Upload a clip you already shot, describe the change in a sentence, and get back the same motion with a different sky, a new outfit or another hour of light. Six video-editing models run in the toolkit, so you can compare how each one reads the same instruction before you keep a version.
You filmed a shot you almost love, but the sky is flat, the jacket is the wrong color, or the light looks like noon when you wanted evening. A reshoot means another day, another location and another round of coordination. Text-based video editing offers a different route: upload the clip, describe the change in a sentence, and get back the same movement with a new sky, a new outfit or a new hour of light.
Traditional fixes for a clip that is almost right are expensive in effort. Changing the weather means compositing rain over every frame. Changing a jacket means masking it and tracking it through the whole shot. Changing the time of day means grading, relighting and hoping the shadows still make sense.
A video-editing model collapses them into one instruction. You supply the footage, which fixes the camera move, the timing and the performance, and the prompt says what should look different. The model rewrites the pixels and tries to leave the motion alone. That makes it a good fit when the footage is the valuable part and only one visual detail is wrong.
Picasso IA lists 28 models in its video-editing category, out of 502 in the catalog, and they all run from the same workspace, the toolkit. That matters here because the models disagree with each other more than you would expect. One handles clothing well, another is better at atmosphere, a third keeps faces steadier. Running the same clip through two or three of them costs some credits and some waiting, which is cheaper than committing to the wrong one.
Most edits fall into a handful of families, and each family leans on different models. Use this table to decide where to start, then adjust after you see the first result.
| What you want to change | Prompt wording that works | Models worth trying first |
|---|---|---|
| Weather | add steady rain, thick fog, or fresh snow on the ground | Wan 2.7 Videoedit, Gen 4 Aleph |
| Time of day | change the scene to golden hour with long warm shadows | Aleph 2, Wan 2.7 Videoedit |
| Outfit or clothing color | change the jacket to a red wool coat, keep everything else | P Video Edit, Lucy Edit 2 |
| Background or setting | replace the studio with a sunlit terrace, same camera move | Aleph 2, P Video Edit |
| Overall look or grade | cinematic teal and orange grade, soft film grain | Luma Modify Video, Wan 2.7 Videoedit |
Two habits help with any of these. First, name what must stay put as clearly as what should change: keep the camera movement, keep the lighting on the face, keep the background. Second, run the same prompt on a short test clip before you spend credits on the full one.
This is the loop that works whether the clip is a ten-second product shot or a travel scene. It assumes a clip in MP4 form and a rough idea of the change.
Generation costs credits, and the amount depends on the model and the settings you choose, so check the pricing page for what a plan includes. New accounts get free credits to try a first edit.
The most common reason an edit disappoints is a prompt that asks for too much. A sentence like make it a rainy evening and give her a black raincoat and a different street asks the model to change the light, the weather, the clothing and the location in one pass. Something will drift, and it is usually the thing you wanted to keep.
Ask for one change, name the constraint, and stack changes across separate runs. Rain first, then time of day, then the coat. Each step is easy to judge, and a failure repeats only that step.
Tip: end every prompt with a keep clause, such as keep the camera movement, the face and the background unchanged. Editing models treat everything you do not mention as negotiable, so the keep clause is the cheapest way to protect the parts of the shot you like.
Reference images help when words are not specific enough. Lucy Edit 2 and P Video Edit accept reference images, which is useful for showing an exact jacket or an exact pair of sunglasses instead of describing them. Luma Modify Video works differently: it has adherence levels, from staying very close to the source to reinterpreting it loosely, so a subtle grade and a heavy restyle are separate settings rather than separate prompts.
The six models overlap, but their limits differ, and the limits often decide which one you can use. Check each model page for current specs.
Aleph 2 is built for longer clips: its model page describes edits across multi-shot footage and accepts optional keyframe images pinned at specific moments, which is handy when you want the first and last frame to match a look. Gen 4 Aleph uses only the first five seconds of the input and lets you set the output aspect ratio, so it suits short social clips. Lucy Edit 2 takes both a prompt and a reference image and is aimed at swapping objects, clothing and styles.
