ClipCanva

FLUX 3 Video Continuation Workflow: 20-Second Clips, Keyframes, and the 15-Second V2V Limit

FLUX 3 video continuation is capped at 15 seconds as of 17 Aug 2026. Use t2v, i2v keyframes, v2v, and draft-enhance without mixing modes.

August 20, 2026ClipCanva Editorial

FLUX 3 Video Continuation Workflow: 20-Second Clips, Keyframes, and the 15-Second V2V Limit

A FLUX 3 video workflow does not start with a 20-second hero prompt. It starts by picking one mode: text-to-video, image-to-video with pinned keyframes, or video continuation. Black Forest Labs’ FLUX 3 video docs put all four jobs on one endpoint: t2v, i2v, v2v, and draft_enhance. On 17 August 2026, BFL capped continuation (v2v) at 15 seconds. Text-to-video, image-to-video, and draft enhance stay at up to 20 seconds. That is an operator rule, not a marketing slogan.

ClipCanva is not affiliated with Black Forest Labs, Canva, Kling, OpenAI, or Google. FLUX 3 video is BFL’s preview / early-access model. ClipCanva’s live FLUX Video AI page still treats FLUX as a stills source, then animates that frame. The FLUX 3 model page is coming soon. Use this guide to write the packet you can run today, and to avoid burning credits on the wrong FLUX 3 mode.

Official facts for FLUX 3 video and nearby tools

Source Public claim Workflow implication
BFL FLUX 3 announcement (23 Jul 2026) One multimodal model for image, video, and audio. Video with native audio up to 20 seconds in a single generation. Write audio in the prompt if you want speech, effects, or ambience. Silence is a setting, not a default.
BFL release notes (4 Aug 2026) Preview video: up to 20 seconds at FHD (1920 × 1088 for 16:9), 24 fps. Four modes: t2v, i2v with 1–10 pinned keyframes, v2v, draft_enhance. Audio on by default. One request, one mode. Do not ask continuation to invent the first frame.
BFL release notes (17 Aug 2026) Maximum duration for video continuation (v2v) is now 15 seconds. t2v, i2v, and draft enhance remain up to 20 seconds. If you need a 20-second beat, generate it as t2v or i2v. Continue only the last usable 5–15 seconds.
FLUX 3 video docs i2v keyframes: one image is the opening frame; two pin start and end; timestamped pairs pin exact times. v2v needs a start_video. Duration is whole seconds, 5–20, or auto. Resolution is hd (default) or fhd. Draft at HD. Spend FHD on the keeper.
FLUX 3 overview Drafts generate faster and cost about a third of a full render. draft_enhance reproduces the picked draft at full quality. Never enhance a clip you have not accepted for identity, product, and text.
Canva AI Video Generator Create a Video Clip, powered by Google Veo-3: one 16:9 clip per prompt, up to eight seconds, with synchronized audio. An 8-second Canva/Veo clip and a 15-second FLUX 3 continuation are different jobs. Do not paste the same prompt into both.

These are vendor-documented capabilities, not a guarantee that your SKU, face, or logo will hold. Access, region, credits, and exact limits change. Confirm the live BFL panel or API before you promise a client 20 seconds, FHD, or continuation.

Pick the mode before you write the prompt

FLUX 3 fails when you treat every brief as text-to-video. The mode decides what is allowed to change.

Job Mode What you must supply What the model is allowed to invent Fail condition
New establishing shot t2v One scene card, one camera move, optional quoted line World, motion, and audio bed Extra products, new wardrobe, unreadable type
Animate an approved still i2v, one keyframe The keeper still plus one verb Motion only The still is redesigned
First frame to last frame i2v, two keyframes Start still and end still The interpolation A third location appears between them
Timed storyboard i2v, up to 10 timestamped frames Ordered stills with seconds In-between motion Keys that fight each other (new face at 0:04)
Extend a keeper clip v2v start_video plus the next action The next 5–15 seconds Asking for a 20-second continuation after 17 Aug 2026
Cheap motion test draft: true, then draft_enhance Same prompt and cache bundle Nothing new on enhance Enhancing a draft that already melted the logo

If the brief is “make it longer,” that is v2v. If the brief is “this bottle must not change,” that is i2v from a locked still. If you do not have a still or a clip yet, that is t2v, and you should expect more drift.

