An AI short film on 8frame runs through five zones of one board: the script broken into shots, the reference sheet for characters and places, a still per shot, a clip per still, and the sound, before the cut happens in the Video Editor. The limit people do not expect is that nothing in this chain carries sound through to the end on its own: the Montage node drops audio, so dialogue, music and effects go back in on the Video Editor timeline.
The SHORT FILM template on /workflows is the starting point for the shot zone: it is built around one image generation that gives you many usable frames to pick from, instead of one prompt per frame. This page expands it into the whole film. For a trailer cut of the same film, see how to make a movie trailer with AI.
TL;DR
| Zone | Node or model | Price in credits | Needs |
|---|---|---|---|
| 1. Script | Sticky notes and frames | not a generation | Your script, split into shots |
| 2. References | Nano Banana Pro / Seedream 5 | 21 at 1K or 2K, 42 at 4K / 6 per image | Character and location descriptions |
| 3. Shot stills | ShotGrid / single image nodes | 42 per 3x3 sheet at 4K / per model | References plus a shot description |
| 4a. Clip from a still | Kling v3 Standard | 86 for 5 s with sound, 171 for 10 s | One start image |
| 4b. Long take | Seedance 2.5 / Wan 3.0 | 157 for 5 s at 720p, up to 30 s / 95 for 5 s at 1080p, 2 to 30 s | One start image |
| 4c. Several characters in one shot | Happy Horse, reference to video | 95 for 5 s at 720p, 189 at 1080p | 1 to 9 character references |
| 5a. Voice | ElevenLabs TTS | 13 per started 1,000 characters | The lines |
| 5b. Effects and score | ElevenLabs SFX / ElevenLabs Music | 1 to 6 by length / 81 per started minute | A text description |
| 6. Cut | Video Editor | not a generation | All clips and audio |
A credit is about $0.01 at pack rate. Failed generations are refunded automatically; a take you do not like is a new run at the same price.
The board, zone by zone
Zone 1: the script as shots
Paste the script into a frame and break it into numbered shots on sticky notes: shot number, what we see, who speaks, length in seconds. The length matters more than people think. It decides which model can make the shot in one take, and a shot written as 25 seconds rules out every model that stops at 15.
Mark each shot with one of three tags: "still and move" (one image, gentle motion), "long take" (over 15 seconds, one continuous move) and "dialogue" (a face speaking to camera). Each tag maps to a different node below.
Zone 2: one reference sheet before any shot
Generate the characters and locations first, in a frame of their own: each character front, three-quarter and full body in the film's costume, each location wide and close. Every later node points at these images. Nano Banana Pro is 21 credits at 1K or 2K, 42 at 4K. Seedream 5 is 6 per image and takes up to 14 reference images, the most on the roster; Nano Banana takes up to 10.
Use fictional characters, or real actors who have given written consent for AI images and video of them, covering the film and how it will be shown.
Zone 3: stills per shot
This is where the SHORT FILM template earns its place. ShotGrid makes a 3x3 contact sheet (nine frames in one image) for 42 credits at 4K, with a prompt of up to 5,000 characters and 1 to 10 reference images, in square (1:1) sheets; no other ratio on the canvas. One sheet per scene gives you nine angles of the same moment in the same light; pick the frames that work, then regenerate each chosen frame at full size and in the film's shape (16:9 or 9:16) on an image node, with the sheet as a reference; a square ninth of a sheet is a guide, not a final frame.
Lay the chosen stills in shot order in a row. That row is your animatic in still form, and you can show it to anyone before a single clip is made.
Zone 4: a clip per still
Connect each still to a video node by its tag:
- Still and move. Kling v3 Standard takes any whole second from 3 to 15, billed per second (17.01 credits per second with sound, 11.34 silent); it opens at 5 seconds with sound for 86 credits, and 10 seconds is 171. It is image-to-video only, which suits this zone because every shot already has its still.
- Long take. Seedance 2.5 runs 4 to 30 seconds and tops out at 720p (157 for 5 seconds, 937 for 30). Wan 3.0 runs 2 to 30 seconds and opens at 1080p (95 for 5 seconds; 9.45 per second at 720p).
- Two or three characters in frame. Happy Horse in reference-to-video mode takes 1 to 9 character references and costs 95 for 5 seconds at 720p, 189 at 1080p.
- A shot that must end on a set frame (a door closing on a specific composition): an end-frame model. On the canvas that is Kling 2.5 Turbo Pro (48 for 5 seconds), Kling 2.1 Pro (67), Kling O1 Image or Kling O3, or Seedance with a start image. Kling v3 does not take an end frame on the canvas.
Zone 5: dialogue and sound
A short film lives on sound, and this zone is all of it.
- Lines. ElevenLabs TTS bills 13 credits per started 1,000 characters, up to 5,000 characters per generation. Each generation starts a new block, so very short lines run one per generation cost 13 each; batching a character's lines into one script and cutting them on the timeline is cheaper.