Wan 2.7 Videoedit works on short clips and offers 720p or 1080p output, plus a choice between keeping the original audio and letting the model regenerate it. Luma Modify Video offers the adhere, flex and reimagine modes and an optional modified first frame. P Video Edit supports up to four reference images, a draft mode for a quick preview before the full render, and keeps the original audio by default.
None of them is the best at everything. For footage you care about, run your two best candidates and compare. You can browse the whole family in the collection or search it in the full catalog.
Be realistic about the edges before you plan a project around it. These are the places where edits break most often:
A full stylistic restyle, like turning a clip into a cartoon, is a different job again, covered in turning videos into anime.
Yes, within limits. Models such as Wan 2.7 Videoedit and Gen 4 Aleph list weather changes like rain, fog and snow among their uses, and you describe the result in plain words. The camera move and the subject usually carry over, while lighting and surfaces change to match. Expect to run it a couple of times, because wet ground, falling flakes and reflections are details the model has to invent for every frame.
It depends on the model, and each one has a cap. Some accept only a few seconds, some work up to roughly fifteen, and a few accept clips up to about thirty seconds, with file size limits as well. Because these specs get updated, read the model page in the toolkit before you upload. For longer footage, split it into single shots, edit each one with the same prompt and seed, and join them afterward.
Often close, but not guaranteed. Faces are the most common place for drift, especially when the head turns, the person moves quickly or the edit is large, like a new outfit plus a new location. Smaller, single-change prompts keep identity steadier. Only edit footage of yourself, people who have agreed to it, or fictional characters, and never use it to put words or looks on a real person without their consent.
You can try, and it is one of the better-supported uses. Say exactly which garment changes and what stays, for example change only the jacket to a navy blazer, keep the face, hair and background unchanged. P Video Edit and Lucy Edit 2 also accept reference images, so you can show the garment instead of describing it. Check sleeves, hands and neckline closely, since those edges are where clothing edits tend to wobble.
It depends on the model and its settings. P Video Edit keeps the original audio by default, and Wan 2.7 Videoedit lets you choose between keeping the original sound or letting the model decide whether to regenerate it. Other models may return video only or alter the sound, so listen to the result rather than assuming. If the audio matters, you can always re-attach the original track in your own editor afterward.
The goal is different. Here, you keep the footage looking like real footage and change one visible condition: the weather, the outfit, the hour of light. A stylistic restyle replaces the whole look of the clip with an illustrated one. Several of the same models can do both, but the prompts, the settings and the expectations are not the same, and the anime route has its own page on turning videos into anime.
Describing an edit works well for adding or replacing things and poorly for erasing. Removal needs the model to invent what was behind the object, frame after frame, while the camera moves. For that job use a workflow built for it, which is covered on the page about removing objects from videos. For a text prompt edit, a safer pattern is to replace the object with something harmless rather than ask for it to vanish.
That depends on the footage you started with, the model's terms and the rules of wherever you publish. If you shot the footage and the people in it agreed, you are in a much better spot than if the clip belongs to someone else. Check the platform's current requirements for AI-edited content, since some ask for a disclosure label, and read the model's terms on its page in the catalog before you rely on it for client work.
No. If you can write a sentence and upload a file you can run an edit, because there is no timeline, mask or keyframe to manage. Experience helps with judgment, like spotting a face that drifted. Start with a short clip, make one change, and compare it with the original before you try anything more ambitious.
Each run uses credits, and the cost depends on the model, the clip length and the settings you pick, so a number written here would be wrong soon. Video work sits higher in the catalog than images. New accounts get free credits to start, and the pricing page lists the current plans. A draft or lower-resolution pass, where a model offers one, is a sensible way to test a prompt cheaply.
Your clip is already shot, so the hard part is done. Open the Picasso IA toolkit, upload it, and describe the one change you wish it had.
Choose your language
