A cinematic short does not need a film set. It needs a clear look, a shot list, and footage that matches the words on screen. This workflow uses AgentReel tools you can already open: generate the shots you don't have, then cut everything to a script with captions, and export for TikTok, Reels, Shorts, or YouTube.
1. Start with the look, not the gear
Pick one mood and one place before you generate anything. "Warm, quiet morning in a small bakery" is a look. "Cinematic" on its own is not, because it gives every clip a different colour and pace.
Write the look down in three lines: the time of day, the light, and the feeling. For example: early morning, soft window light, calm and unhurried. Keep those three lines in front of you for every shot. If a clip breaks the look, cut it, even if it is pretty.
Keep the story to one idea. A 45-second short can hold one morning, one job, or one small change. It cannot hold a whole day.
2. Write a shot list your tools can actually shoot
Break the short into shots, not scenes. One shot, one action. "She opens the shop" is one shot. "She opens the shop, makes coffee, and serves a customer" is three shots, and it will fall apart if you try to generate it as one clip.
A simple 45-second structure works well:
| Shot | Framing | Action |
|---|---|---|
| 1 | Wide | The street at dawn, shutters still down |
| 2 | Medium | The baker unlocks the door and steps inside |
| 3 | Close-up | Hands dusting flour over a worktop |
| 4 | Close-up | Steam rising from a tray out of the oven |
| 5 | Medium | The first customer takes a loaf, and smiles |
Write each shot as subject, action, camera, light. For example: "A baker in an apron slides a tray into the oven, slow push-in, warm morning light through the window, shallow depth of field." Camera words do real work here: dolly in, pan left, tracking shot, handheld, static tripod. Style words set the look: warm, muted, high contrast, soft light. Keep one clear action per clip, because three events packed into a 5-second clip will blur into none of them.
Note the limits before you plan: most clips in the AI Video Generator are 5, 10, or 15 seconds; Veo 3.1 always renders 8 seconds, and Vidu Q4 reaches 16 seconds from a start image. Plan in shots that fit those lengths. A cinematic short is a sequence of short shots, not one long take.
3. Generate only the shots you don't have
Open the AI Video Generator and work shot by shot. Start from what you have: with text alone, the model invents the whole scene; with a start image, it animates that picture, so a product shot or a portrait keeps its exact look. Some models also take a second image as the last frame, which lets you plan a move between two fixed points. Gemini Omni Video accepts up to seven reference images, the most in the generator, which helps when a face or a product must stay consistent.
Draft cheap, then finish expensive. Test framing and motion at 480p or 720p with a short length and a low-cost model, for example a 5-second 720p Runway clip costs 3 credits, and render only the keeper at 1080p or 4K. Veo 3.1 and the Gemini Omni models render 4K; most other models top out at 1080p or lower, so check the model card before you promise a final resolution. The Seedance models, MiniMax H3, PixVerse V6, and HappyHorse 1.1 also offer 21:9, a cinematic letterbox shape; Veo 3.1 and Gemini keep to 16:9 and 9:16.
Sound is a choice, not an afterthought. Veo 3.1, the Seedance models, Wan 3.0, Wan 3.0 Prime, and Vidu Q4 generate audio with every clip. Kling 3.0, Kling 3.0 Omni, and PixVerse V6 let you turn sound on or off, and turning it off lowers the price. If you will add your own music or voiceover later, a silent clip is the cheaper choice. The exact price appears on the Generate button before you start, and a failed generation refunds its credits automatically.
If you have no start frame yet, make one in the AI Image Generator first. A still that already has the right framing, colour, and subject usually gives a steadier start than a long text prompt, because the model no longer has to guess what the first second should look like.
4. Cut it to a script with the B-Roll Editor
Generated inserts alone are a montage. A script turns them into a film. In the AI B-Roll Editor (B-Roll Editor), drop in your footage, up to 500 clips, and paste your voiceover script, in English or Chinese, up to about 8 minutes of speech. The editor splits the script into caption-sized phrases and matches a shot to every line, so the picture follows the words instead of drifting.
