text requirements for your own, choose duration, resolution, and aspect ratio, and you can generate a new video. A remake inherits the original's structure and rhythm — it is not a frame-by-frame copy.

Key takeaways

• Shot breakdown and video generation are two separate stages: first get a reviewable shot plan and merged script, and only submit for generation once you are happy with the edits.
• Input is MP4, MOV, or WEBM, up to 100MB and 60 seconds; output duration is selectable from 5 to 60 seconds.
• Generation runs on one merged script: when the duration stays within the model's single-segment limit, one run produces a complete video; beyond the limit, it is automatically split at shot boundaries and chained, not generated per shot and stitched together afterward.
• If all you want is a text prompt or standalone shot files, use Video to Prompt or the online video splitter instead of running a full remake.

Start with a reference clip that has clear structure

The ceiling of a remake is set when you pick the clip. A video with clear changes between shots and a rhythm that rises and falls yields a plan with genuinely reusable structure: what the opening uses to grab attention, which second the subject switches, and what action the ending lands on. Product ads, product showcases, and social-platform hook videos all meet this bar; single-take footage with nothing but slow camera movement has no meaningful shot boundaries to extract and doesn't work as a remake reference.

The current workspace accepts MP4, MOV, and WEBM files, up to 100MB and 60 seconds, and both limits must be met at once. Platform videos from TikTok, Reels, and the like must be downloaded as source files before uploading — the workspace can't import a platform page link directly. Both breakdown and generation require signing in to a PixMind account.

Sample reference clip (AI-generated video):

Upload and generate the shot breakdown plan

This step's deliverable is the shot breakdown plan — preparation material before generation — and it doesn't consume generation credits yet.

1. Drag the reference clip into the upload area, or click to select a file.
2. Choose the prompt language. Simplified Chinese, Traditional Chinese, English, Japanese, Korean, and more — 15 languages in total — are currently supported, and the shot descriptions from the breakdown will use the selected language.
3. Click "Generate remake storyboard." The file uploads first, then the backend analyzes it while the page shows progress.
4. When it finishes, review the shot cards one by one: each card carries a number, a reference frame, and a duration, plus scene, shot size, camera movement, and color tone, and shots with dialogue or sound effects also list the corresponding text. The number of shots kept in the breakdown and the number of reference images submitted with the script follow the number of reference images the selected model supports — the generation page's current guidance is authoritative.

The shot cards are read-only. What you actually edit sits lower on the page: the "editable video remake script." The system has already merged all shots into this combined script, and it is exactly what the new video will be generated from. Analysis is billed by reference clip duration, the same as Video to Prompt; generation credits get a separate estimate before you submit.

Rewrite the breakdown into your own production brief

There is one rule for rewriting: keep the structure for yourself, replace the content with yours. Shot order, each shot's approximate duration, and camera-movement logic are the most borrow-worthy parts of the reference; subjects, scenes, text, and style can all be swapped.

1. Read through the merged script and confirm each shot paragraph maps one-to-one to the cards above, keeping the original order and rhythm descriptions for now.
2. Replace subjects and scenes: change the original's protagonists, products, and locations into your own, keeping descriptions specific — spell out shape, color, and material.
3. Rewrite text and audio requirements: dialogue, text that must appear on screen, and sound-effect notes all change to match the new content.
4. Adjust style and trim: color tone, lighting, and camera movement can stay or be rewritten; delete shot paragraphs you don't need so the information load moves toward your target duration.

Watch cross-shot references while rewriting: consecutive shots must keep the same subject, and a single description shouldn't cram in multiple actions. You can rewrite paragraphs wholesale, but keeping clear shot numbers makes it easier for the result to hold the original's rhythm.

Choose output settings and generate a new video

Generation runs on this merged script as a unit, and the reference frames are submitted along with it to guide composition for the whole video.