- A face speaking to camera. The lip-sync nodes make a new clip from a still portrait plus audio: Pruna Avatar (3.375 per second at 720p, up to 35 seconds), OmniHuman v2 (71 per 5 seconds, billed per second of the audio, 4 to 35 seconds), Hailuo H3 Max Lip Sync (54 per 5 seconds at 768p, audio clipped at about 14.8 seconds). They do not re-sync the mouth in a clip you already made. We quote price and settings, not results.
- Effects. ElevenLabs SFX bills 1 to 6 credits by length (0.5 to 22 seconds).
- Score. ElevenLabs Music is 81 per started minute; MiniMax Music 3 is 0.27 per second.
Kling v3, Seedance and Hailuo H3 clips come with their own audio tracks. Listen before you keep or mute them: generated ambience can help a shot or fight your score.
Zone 6: the cut
Open the Video Editor mode. Lay the clips in shot order, then the dialogue, effects and score on sound layers, then titles and captions as text layers. The Filter Preset menu (Cinematic, Film Noir, Kodak, Super 8 and others) and a per-clip Brightness slider help clips from different models sit in one grade; there are no curves, colour wheels or LUTs.
You can preview a rough assembly with the Montage node (2 to 10 clips, at most 15 seconds of each, 10 credits per run), but treat it as picture only: the merged file has no sound.
Which model for which shot
| Shot | Model | Lengths on 8frame | Resolution | End frame on the canvas |
|---|---|---|---|---|
| Close-up, small move | Kling v3 Standard | 3 to 15 s | 720p | No |
| Wide establishing, slow drift | Wan 3.0 | 2 to 30 s | 480p, 720p, 1080p | No (canvas has one image slot) |
| Long continuous move | Seedance 2.5 | 4 to 30 s | 480p, 720p | Yes, with a start image |
| Two characters together | Happy Horse | 3 to 15 s | 720p, 1080p | No |
| Shot that lands on a set frame | Kling 2.5 Turbo Pro | 5 or 10 s | 1080p | Yes |
| 2K close-up | Hailuo H3 | 5, 10, 15 s | 2K | No |
No video model on 8frame outputs 4K, and there is no video upscaler. If the film is going to a big screen, pick the resolution per shot at generation time.
Festivals and awards: what the Academy now says
If the film is headed for festivals, read each festival's own rules before you start. As one reference point, the Academy's rules for the 99th Oscars, reported by NPR on 2026-05-03, say that only roles "demonstrably performed by humans with their consent" are eligible for acting awards, that screenplays must be "human-authored", and that the Academy "reserved the right to investigate the use of generative AI in any submission". The reported rules speak of performances and screenplays, not of images.
For a short film on 8frame that means: write the script yourself, and keep the board, because it shows which shots were generated and from which inputs if anyone asks.
Where it breaks
- Montage has no sound. It joins picture only; dialogue and score go on the Video Editor timeline.
- No mouth re-sync. A generated clip cannot be made to say a new line; make a lip-sync clip from a still instead.
- 720p ceilings. Seedance and Kling v3 Standard return 720p; mix them with 1080p shots knowingly.
- No transcription. The Video Editor's captions come from pasted text, timed at an estimated 160 words per minute, not from the audio. There is no SRT export.
- Consistency drifts with length. Long takes give a character more time to change. Cut long dialogue scenes into shorter shots from the same references.
- One error message. Almost every failed generation shows "Sensitive content detected. Please try changing your prompt or image." The credits come back; change the input and retry.
FAQ
Can I make a whole short film on one 8frame board? Yes: script notes, references, stills, clips and sound all live on the board, and the cut happens in the Video Editor mode. The board itself has no timeline.
Which model is best for long shots? For one continuous take, Seedance 2.5 (4 to 30 seconds, up to 720p) or Wan 3.0 (2 to 30 seconds, up to 1080p). Kling v3 Standard stops at 15 seconds.
How do I keep characters the same across shots? Make a reference sheet first and connect it to every image and video node. Happy Horse takes 1 to 9 character references in one shot.
Can an AI short film enter the Oscars? The Academy's rules for the 99th awards, as reported by NPR, make only human, consented performances eligible for acting awards and require human-authored screenplays. Check each festival's rules separately.
Why is there no sound after I join clips? The Montage node outputs picture only. Put your dialogue, effects and music on the Video Editor's sound layers.
Sources
- NPR via OPB, "Academy announces major overhaul to rules" (99th Oscars, AI eligibility), 2026-05-03: opb.org/article/2026/05/03/new-academy-rules-for-99th-oscars-in-2027, checked 2026-11-09
- 8frame SHORT FILM template description: /workflows/short-film, checked 2026-11-09
- 8frame prices, defaults and limits: the 8frame app, checked 2026-11-09
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.