FLUX Video Edit [fast]: Edit a Video With a Text Prompt

FLUX Video Edit [fast] changes an existing video from one written instruction: swap or remove an object, change the setting, restyle the clip, rewrite a sign or change what someone says. The motion, camera, timing and original audio stay as they were. On VdoBloom it is the default model in the AI Edit tab, takes clips from 0.7 to 15 seconds and costs 2.4 credits per second of your clip, so a 5-second edit is 13 credits.

Disclosure: VdoBloom is our platform.

Most AI video tools make a new clip from scratch. FLUX Video Edit [fast] does the opposite: you give it footage you already have and describe one change, such as “replace the red car with a white horse”. Everything you don’t mention is meant to come back untouched. This guide covers what the model is, how to use it on VdoBloom, the exact price, how to phrase edits so they land, its real limits, and how it compares with the other editors in the same tab.

What FLUX Video Edit [fast] is

FLUX Video Edit [fast] is a video-to-video editing model from Black Forest Labs, the company behind the FLUX image models. It was released in September 2026, and VdoBloom added it on September 25, 2026. It is not a text-to-video generator. According to Black Forest Labs’ FLUX Video Edit page, “length, framing, camera movement, timing and audio all come from the source clip”. As of September 2026, the page lists these kinds of edit:

• Remove, add or replace objects and characters: take a bin out of a street shot, put a plant on a desk, swap one product for another.
• Rebuild the setting: move the same action to a snowy night, a beach or a different room.
• Rewrite on-screen text: change what a sign, label or screen says.
• Change colours, materials and effects: a blue jacket instead of a red one, marble instead of wood.
• Change or translate dialogue with lip sync: new words for an existing line, or the same line in another language.
• Restyle the whole clip: turn live action into animation or a film-stock look.
• Change what happens, or stack several edits in one prompt.

The output keeps your clip’s length and aspect ratio. Black Forest Labs states the output is up to 720p at 24 fps, and that source audio “carries through untouched unless the prompt asks for a dialogue or sound change”.

How to use it on VdoBloom

1. Open the AI Edit tab. FLUX Video Edit [fast] is the first model in the picker and is selected by default.
2. Upload your clip (MP4, MOV or WebM) or paste a public video link.
3. Write the change you want in the prompt box, up to 4,096 characters. One or two sentences is usually enough.
4. Check the credit cost shown before you generate. It is based on the length of your clip.
5. Press generate. The edited video appears in the same panel and in your library.

What VdoBloom accepts for this model:

• Length: 0.7 to 15 seconds. Longer clips are refused before any credits are taken. They are not trimmed for you.
• File size: up to 50 MB.
• Size: at least 160 pixels on each side. Anything above 720p comes back at 720p.
• No reference image and no seed. The edit is driven by your text alone, so the reference-image and seed fields that other editors in the tab show are hidden for this model.

What comes back:

• Same length and shape. A 9:16 phone clip stays 9:16. The frame can come back a few pixels smaller: in VdoBloom’s integration runs on September 25, 2026, a 640 x 360 clip returned at 640 x 352.
• 24 fps. The model outputs 24 frames per second whatever the source rate, so a 60 fps clip loses its extra frames.
• Your original audio. Music, ambience and speech carry over unless you ask for a dialogue change.
• Speed. A 3-second edit took about 26 seconds in the same integration runs. Black Forest Labs quotes about 50 seconds for a 10-second clip.

How much FLUX Video Edit [fast] costs

You pay 2.4 credits per second of your source clip, rounded up to a whole credit, with a minimum of 3 credits. The edited video runs one frame (1/24 of a second) longer than the source, and that frame is billed too, which is why a 5-second clip is 13 credits rather than 12.

Three worked examples:

• A 3-second clip with one object swapped: 8 credits. New accounts get 10 free credits, so the first short edit costs nothing.
• A 10-second product clip re-set in three seasons (three separate runs): 3 × 25 = 75 credits.
• Testing a prompt on a 4-second cut first (10 credits), then running the full 15-second clip once (37 credits): 47 credits, instead of paying 37 for every attempt on the full clip.

The price doesn’t depend on what you ask for. A colour change and a full restyle cost the same. Prompts go through VdoBloom’s content check before any credits are charged, and if an edit fails, the credits go back to your balance automatically.

How to write edit prompts that work

The provider’s prompting guides make one point again and again: describe the change, not the whole shot. The model already sees the video. Your words should say what is different.