1. Choose the generation model — Seedance 2.5 is the default, and available options follow the workspace's current list.
2. Choose the final duration (5 to 60 seconds). The default follows the reference clip's total length, which you can keep as is or shorten to keep only the core shots.
3. Choose resolution and aspect ratio. The aspect ratio auto-matches the reference clip's proportions — a vertical clip automatically selects 9:16, for example — so there's nothing to align by hand.
4. Decide whether to keep the original audio. The toggle is on by default: after generation, the reference clip's audio track replaces the generated video's audio so the music and voice match the original; this option is unavailable in segmented mode.
5. Review the credit estimate, then click "Generate remake video."

How generation runs depends on how the target duration relates to the model's single-segment limit. Seedance 2.5 and Wan 3.0 both cap a single generation at 30 seconds: within the cap, one merged script generates one complete video in a single run; beyond 30 seconds, the workspace groups adjacent shots into segments at shot boundaries, each within the cap, then generates them in sequence, chained. The last frame of each segment becomes the first frame of the next, with added consistency descriptions of the protagonist's appearance and the scene so characters, clothing, and lighting stay as consistent as possible across segments. Either way, the deliverable is a finished video — there is no step where each shot is generated as its own clip and then auto-edited together.

If generation fails, the error message appears on the result card and you can retry directly without redoing the breakdown; clicking "Regenerate" returns to the script page with the shot plan and output settings preserved, ready for small edits and resubmission.

Compare against the original to verify the structure survived

Once you have the new video, a structural comparison is more useful than replaying overall impressions.

1. Place the original and the result side by side, and mark three key points on the original's timeline: the opening hook, the first shot transition, and the closing action.
2. Compare one by one: does the result's opening establish attention just as early, does the subject switch happen at a similar timestamp, and does the ending land on the same kind of action?
3. Find the biggest deviation, go back to the merged script and change only that, then generate again. Shot-order problems outrank style problems: once the order drifts, no amount of stylistic similarity reads as structural similarity.

If the rhythm still doesn't match after a few rounds of edits, the reference clip itself is probably too long per shot or too information-dense. Switch to a shorter reference clip, or set the target duration closer to the breakdown's total length.

How to choose the right entry point when you need prompts or clips

Video Remake is the heaviest of the reference-based video tools, and picking the wrong entry point wastes credits. Judge by the deliverable you want:

For text, use Video to Prompt; for clips, use the splitter; step into Remake only when you want a finished video. To learn prompt extraction systematically, read the complete guide to Video to Prompt. Two advanced posts cover the same theme: "How to Adapt a Reference Ad's Shot Structure into a Product Film of Your Own" covers mapping a reference structure onto your own product, and "What to Do When Your Video Remake Doesn't Feel Like the Original: Tuning Shot Rhythm and the Brief" covers troubleshooting result drift.

FAQ

Does a remake copy the original one-to-one?

No. The reference clip is converted into a storyboard plan and a set of reference frames that guide the generation of a new video. The result can inherit the original's shot order, rhythm, and visual direction, but it won't be copied frame by frame — and once you swap subjects and scenes for your own content, the imagery is supposed to differ from the original.

How is it different from Video to Prompt?

Video to Prompt's deliverable is text: it extracts reusable visual prompts and production notes from a reference video without generating one. Video Remake goes further after the breakdown: review the shot plan, rewrite the merged script, choose output settings, and finally generate a new video. If you want a prompt to feed other generation tools, pick the former; if you want a finished video right away, pick the latter.

Does it generate a separate video for each shot?

No. Generation runs on the integrated remake script as one unit: when the duration stays within the model's single-segment limit (30 seconds for Seedance 2.5, for example), one run produces a complete video; beyond the limit, it splits at shot boundaries and chains segments — still aiming at one coherent finished video, not one clip per shot stitched back together.

Upload a clip and start with the breakdown plan

Find a 20-to-60-second reference clip with clear shot changes and upload it, then do the breakdown alone first: check whether the shot plan correctly identifies the opening, the transitions, and the ending before touching the script. Analysis is billed by reference clip duration, and generation credits get a separate estimate before you submit.

Open Video Remake.