Continuation card before you spend FHD credits

Write the card before you set fhd or chain a second v2v. BFL’s own loop is draft, pick, enhance. Continuation is a second generation, not a free extra.

Job: 10-second 16:9 product clip, then one 8-second continuation.
Viewer takeaway: same bottle, one press, then a slow push-in.
Locked identity: approved pack shot. Label text is real, not generated.
Mode 1: i2v. Keyframe 1 = front three-quarter still. Verb = one pump press.
Mode 2: v2v, 8 seconds, not 20. Start_video = the accepted i2v clip.
Audio: pump click + quiet room tone. No invented slogan.
Camera: static for the press, then 8-second push-in. One move per request.
Avoid: extra SKUs, extra hands, price overlay, second language on glass.
Review: would a buyer recognize the SKU with the logo cropped out?
Enhance: only the draft that passes identity at 100%.

If the continuation needs a new camera height or a new room, stop. That is a new t2v or a new i2v, not v2v. Continuation is for momentum, framing, and scene logic from the last frames.

A prompt formula that travels across t2v, i2v, and v2v

Keep one skeleton. Change only the mode-specific block.

Mode: [t2v | i2v | v2v].
Duration: [5–20 for t2v/i2v; 5–15 for v2v]. Resolution: hd draft, fhd keeper.
Aspect: [16:9 | 9:16 | 1:1]. Audio: [on, quoted line / off].

Subject: [SKU or person, materials, colors, distinctive marks].
Locked identity: [which still or clip defines face / label / silhouette].
Action: [one verb]. Camera: [one move]. End state: [what is in frame at cut].

If i2v:
Keyframes: [count]. 0s = [still]. [optional seconds = still]. Last = [still].
If v2v:
Start from the last frames of [clip]. Continue only [next action]. Do not recast.

Lighting: [source, direction]. Hold it.
Text: [exact words that must stay, or “no generated type”].
Avoid: [extra products, extra limbs, fake UI, new logos].

Original starter for a serum stills-to-continuation packet:

Mode: i2v, then v2v.

Create a 9:16 product clip, 8 seconds, hd draft.
Subject: 30ml frosted glass serum, gold pump, navy label that reads “NIGHT REPAIR”.
Locked identity: approved pack shot. Do not change bottle shape or type.
Action: a hand enters from camera-right and presses the pump once. One bead forms.
Camera: locked three-quarter. No orbit.
Audio: on. Soft pump click, no music, no voiceover.
Avoid: extra bottles, bathroom clutter, price tags, second languages.

Then v2v, 8 seconds, not 15:
Continue from the last frames. Slow 10-percent push-in. Hand holds still.
Do not add a new location. Do not rewrite the label.

Save the identity line and avoid list in Prompt Ideas. If the clip is a product ad, start from the product commercial prompt pack.

Stills vs video vs continuation

Use the cheapest lock that answers the review question.

Question Do this first Then
Does the SKU look like the SKU? Still in Text to Image or Image to Image Image to Video
Does one motion beat work? Short i2v or ClipCanva FLUX Video AI draft draft_enhance or a higher-cost pass
Do we need the next beat without a cut? Accepted clip as start_video v2v at ≤15 seconds
Do we need a new shot size? New still or new t2v Edit, do not continue
Do we need a script before pixels? AI Script Generator Scene cards, then stills

Canva’s public Veo-3 clip is one 16:9 generation, up to eight seconds, one video per prompt. FLUX 3’s continuation is a different contract: you hand it a finished clip and ask for the next seconds. Mixing those mental models is how teams request “one 20-second continuation” and hit the 17 August cap.