Habits that help

• Point by appearance and position. “The man in the blue shirt on the left” beats “the guy who walked in earlier”. The model sees positions in the frame, not the story.
• Describe the finished state. “A matte black mug” works better than “darken the mug”.
• For removals, name what is behind the object. “The wall continues behind it” or “the pavement continues” gives the model something to fill with. Add “for the whole clip” for things that move, so they don’t fade back in halfway.
• When removing a person, mention their shadow and reflection, or those can stay behind.
• Write separate edits as separate sentences. “Add a fig tree by the window. Paint the wall sage green. Make it evening.” Avoid two instructions about the same area, such as adding a lamp and then describing its light in a separate sentence.
• Don’t pad with “keep everything else the same”. Camera, timing and lighting are kept anyway. Only name nearby objects that you have seen drift.
• Start short. The shortest prompt that names the change is usually right. Add detail only if the first try misses.

Changing dialogue and on-screen text

• Put new lines in quotation marks and name the speaker: Make the woman in the yellow coat say “We open at nine tomorrow.”
• Fit the new line to the old one. The new speech has to fit the time the original speech took. The provider’s guide suggests about two to three words per second of the original talking, not of the whole clip. A line that is too long gets cut off mid-sentence.
• For translation, ask for the same sentence entirely in the new language, with no English words. Some languages run longer than English, so check the longest one first.
• For signs and labels, name the surface and the exact words. Flat, well-lit, front-facing text works best; curved, tiny or partly hidden text is less reliable.

Only the spoken words are rewritten. Background sound such as traffic or crowd noise passes through and can’t be removed with this model. If you want a whole video dubbed into another language, the Translate & Dub tool is built for that.

Honest limits

• 720p ceiling. A 1080p or 4K source comes back at 720p. If you need a sharper final, edit first and then run the result through the Upscale tab. The provider’s own advice is to edit before upscaling, not after.
• 24 fps output. Slow-motion or 60 fps footage loses frames.
• 15 seconds per run. For a longer video, cut it into pieces under 15 seconds, edit each, and join them. Avoid cuts in the middle of a spoken line.
• No seed. The same prompt on the same clip gives a slightly different result each time. Keep the version you like instead of expecting to reproduce it.
• Large, still objects are hard to remove. If something big never moves off what it covers, the model has no view of the background. Replacing it with something plausible usually works better than deleting it.
• Every pass re-encodes the whole clip. Chaining many edits one after another can make untouched areas drift. Stack related changes into one prompt where you can.
• It can’t change length, framing or aspect ratio. It edits the shot you have. It won’t extend it, reframe it or add a new camera move.

Editing footage of real people

Only edit footage you have the right to use, and get consent before changing what a real person says or does. Never use footage of minors. Don’t make intimate or sexual content of real people, or edits of public figures meant to deceive. VdoBloom blocks content involving minors and non-consensual content, and prompts are screened before a run starts.

FLUX Video Edit vs the other video editors on VdoBloom

The AI Edit tab has several models. Prices below are for a 5-second clip at the lowest resolution each one offers, at the time of writing.

How to choose:

• You can describe the change in words and 720p is fine: FLUX Video Edit [fast] is the cheapest option in the tab and keeps your audio.
• You need the new item to match a specific photo (your product, a particular outfit): pick a model that takes reference images, or Genjutsu, which swaps labelled items from reference photos in clips up to 30 seconds. Our guide to changing an outfit in a video walks through it.
• You need 1080p or 4K straight out of the edit: HappyHorse or Gemini Omni, at a higher price, or FLUX followed by an upscale.

If you edit photos as well, FLUX Kontext Pro applies the same instruction-style editing to still images.

Frequently asked questions

Is FLUX Video Edit [fast] free?

Not unlimited, but you can try it without paying. New VdoBloom accounts get 10 free credits, enough for one edit of a clip up to 4 seconds (10 credits). After that it is 2.4 credits per second of clip.

How long does an edit take?

Usually under a minute for short clips. A 3-second edit took about 26 seconds in VdoBloom’s integration runs on September 25, 2026, and Black Forest Labs quotes about 50 seconds for a 10-second clip.

Does it keep the original audio?

Yes. Music, ambience and speech come back as they were, unless your prompt asks to change what someone says. Background sound can’t be removed with this model.

Can I edit a video longer than 15 seconds?

Not in one run. Split it into sections under 15 seconds, edit each one with the same prompt, and join them afterwards. Each section is billed for its own length.

Can FLUX Video Edit output 1080p or 4K?

No. Output is capped at 720p, and larger sources are scaled down. For a sharper final file, upscale the edited clip afterwards on the Upscale tab.

Can it change what someone says in a video?

Yes, with lip sync. Put the new line in quotes, name the speaker, and keep it about as long as the original line. Only use this on footage of people who have agreed to it.