Choose the pace on purpose. Fast pace cuts about every 1.5 seconds; slow pace lets each shot breathe. For a calm, cinematic feel, choose slow. For a punchy vertical short, choose fast and 9:16. The editor's plan moves from a wide establishing shot to medium shots and close-ups of the same scene, and the finished MP4 comes out in 16:9 (1920 x 1080) or 9:16 (1080 x 1920) with captions burned in, so it reads with the sound off.
Check the timeline phrase by phrase before you render. Every row lists its shots with the clip name, source time, and length, and filler shots, phrases with no matching footage, are marked. A timeline full of fillers is a footage problem, not an editing problem: upload clips for the phrases that need them, or generate those inserts, and match again. Changing the script or the pace clears the match, so settle both before you match, or you will pay for the match twice.
5. Cut the Shorts version from the same material
A cinematic short and a vertical short are the same story, cut differently. Keep one idea per clip, one joke, one tip, one reaction, and start right on the moment, not on the build-up. Viewers decide in the first seconds whether to keep watching.
If you already have a long recording, the existing clip tools cut your own footage into short MP4 parts for TikTok, Reels, and Shorts, with subtitles drawn into the picture so the clip reads on mute. Use those cuts as the raw material, then assemble the best parts with voiceover and captions in the B-Roll Editor or the combine-clips tools. Do not re-shoot the short for every platform: cut once, reframe to 9:16, and re-check the captions.
6. Export checklist
- Aspect ratio: 9:16 for TikTok, Reels, and Shorts; 16:9 for YouTube and websites; 1:1 for feeds.
- Captions: burned in and readable on mute; check the first line, not just the last.
- Sound: one voice, one music bed, or silence, never all three fighting.
- Colour: the same warm or muted look in every shot; one off-colour clip breaks the film.
- Credits: check the price shown before each generation and each B-Roll step, and confirm it. A failed step refunds its credits, and work already done is kept.
- Rights: use only footage you shot, footage you licensed, or clips you generated. Do not use other people's videos as if they were yours.
FAQ
How long should each generated clip be?As long as one action needs, and no longer. Most models offer 5, 10, or 15 seconds; Veo 3.1 is always 8 seconds; Vidu Q4 reaches 16 seconds from an image. If your shot needs more time, split it into two shots with a cut between them.
How do I keep a character or product looking the same?Start from a reference image instead of text alone, and reuse the same start image for every shot of that subject. Gemini Omni Video takes up to seven reference images; most other models take up to two frames. Do not expect a text prompt alone to hold a face steady across ten clips.
What if I have no footage at all?Generate stills first, then animate them. Image-to-video models such as Hailuo 2.3 and Vidu Q4 work from a single image, and most other models accept a start image too. Generate the inserts, then cut them to a script as in step 4.
What does a cinematic workflow cost on AgentReel?It depends on the model, length, and resolution of every clip, plus the B-Roll steps, so there is no single total. Every price is shown before you start, on the Generate button for video clips, and in the editor before each B-Roll step. Credit packs start at $11.99 for 100 credits; see the pricing page for current plans and packs, and check the price at publish time.
Can I make a 21:9 letterbox short?Yes, for generated inserts, on the Seedance models, MiniMax H3, PixVerse V6, and HappyHorse 1.1. The finished B-Roll cut itself exports in 16:9 or 9:16, so treat 21:9 as a look for inserts, not as a B-Roll export shape, unless you add letterbox bars in your own editor.
How we compared
This guide was checked against AgentReel's own live tool pages and pricing page on 2026-10-28, and against the repository routes for the AI Video Generator, the AI Image Generator, and the B-Roll Editor. No competitor copy, prompt, example scene, or price was carried into this post. Model durations, resolutions, aspect ratios, and sound behaviour are the values stated in the approved draft, which was verified against those live pages. Prices are stated only where the live pricing page confirms them at check time, and the post tells readers to check the price shown before each job, because prices can change.