Creator / operator checklist

  1. Name the mode in the brief. t2v, i2v, or v2v. If two people say “extend it,” write the mode anyway.
  2. Lock a still before motion. FLUX 3 can do text-to-video. Your brand still should not be invented twice.
  3. One verb per request. Press, turn, walk, look. Not “cinematic energy.”
  4. Draft at HD. BFL documents drafts as faster and about one-third the cost of a full render.
  5. Cap continuation at 15 seconds. That is the 17 August 2026 API limit, not a taste preference.
  6. Pin keyframes on purpose. One still is an opening frame. Two stills are a transition. Ten stills are a board, not a mood dump.
  7. Quote speech if you need speech. Audio is on by default. Invented slogans are a legal review item.
  8. Review faces, hands, labels, and type at 100%. Typography is a claimed strength. It is still a gate.
  9. Enhance only the accepted draft. draft_enhance is a commit, not a second idea.
  10. Keep the packet. Card, refs, start clip, avoid list. The model version can change.
  1. Write the beats. Turn the brief into a four-line shot list in the AI Script Generator. If the source is a long cut, pull the usable beats with the AI Video Summarizer first.
  2. Store the language. Save the identity line, lighting, and avoid list in Prompt Ideas.
  3. Make the still you can run today. Use Text to Image, Image to Image, or GPT Image 2.
  4. Animate the keeper on the live path. Send the accepted frame to FLUX Video AI, Image to Video, or the AI Video Generator.
  5. If you need a second beat without a cut, use Video to Video on ClipCanva, or FLUX 3 v2v on BFL if you actually have early access. Do not wait for the FLUX 3 coming-soon page to ship.
  6. Compare models you can generate. Route the same packet through Compare or a known video route such as Kling 3.0. FLUX 3 video on ClipCanva is not a live generation model as of this writing.

The output you keep is the packet: identity still, mode, duration cap, start clip, motion verb. The endpoint can change. The packet should not.

Limits to keep in the brief

Official pages describe capabilities. They do not certify your commercial, legal, or brand outcome.

  • Availability is separate from the blog post. BFL announced FLUX 3 on 23 July 2026 and opened video as a preview on 4 August 2026. ClipCanva does not currently generate FLUX 3 video. Confirm account, region, and model name before you promise native 20-second FLUX 3 clips.
  • Continuation is shorter than generation. After 17 August 2026, v2v maxes at 15 seconds. A 20-second story is two shots or a t2v/i2v request, not one over-long continuation.
  • FLUX 3 Image is not the same rollout. BFL said image early access would follow video. Do not assume the video endpoint edits stills the way FLUX.2 did.
  • Draft enhance is not a style transfer. It is supposed to reproduce the draft you picked. If you want a different camera, start a new draft.
  • Do not invent list prices. BFL documents a settled cost in credits and a draft that is about one-third of a full render. If a third-party post lists a per-second dollar figure the official page does not state, leave it out.
  • MCP and agent skills are optional. The 7 August and 11 August notes add Agent Skills and MCP generate_video. They do not change the duration caps.

FAQ

What is FLUX 3 video?

FLUX 3 is Black Forest Labs’ multimodal model trained across image, video, and audio. The video preview generates clips with synchronized sound from text, pinned images, or an existing clip. BFL documents up to 20 seconds at FHD for text-to-video and image-to-video.

What changed on 17 August 2026?

BFL’s release notes set the maximum duration for video continuation (v2v) to 15 seconds. Text-to-video, image-to-video, and draft enhance stayed at up to 20 seconds. If a tutorial still says “continue for 20 seconds,” it is stale.

How many keyframes can I pin?

The 4 August 2026 notes describe image-to-video with 1–10 pinned keyframes. The video docs say one image starts the clip, two pin start and end, and [seconds, image] pairs pin exact times. More keys only help if each still is already approved.

Should I turn audio off?

Only if the deliverable is silent B-roll. Audio is on by default. Quoted dialogue, pump clicks, and room tone belong in the prompt. Generated slogans, prices, and legal lines still get replaced in edit.

Can I run this workflow on ClipCanva today?

You can do the planning and the live-model handoff today: script, prompt library, stills, image-to-video, and video-to-video. FLUX 3 video itself is BFL early access. ClipCanva’s FLUX Video AI path animates a FLUX or uploaded still. The FLUX 3 model page is coming soon. Keep the same packet if that model opens later.

Sources